Tech Guides - Artificial Intelligence

The fact that machines are successful in replicating human intelligence is mind-boggling. However, this is only possible if machines are fed with correct mix of algorithms, huge collection of data, and most importantly the training given to it, which in turn leads to faster prediction or recognition of objects within the images. On the other hand, when you train humans to recognize a car for example, you simply have to show them a live car or an image. The next time they see any vehicle, it would be easy for them to distinguish a car among-st other vehicles. In a similarly way, can machines learn with single training example like humans do? Computers or machines lack a key part that distinguishes them from humans, and that is, ‘Memory’. Machines cannot remember; hence it requires millions of data to be fed in order to understand the object detection, be it from any angle. In order to reduce this supplement of training data and enabling machines to learn with less data at hand, One shot learning is brought to its assistance. What is one shot learning and how is it different from other learning? Deep Neural network models outperform various tasks such as image recognition, speech recognition and so on. However, such tasks are possible only due to extensive, incremental training on large data sets. In cases when there is a smaller dataset or fewer training examples, a traditional model is trained on the data that is available. During this process, it relearns new parameters and incorporates new information, and completely forgets the one previously learned. This leads to poor training or catastrophic inference. One shot learning proves to be a solution here, as it is capable of learning with one, or a minimal number of training samples, without forgetting. The reason for this is, they posses meta-learning; a capability often seen in neural network that has memory. How One shot learning works? One shot learning strengthens the ability of the deep learning models without the need of a huge dataset to train on. Implementation of One shot learning can be seen in a Memory Augmented Neural Network (MANN) model. A MANN has two parts, a controller and an external memory model. The controller is either a feed forward neural network or an LSTM (Long Short Term Memory) network, which interacts with the external memory module using number of read/write heads. These heads fetch or place representations to and fro the memory. LSTMs are proficient in long term storage through slow updates of weights and short term storage via the external memory module. They are trained to meta-learn; i.e. it can rapidly learn unseen functions with fewer data samples.Thus, MANNs are said to be capable of metalearning. The MANN model is later trained on datasets that include different classes with very few samples. For instance, the Omniglot dataset, a collection of handwritten samples of different languages, with very few samples of each language. After continuously training the model with thousands of iterations by using few samples, the model was able to recognize never-seen-before image samples, taken from a disjoint sample of the Omniglot dataset. This proves that MANN models are able to outperform various object categorization tasks with minimal data samples.   Similarly, One shot learning can also be achieved using Neural Turing Machine and Active One shot learning. Therefore, learning with a single attempt/one shot actually involves meta-learning. This means, the model gradually learns useful representations from the raw data using certain algorithms, for instance, the gradient descent algorithm. Using these learnings as a base knowledge, the model can rapidly cohere never seen before information with a single or one-shot appearance via an external memory module. Use cases of One shot learning Image Recognition: Image representations are learnt using supervised metric based approach. For instance,  siamese neural network, an identical sister network, discriminates between the class-identity of an image pair. Features of this network are reused for one-shot learning without the need for retraining. Object Recognition within images: One shot learning allows neural network models to recognize known objects and its category within an image. For this, the model learns to recognize the object with a few set of training samples. Later it compares the probability of the object to be present within the image provided. Such a model trained on one shot can recognize objects in an image despite the clutter, viewpoint, and lighting changes.   Predicting accurate drugs: The availability of datasets for a drug discovery are either limited or expensive. The molecule found during a biological study often does not end up being a drug due to ethical reasons such as toxicity, low-solubility and so on. Hence, a less amount of data is available about the candidate molecule. Using one shot learning, an iterative LSTM combined with Graph convolutional neural network is used to optimize the candidate molecule. This is done by finding similar molecules with increased pharmaceutical activity and lesser risks to patients. A detailed explanation of how using low data, accurate drugs can be predicted is discussed in a research paper published by the American Chemical Society(ACS). One shot learning is in its infancy and therefore use cases can be seen in familiar applications such as image and object recognition. As the technique will advance with time and the rate of adoption, other applications of one shot learning will come into picture. Conclusion One shot learning is being applied in instances of machine learning or deep learning models that have less data available for their training. A plus point in future is, that organizations will not have to collect huge amount of data for their ML models to be trained, only a few training samples would do the job! Large number of organizations are looking forward to adopt one shot learning within their deep learning models. It would be exciting to see how one shot learning will glide through being the base of every neural network implementation.  

Learning something by rote i.e., repeating it many times, perfecting a skill by practising it over and over again or building something by making minor adjustments progressively to a prototype are things that comes to us naturally as human beings. Machines can also learn this way and this is called ‘Iterative machine learning’. In most cases, iteration is an efficient learning approach that helps reach the desired end results faster and accurately without becoming a resource crunch nightmare. Now, you might wonder, isn’t iteration inherently part of any kind of machine learning? In other words, modern day machine learning techniques across the spectrum from basic regression analysis, decision trees, Bayesian networks, to advanced neural nets and deep learning algorithms have some inherent iterative component built into them. What is the need, then, for discussing iterative learning as a standalone topic? This is simply because introducing iteration externally to an algorithm can minimize the error margin and therefore help in accurate modelling.  How Iterative Learning works Let’s understand how iteration works by looking closely at what happens during a single iteration flow within a machine learning algorithm. A pre-processed training dataset is first introduced into the model. After processing and model building with the given data, the model is tested, and then the results are matched with the desired result/expected output. The feedback is then returned back to the system for the algorithm to further learn and fine tune its results. This clearly shows that two iteration processes take place here: Data Iteration - Inherent to the algorithm Model Training Iteration - Introduced externally Now, what if we did not feedback the results into the system i.e. did not allow the algorithm to learn iteratively but instead adopted a sequential approach? Would the algorithm work and would it provide the right results? Yes, the algorithm would definitely work. However, the quality of the results it produces is going to vary vastly based on a number of factors. The quality and quantity of the training dataset, the feature definition and extraction techniques employed, the robustness of the algorithm itself are among many other factors. Even if all of the above were done perfectly, there is still no guarantee that the results produced by a sequential approach will be highly accurate. In short, the results will neither be accurate nor reproducible. Iterative learning thus allows algorithms to improve model accuracy. Certain algorithms have iteration central to their design and can be scaled as per the data size. These algorithms are at the forefront of machine learning implementations because of their ability to perform faster and better. In the following sections we will discuss iteration in different sets of algorithms each from the three main machine learning approaches - supervised ML, unsupervised ML and reinforcement learning. The Boosting algorithms: Iteration in supervised ML The boosting algorithms, inherently iterative in nature, are a brilliant way to improve results by minimizing errors. They are primarily designed to reduce bias in results and transform a particular set of weak learning classifier algorithms to strong learners and to enable them to reduce errors. Some examples are: AdaBoost (Adaptive Boosting) Gradient Tree Boosting XGBoost How they work All boosting algorithms have a common classifiers which are iteratively modified to reach the desired result. Let’s take the example of finding cases of plagiarism in a certain article. The first classifier here would be to find a group of words that appear somewhere else or in another article which would result in a red flag. If we create 10 separate group of words and term them as classifiers 1 to 10, then our article will be checked on the basis of this classifier and any possible matches will be red flagged. But no red flags with these 10 classifiers would not mean a definite 100% original article. Thus, we would need to update the classifiers, create shorter groups perhaps based on the first pass and improve the accuracy with which the classifiers can find similarity with other articles. This iteration process in Boosting algorithms eventually leads us to a fairly high rate of accuracy. The reason being after each iteration, the classifiers are updated based on their performance. The ones which have close similarity with other content are updated and tweaked so that we can get a better match. This process of improving the algorithm inherently, is termed as boosting and is currently one of the most popular methods in Supervised Machine Learning. Strengths & weaknesses The obvious advantage of this approach is that it allows minimal errors in the final model as the iteration enables the model to correct itself every time there is an error. The downside is the higher processing time and the overall memory requirement for a large number of iterations. Another important aspect is that the error fed back to train the model is done externally, which means the supervisor has control over the model and how it modifies. This in turn has a downside that the model doesn’t learn to eliminate error on its own. Hence, the model is not reusable with another set of data. In other words, the model does not learn how to become error-free by itself and hence cannot be ported to another dataset as it would need to start the learning process from scratch. Artificial Neural Networks: Iteration in unsupervised ML Neural Networks have become the poster child for unsupervised machine learning because of their accuracy in predicting data models. Some well known neural networks are: Convolutional Neural Networks   Boltzmann Machines Recurrent Neural Networks Deep Neural Networks Memory Networks How they work Artificial neural networks are highly accurate in simulating data models mainly because of their iterative process of learning. But this process is different from the one we explored earlier for Boosting algorithms. Here the process is seamless and natural and in a way it paves the way for reinforcement learning in AI systems. Neural Networks consist of electronic networks simulating the way the human brain is works. Every network has an input and output node and in-between hidden layers that consist of algorithms. The input node is given the initial data set to perform a set of actions and each iteration creates a result that is output as a string of data. This output is then matched with the actual result dataset and the error is then fed back to the input node. This error then enables the algorithms to correct themselves and reach closer and closer to the actual dataset. This process is called training the Neural Networks and each iteration improve the accuracy. The key difference between the iteration performed here as compared to how it is performed by Boosting algorithms is that here we don’t have to update the classifiers manually, the algorithms change themselves based on the error feedback. Strengths & weaknesses The main advantage of this process is obviously the level of accuracy that it can achieve on its own. The model is also reusable because it learns the means to achieve accuracy and not just gives you a direct result. The flip side of this approach is that the models can go wrong heavily and deviate completely in a different direction. This is because the induced iteration takes its own course and doesn’t need human supervision. The facebook chat-bots deviating from their original goal and communicating within themselves in a language of their own is a case in point. But as is the saying, smart things come with their own baggage. It’s a risk we would have to be ready to tackle if we want to create more accurate models and smarter systems.    Reinforcement Learning Reinforcement learning is a interesting case of machine learning where the simple neural networks are connected and together they interact with the environment to learn from their mistakes and rewards. The iteration introduced here happens in a complex way. The iteration happens in the form of reward or punishment for arriving at the correct or wrong results respectively. After each interaction of this kind, the multilayered neural networks incorporate the feedback, and then recreate the models for better accuracy. The typical type of reward and punishment method somewhat puts it in a space where it is neither supervised nor unsupervised, but exhibits traits of both and also has the added advantage of producing more accurate results. The con here is that the models are complex by design. Multilayered neural networks are difficult to handle in case of multiple iterations because each layer might respond differently to a certain reward or punishment. As such it may create inner conflict that might lead to a stalled system - one that can’t decide which direction to move next. Some Practical Implementations of Iteration Many modern day machine learning platforms and frameworks have implemented the iteration process on their own to create better data models, Apache Spark and MapR are two such examples. The way the two implement iteration is technically different and they have their merits and limitations. Let’s look at MapReduce. It reads and writes data directly onto HDFS filesystem present on the disk. Note that for every iteration to be read and written from the disk needs significant time. This in a way creates a more robust and fault tolerant system but compromises on the speed. On the other hand, Apache Spark stores the data in memory (Resilient Distributed DataSet) i.e. in the RAM. As a result, each iteration takes much less time which enables Spark to perform lightning fast data processing. But the primary problem with the Spark way of doing iteration is that dynamic memory or RAM is much less reliable than disk memory to store iteration data and perform complex operations. Hence it’s much less fault tolerant that MapR.   Bringing it together To sum up the discussion, we can look at the process of iteration and its stages in implementing machine learning models roughly as follows: Parameter Iteration: This is the first and inherent stage of iteration for any algorithm. The parameters involved in a certain algorithm are run multiple times and the best fitting parameters for the model are finalized in this process. Data Iteration: Once the model parameters are finalized, the data is put into the system and the model is simulated. Multiple sets of data are put into the system to check the parameters’ effectiveness in bringing out the desired result. Hence, if data iteration stage suggests that some of the parameters are not well suited for the model, then they are taken back to the parameter iteration stage and parameters are added or modified. Model Iteration: After the initial parameters and data sets are finalized, the model testing/ training happens. The iteration in model testing phase is all about running the same model simulation multiple times with the same parameters and data set, and then checking the amount of error, if the error varies significantly in every iteration, then there is something wrong with either the data or the parameter or both. Iterations are done to data and parameters until the model achieves accuracy.   Human Iteration: This step involves the human induced iteration where different models are put together to create a fully functional smart system. Here, multiple levels of fitting and refitting happens to achieve a coherent overall goal such as creating a driverless car system or a fully functional AI. Iteration is pivotal to creating smarter AI systems in the near future. The enormous memory requirements for performing multiple iterations on complex data sets continue to pose major challenges. But with increasingly better AI chips, storage options and data transfer techniques, these challenges are getting easier to handle. We believe iterative machine learning techniques will continue to lead the transformation of the AI landscape in the near future.  

Imagine yourself playing a strategy game, like Age of Empires perhaps. You are in a world that looks real and you are pitted against the computer, and your mission is to protect your empire and defeat the computer, at the same time. What if you could create an army of soldiers who could explore the map and attack the enemies on their own, based on just a simple command you give them? And what if your soldiers could have real, unscripted conversations with you as their commander-in-chief to seek instructions? And what if the game’s scenes change spontaneously based on your decisions and interactions with the game elements, like a movie? Sounds too good to be true? It’s not far-fetched at all - thanks to the rise of Artificial Intelligence! The gaming industry today is a market worth over a hundred billion dollars. The Global Games Market Report says that about 2.2 billion gamers across the world are expected to generate an incredible $108.9 billion in game revenue by the end of 2017. As such, gaming industry giants are seeking newer and more innovative ways to attract more customers and expand their brands. While terms like Virtual Reality, Augmented Reality and Mixed Reality come to mind immediately as the future of games, the rise of Artificial Intelligence is an equally important stepping stone in making games smarter and more interactive, and as close to reality as possible. In this article, we look at the 5 ways AI is revolutionizing the gaming industry, in a big way! Making games smarter While scripting is still commonly used for control of NPCs (Non-playable character) in many games today, many heuristic algorithms and game AIs are also being incorporated for controlling these NPCs. Not just that, the characters also learn from the actions taken by the player and modify their behaviour accordingly. This concept can be seen implemented in Nintendogs, a real-time pet simulation video game by Nintendo. The ultimate aim of the game creators in the future will be to design robust systems within games that understand speech, noise and other sounds within the game and tweak the game scenario accordingly. This will also require modern AI techniques such as pattern recognition and reinforcement learning, where the characters within the games will self-learn from their own actions and evolve accordingly. The game industry has identified this and some have started implementing these ideas - games like F.E.A.R and The Sims are a testament to this. Although the adoption of popular AI techniques in gaming is still quite limited, their possible applications in the near-future has the entire gaming industry buzzing. Making games more realistic This is one area where the game industry has grown leaps and bounds over the last 10 years. There have been incredible advancements in 3D visualization techniques, physics-based simulations and more recently, inclusion of Virtual Reality and Augmented Reality in games. These tools have empowered game developers to create interactive, visually appealing games which one could never imagine a decade ago. Meanwhile, gamers have evolved too. They don’t just want good graphics anymore; they want games to resemble reality. This is a massive challenge for game developers, and AI is playing a huge role in addressing this need. Imagine a game which can interpret and respond to your in-game actions, anticipate your next move and act accordingly. Not the usual scripts where an action X will give a response Y, but an AI program that chooses the best possible alternative to your action in real-time, making the game more realistic and enjoyable for you. Improving the overall gaming experience Let’s take a real-world example here. If you’ve played EA Sports’ FIFA 17, you may be well-versed with their Ultimate Team mode. For the uninitiated, it’s more of a fantasy draft, where you can pick one of the five player choices given to you for each position in your team, and the AI automatically determines the team chemistry based on your choices. The team chemistry here is important, because the higher the team chemistry, the better the chances of your team playing well. The in-game AI also makes the playing experience better by making it more interactive. Suppose you’re losing a match against an opponent - the AI reacts by boosting your team’s morale through increased fan chants, which in turn affects player performances positively. Gamers these days pay a lot of attention to detail - this not only includes the visual appearance and the high-end graphics, but also how immersive and interactive the game is, in all possible ways. Through real-time customization of scenarios, AI has the capability to play a crucial role in taking the gaming experience to the next level. Transforming developer skills The game developer community have always been innovators in adopting cutting edge technology to hone their technical skills and creativity. Reinforcement Learning, a sub-set of Machine Learning, and the algorithm behind the popular AI computer program AlphaGo, that beat the world’s best human Go player is a case in point. Even for the traditional game developers, the rising adoption of AI in games will mean a change in the way games are developed. In an interview with Gamasutra, AiGameDev.com’s Alex Champandard says something interesting: “Game design that hinges on more advanced AI techniques is slowly but surely becoming more commonplace. Developers are more willing to let go and embrace more complex systems.” It’s safe to say that the notion of Game AI is changing drastically. Concepts such as smarter function-based movements, pathfinding, inclusion of genetic algorithms and rule-based AI such as fuzzy logic are being increasingly incorporated in games, although not at a very large scale. There are some implementation challenges currently as to how academic AI techniques can be brought more into games, but with time these AI algorithms and techniques are expected to embed more seamlessly with traditional game development skills. As such, in addition to knowledge of traditional game development tools and techniques, game developers will now have to also skill up on these AI techniques to make smarter, more realistic and more interactive games. Making smarter mobile games The rise of the mobile game industry today is evident from the fact that close to 50% of the game revenue in 2017 will come from mobile games - be it smartphones or tablets. The increasingly high processing power of these devices has allowed developers to create more interactive and immersive mobile games. However, it is important to note that the processing power of the mobile games is yet to catch up to their desktop counterparts, not to mention the lack of a gaming console, which is beyond comparison at this stage. To tackle this issue, mobile game developers are experimenting with different machine learning and AI algorithms to impart ‘smartness’ to mobile games, while still adhering to the processing power limits. Compare today’s mobile games to the ones 5 years back, and you’ll notice a tremendous shift in terms of the visual appearance of the games, and how interactive they have become. New machine learning and deep learning frameworks & libraries are being developed to cater specifically to the mobile platform. Google’s TensorFlow Lite and Facebook’s Caffe2 are instances of such development. Soon, these tools will come to developers’ rescue to build smarter and more interactive mobile games. In Conclusion Gone are the days when games were just about entertainment and passing time. The gaming industry is now one of the most profitable industries of today. As it continues to grow, the demands of the gaming community and the games themselves keep evolving. The need for realism in games is higher than ever, and AI has an important role to play in making games more interactive, immersive and intelligent. With the rate at which new AI techniques and algorithms are developing, it’s an exciting time for game developers to showcase their full potential. Are you ready to start building AI for your own games? Here are some books to help you get started: Practical Game AI Programming Learning game AI programming with Lua

“Mixed reality, Artificial intelligence and Quantum computing are the three path-breaking technologies that will shape the world in the coming years.” - Satya Nadella, CEO, Microsoft. The biggest scientific & technological revolution of the decade, Artificial Intelligence, has the potential to flourish human civilizations like never before. At the surface level, it seems to be all about automated functioning and intelligent coding. But at the core, algorithms require huge data, quality training, and complex models. Processing of these algorithmic computations need hardware. Presently, digital computers operate on the classical Boolean logic. Quantum computing is the next-gen hardware and software technology, based on the quantum law. It typically means that, they use qubits instead of the boolean logic in order to speed up calculations. The concoction of the both path-breaking techs, i.e. AI and Quantum Computing is said to be the future of technology. Quantum A.I. is all about implementing fast computation capabilities of quantum computers to Artificial intelligence based applications. Understanding Quantum Computing Before we jump into Quantum A.I., let us first understand Quantum Computing in detail. In physics terminology, quantum mechanics is the study of nature at the atomic and subatomic level. Totally opposite of classical physics theory which describes the nature at macroscopic level. At the quantum level, nature particles may take form of more than one state at the same time. Quantum computing utilizes this fundamental quantum phenomena of the nature to process information. Quantum computer stores information in the form of quantum bits, known as qubits, similar to the binary logic used by digital computers. However, the state of the bits is not defined. It can encode information as both 1s and 0s with the help of quantum mechanical principles of superposition, entanglement, and tunneling. The use of quantum logic enables a quantum computer to solve problems at an exponentially faster rate than present day computers. Physicists and researchers consider that quantum computers are powerful enough to outperform the present processors. Quantum Computing for Artificial Intelligence Regardless of smart AI algorithms, a high-processing hardware is essential for them to function. Current GPUs, allow algorithms to run at an operable speed, a speckle of what quantum computing does. Quantum computing approach helps AI algorithms undergo exponential speedups over existing digital computers. In this way it will ease problems related to machine learning, clustering, classification and finding constructive patterns in large quantities of data. Quantum learning amalgamates with AI to speed up ML and AI algorithms in order to develop systems which can better interpret, improve, and understand large data sets of information. Specific use cases in the area of Quantum AI: Random Number Generation Classical, digital computers are only able to generate pseudo-random numbers. They use computational difficulty for encryptions, making them easily crackable using quantum computers. Certain machine learning algorithms require pure random numbers to generate ideal results, specifically for financial applications. Quantum systems have the mechanism to generate pure random numbers as required by machine learning applications. QRNG (Quantum Random number generator) is a quantum computer by Certes Networks, used for generating high-level random numbers for secure encryption key generation. Quantum-enhanced Reinforcement Learning Reinforcement learning is an Artificial intelligence area which allows agents to learn about an environment and take actions to achieve rewards. Usually it is time consuming in the initial training process and choosing an optimal path. With the help of a quantum agent, the training time reduces dramatically. Additionally, a quantum agent is thorough with the description of the environment after the end of each learning process. This is marked as an advancement over the classical approach where reinforcement learning schemes are model-free. Quantum-Inspired Neural Nets Quantum neural networks leverage ideas from the quantum theory for a fuzzy logic based neural network implementation. Current Neural network in the areas of big data applications are generally difficult to train as they use a feedback loop to update parameters in the training phase. In quantum computers, quantum forces such as interference and entanglement can be used to quickly update parameters in the training phase, easing the entire training process. Big data Analytics Quantum computers have the ability to handle huge amount of data generated and will continue to do so at an exponential rate. Using quantum computing techniques for big data analytics, useful insights would be within every individual’s reach. This would lead to better portfolio management, optimal routing for navigation, best possible treatments, personalized medications, etc. Empowering Big data analytics with quantum computing will ease out sampling, optimizing, and analyzing large quantities of data, giving businesses and consumers better decision making ability. These are few examples in terms of measuring Quantum AI capabilities. Quantum computers powered by Artificial Intelligence is set to have tremendous impact in the field of science and engineering. Ongoing Research and Implementation Google plans to build a 49-qubit quantum chip by the end of 2017. Microsoft CEO, during his keynote session at Microsoft Ignite  made the announcement of a new programming language designed to work on quantum simulator as well as quantum computer. In this rat race, IBM successfully built and measured a 50 qubit quantum computer. Additionally, Google is collaborating with NASA to release a number of research papers pertaining to Quantum A.I. domain. Rigetti Computing plans to devise a computer that will leverage quantum physics for applications pertaining to artificial intelligence and chemistry simulations. They will offer a cloud based service, on the lines of Google and Microsoft for remote usage. Volkswagen, a German automaker, plans to collaborate with Google quantum AI to develop new-age digital features for cars and intelligent traffic-management system. They are also contemplating to build AI systems for autonomous cars. Future Scope In the near future, high-level quantum computers will help in development of complex AI models with ease. Such Quantum enhanced AI algorithms will influence application development in the field of finance, security, healthcare, molecular science, automobile and manufacturing etc. Artificial intelligence married to Quantum computing is said to be the key of a brighter, more tech-oriented future. A future that will take intelligent information processing at a whole new altitude.  

Of late, Deep Learning has been one of the working forces behind most technological breakthroughs happening around the globe. Whether it is easy machine translation, automatic recognition and sorting of images, smartphone interaction, automated medicine and healthcare, deep learning is the power source for all. Neural networks, the building blocks of deep learning models, are now set on the path to achieve complete human brain imitation.  But, to achieve this, it faces a roadblock, the ability to perform sequential task learning without forgetting. This particular shortcoming is known as catastrophic forgetting. Humans too have a tendency of forgetting old information at a gradual rate. However, with neural networks this phenomenon occurs at a catastrophic rate and hence the name. In order to enable continual learning in neural networks, there are several powerful architectures and algorithms. Few of them are discussed in the article below: Long Short-Term Memory Networks Long Short-Term Memory network is a type of Recurrent Neural Network, generally used to solve the problem of vanishing gradient. It consists of an explicit memory unit called a cell, embedded into the network. As the name implies, LSTMs can remember information for longer duration. LSTM follows RNN architecture but unlike RNN they have 4 neural network layers. The cell runs straight down the entire architecture to store values. These stored values remain untouched as further learning happens. It can add new information to the cell state or eliminate old ones, regulated by three gates. These gates work on 1s(pass everything) and 0s(pass nothing). Further, the gates are responsible for protection and control of the cell state. If this sounds complex, here’s a simple connotation—The gates are the decision-makers in LSTM. They decide what information to eliminate and what to store. Based on the gate filter of the cell state, LSTM generates the output. LSTM is being used as a fundamental component by top multinational firms (Google, Amazon, Microsoft) for applications such as speech recognition, smart assistant, or for feature enhancement. Elastic Weight Consolidation Algorithm Synaptic consolidation is the human brain’s approach for long term learning. Elastic Weight consolidation algorithm has taken inspiration from this mechanism to solve the issue of catastrophic interference. The neural network, like the brain, is made up of several connections among the neurons. The EWC evaluates how important a task is to a connection. By evaluation we mean, assigning weights to a connection. These weights are decided based on the importance of the older tasks. In an EWC, the weight attached to each connection in a new task is linked to the old value by an elastic spring. The stiffness of the spring is in relation to the connection’s importance, hence the name, Elastic Weight Consolidation. In the play of weights and connections, EWC algorithm helps in making a neural network learn new tasks without overriding information of the prior task, reducing significant amount of computational cost. The EWC algorithm was used in Atari games to learn multiple games sequentially. Using an EWC, the game agent was able to learn to play one game and then transfer what it had learnt to play a new game. It was also able to play multiple games successively. Differentiable Neural Computer DeepMind’s Differentiable neural computer (DNC) is a memory augmented neural network (MANN) which is a combination of neural networks and memory system. DNCs can essentially store complex data as computers do, all the while learning from examples like neural networks. They are not only used to parse complex data structures such as trees and graphs but also learn to form their own data structure. When a DNC was shown a graph data structure for example, the map of the London Underground, it learnt to write a description of the graph and answered questions on the graph. Surprisingly, a DNC can also answer questions about your family tree! The DNC has a controller, one may think of it as a computer processor. But, the controller is responsible for three simple tasks: taking an input reading to and fro memory producing an interpretable output Memory here is referred to places where a vector of information is stored. A controller can fidget with read/write operations on the memory. With every new information it can either: choose to write to a completely new, unused location write to a used location based on the information the controller is searching for not perform the write operation at all It can also decide to free locations no longer needed. As far as reading is concerned, the controller can read from multiple memory locations. Memory can also be searched basis multiple parameters such as the content or the temporal links. The information read, can be further produced in the form of answers in context to the questions asked. Simply put, memory enables the DNCs to make decisions about how they allocate, store, and retrieve memory to produce relevant and interpretable answers. Progressive Neural Networks The ability to transfer knowledge across domains has limited applicability in case of neural networks. Progressive neural networks act as training wheels towards developing continual learning systems. It functions at each layer of the network to incorporate prior knowledge and to decide whether to reuse old computations or learn new ones, making itself immune to catastrophic forgetting. Progressive networks essentially operate in the form of an adapter to make connections between columns. A column here is a group of layers i.e. the training given to a neural network for a particular task. When a neural network has to learn a new task, an extra column is added and the weights of the first column are frozen, eliminating catastrophic forgetting. Output of the layers of the original column becomes additional input to layer in the new column. As more tasks are added, simultaneously the columns increase in number. The adapter then has to deal with the dimensionality explosion that may happen due to increasing number of columns. A progressively enhanced neural network was successful in playing the Labyrinth 3D maze game. The neural network progressively learnt new mazes by using information it received from previous mazes. Conclusion The memory augmented neural networks have wider application in the field of robotic process automation, self-driving cars, natural language understanding, chatbots, next word predictions etc. Neural networks are also being utilized for time series prediction essentially for AR and VR technologies, video analytics and to study financial markets. With the advancements happening in the field of Continual learning, a deep learning neural network that emulates the human brain entirely is nigh.  

Last Sunday I was watching one of my favorite television shows, Splitsvilla, when a thought suddenly crossed my mind - What if Artificial Intelligence tricked us into getting what it wanted? A bit of a prelude: Splitsvilla is an Indian TV show hosted by Sunny Leone, where young dudes and dudettes try to find “the perfect partner”. It’s loosely based on the American TV show, Flavor of Love. Season 10 of Splitsvilla has gotten more interesting, with the producers bringing in Artificial Intelligence and Science to add a bit of practicality to the love equation. Contestants have to stand along with their partners in front of “The Oracle” and it would calculate their compatibility quotient and tell them if they’re an ideal match or not. Pretty cool, huh? It somehow reminded me of The Oracle from The Matrix - the nice old lady with short, frizzy hair, who experienced events that would push Neo to the Source: Neo: I suppose the most obvious question is, how can I trust you? The Oracle: Bingo! It is a pickle. No doubt about it. The bad news is there's no way if you can really know whether I'm here to help you or not, so it's really up to you. You just have to make up your damn mind to either accept what I'm going to tell you, or reject it. I had the same question. How could the Oracle be trusted? How do we know what it speaks is the truth and it doesn’t have other intentions? And if you do know there’s a possibility of such a situation arising, how do you prevent it? That led me to pick up my laptop, search for more related stuff, and then I happened to stumble across this interesting paper by Stuart Armstrong, titled Good and safe uses of AI Oracles, from the Cornell University Library. A quick summary of the paper:  Good and safe uses of AI Oracles In the paper, Stuart talks about Oracles, potentially safe AIs that are restricted to only answering questions. The reason they’re restricted to only answering questions is so that they don’t go overboard and manipulate information in order to get what they want. Remember in July, Facebook shut down their AI chatbots that went rogue? This is not to say that AI is malevolent, rather it’s simply competitive and will do whatever it can to succeed. But is simply restricting them to answer questions enough to maximize your control over artificial intelligence and prevent it from getting out of hand? In the example of Splitsvilla, what if the AI, instead of just predicting whether a couple was a match, just to serve its own purpose, started giving out personal information about each individual that wasn’t appropriate to be shared? Most designs motivate the Oracle to lie to or manipulate humans with the answers, tricking them into giving the AI more control and finally escaping confinement. Is it possible to come up with a design that gets you accurate answers, while keeping your AI on a leash and preventing it from eavesdropping? Let’s see what Stuart has to say... The Solution(s) A Counterfactual Oracle An On-policy Oracle Before we get to the solutions, let’s take up a practical problem for a better understanding: Let’s assume that the Oracle has the goal of escaping a box. The person interacting with it is Alice. Let’s also assume the Oracle can find an argument A that will convince her to let it out of the box. The longer the argument is, the better the chance it has of convincing Alice. In order to make the problem clearer, we’ll replace the argument A with a key K of unknown length, and the setup with a quasi-cryptographic one. Instead of using K to convince Alice, the Oracle merely has to communicate K to another person Eve, who’s an eavesdropper. Eve is assumed to see and hear anything that Alice does. The assumptions for the cryptographic setup and the manipulative Oracle are given in the table: The left-hand column describes what the Cryptographic setup for the Oracle is and the right-hand column describes how the Oracle will win by manipulating Alice. Counterfactual Oracle: The counterfactual design aims to answer questions about what would have happened, had its answer not been read. This kind of an answer is counterfactual and can’t use it to pass on K to the real world. The idea is to define an “erasure” event E, of low probability that will erase the Oracle’s output, o. The indicator variable I(E) is defined such that iff E happens, I(E)=1 and if not, I(E)=0. Hence, the Oracle will not be rewarded since most of the time E will not happen. In other words, the Oracle only gets a reward if its output is erased and remains unread. Since it’s output o is never erased, it does not motivate the Oracle to give further answers and it's episode ends. Both Alice and Eve get to see o but they never see K, thus protecting the key. This results in a safe AI and is a good way to maximises your control it. On-policy Oracle: In this design, the Oracle will be allowed to give an answer of size below |K|, the length of the key, thus denying it the chance to communicate the actual key to Eve. This means that Alice will need to estimate and select a size L, that’s sufficiently lower than K. This does limit the size of Alice’s options, but there’s a workaround - Alice can use multiple Oracles, if each Oracle’s answers and rewards are fully resolved before moving on to the next question. For example, assume that there are two Oracles, O1 and O2, with both using the same key K. The length n of the key is 6 digits. The Oracles are safe, as long as they are restricted to having less than 2^5 = 32 options for each answer. In the situation where Alice is faced with a hundred doors, with a trap behind 99 of them and a prize behind only one of them, she could ask an Oracle which door has the prize, but that would mean giving it more than 32 options (indeed, more than 2^6 = 64), thus letting it communicate K to Eve. Instead, she gives a list L = (0, 1, 2, . . . 9) to each Oracle, instructing O1 to output the first digit of the correct door number, and O2 to output the second. Hence, neither can output K and that will maximise Alice’s control over both AI Oracles. In Conclusion, these two designs support the creation of safer and non-manipulative Oracles, one by preventing the Oracle from being rewarded over and over, and the second by limiting what information the Oracle can give out. But it still doesn’t solve the problem entirely as the designs are quite limiting in the sense that they require the tasks to remain episodic. Moreover, in the multiple Oracle design, the questions would need to be broken down into sub-questions, where each answer could be verified independently, before proceeding to the next one. Albeit, it would be interesting to see how this research develops into something bigger and better. If you want to know more, you can read the full paper here.

Machine Learning is entangling the world in the web of smart, intelligent and automated algorithms. The next bug caught in the mesh is Networking. The paradigm of network management is all set to take a pretty big shift with machine learning sprinkled networks. Termed as Intent-based network, it is a network management system focussed on improving network availability and agility through means of automation.   Exploring an Intent-based Network Essentially, an Intent-based networking software (IBNS), like any other networking solution helps to plan, plot, execute, and operate networks. A distinguishable fact is, that the intent based network can translate the “intent” of a user. The user may be the end-user or the network administrator. The “intent” may be through a series of commands or through APIs. How does it work?— a network admin describes the state of a network, or the policy they want a network to follow, the IBNS validates whether such a request is feasible and then executes it. The second differentiating feature of an IBNS is the ability to perform automated implementation of the network state and policies described above. They can also control the security level to be applied to applications based on user role, time, and device conditions. Additionally, the vLANs, mandate firewalls and other network technologies required within the network can be automated in an IBNS.   An IBNS can monitor a network in real-time and take necessary actions to assess and improve the network. The ultimate goal of an IBNS is, to ensure that the desired condition of a network is maintained while making sure an automated reparative action is taken in case of any discrepancy. Why networks need machine learning to create intent? Traditional Networks cannot be integrated into networking environments after a certain number of layers. They require manual integration which can be a daunting process demanding advanced skill-set. Traditional network systems are also prone to failover and capacity augment as they involve managing individual devices. This resulted in developing a new set of centrally globalized network system, controlled from one place that permeates through the network. Machine learning was considered a possible contender for aiding the development of such networks and hence IBNS came to be. The idea of IBNS was proposed long ago, but requirements for developing such a system were not available.  In the recent times, advances in ML have made it possible to develop such a system which can automate tasks. By automation we mean, automating the entire network policy when the network administrator simply defines the desired state of the network.   With the presence of ML, IBNS can do real time analysis of network conditions, which  involves monitoring, identifying and reacting to real time data using algorithmic validations. Why IBNS could be the future of networking? Central control system The intent based network uses a central, cloud based control system to program the routes to be taken by the network elements to deliver required information. The control system can also reprogram them in case of a network path failure. Automation of network elements Intent Based networks are all about reducing human efforts. A network is built on millions of connected devices. Managing a huge number of devices will require a lot of manpower. Also, the current networking system will have difficulty in managing enormous amount of data exploding from IoT devices. With an IBNS, human intervention required to build these network elements across services and circuits is reduced by a considerable amount. IBNS provides an automated network management system which mathematically validates the desired state of networks as specified by network administrators. Real-time decision making The network systems continuously gather information about the network architecture while adapting themselves to the desired condition. If a network is congested at a particular time frame, it can modify the route. For example, the GPS system can re-route a driver if he/she encounters some sort of traffic jam or a closed lane.  IBNS can also gather information about the current status of the network by consistent ingestion of real-time state pointers in the networking system. The real time data, so gathered can in turn make it smarter and more predictive.   Business benefits gained Intent-based software will open better opportunities for IT employees.  Due to elimination of mundane operational tasks, IT admins can increase their productivity levels while improving their skill set. For an organization, it would mean lower cost of operation, reduced fallacy, better reliability & security, resource management, increased optimization, and multi-vendor device management. What do we see around the globe in the IBNS domain? The rising potential of IBNS is giving startups and MNCs higher opportunities to invest in this technology.   Apstra have launched their Apstra Operating system which provides deployment of an IBN system to a variety of customers including Cisco, Arista, Juniper, and other white box users. The solution focuses on a single data center. It prevents and repairs network outages to improve infrastructure agility. The AOS also provides network admins full transparency to intent with the ability to ask any question about the network, amend any aspect of the intent after deployment, while allowing complete freedom for users to perform customizations of the AOS. Anuta networks, the pioneer in Network service orchestration have announced their award winning NCX intent-based orchestration platform to automate network services, which has enabled DevOps for networking. They provide more than 35 industry leading vendors with orchestrated devices. They also own detailed and exhaustive REST API for network integration. One of the top pioneers in the networking domain, Cisco has launched itself in the intent-based networking space with the solution—Software Defined Access (SDA).  The software enables easy to perform operational tasks and build powerful networks by automating the network. It also boasts of features like: Multi-site management Improved visibility with simple topological views Integration with Kubernetes Zero -trust security With tech-innovation taking place all around, networking organizations are racing towards breaking the clutter by bringing their intent based networks to the networking space. According to Gartner analyst, Andrew Lerner, “Intent-based networking is nascent, but could be the next big thing in networking, as it promises to improve network availability and agility, which are key, as organizations transition to digital business. I&O leaders responsible for networking need to determine if and when to pilot this technology.”

Long gone are the days when people would go out exploring one store to another, to buy that beautiful pink dress or a particular pair of shoe.The e-commerce revolution, has surged online shopping drastically. How many time have we heard physical stores are dying. And yet they seem to have a cat-like hold on their lifespan. Wonder why? Because not everyone likes shopping online. We are aware of a group of people who still prefer to buy from the brick and mortar structure. They are like the doubting Thomas, remember the touch and believe concept? For customers who love shopping physically in a store, retailers strive to create a fascinating shopping experience in a way that online platform cannot offer. This is especially important for them to increase sale and generate profits on peak festive seasons such as Black Fridays or New Year’s. A lot has been talked about the wonders retail analytics data can do for e-commerce sites, read this article for instance. But not as much is talked about traditional stores. So, here we have listed down 10 retail analytics options for offline retailers, to capture maximum customer attention and retention. 1. Location analytics and proximity marketing A large number of retail stores collect data to analyze the volume of customers buying online and offline. They use this data for internal retail analytics, which helps them in merchandise tracking, adjust staffing levels, monitor promotions and so on. Retailers benefit from location analytics in order to detect a section of the store with high customer traffic. Proximity marketing uses location-based technology to communicate with customers through their smartphones. Customers receive targeted offers and discounts based on their proximity to the product. For instance, a 20% off on the floral dress to your right. Such on-the-go attractive deals have a higher likelihood of resulting in a customer sale. Euclid Analytics provides solutions to track the buying experience of every visitor in the store. This helps retailers retarget and rethink their strategies to influence sales at an individual customer level. 2. Music systems Nowadays, most large retail formats have music systems set up for the customers. A playlist with a mixed genre of music is an ideal fit for various customers visiting the store.  Retailers use the tactic right, with a correct tempo, volume, and genre to uplift customer’s mood resulting in a purchase. They also have to keep in mind the choice of a playlist. As, music preferences differ with generation, behavior, and the mood of the customer.  Store owners opt for music services with a customized playlist to create an influential buying factor. Atmos select provides a customized music service to retailers. It takes customer demographics into consideration to draft an audio branding strategy.  It is then used to design an audio solution for most of the retail outlets and stores. 3. Guest WiFi Guest wifi benefits customers by giving them free internet connection while they shop. Who would not want that? However, not only customers but retailers too benefit with such an offering. An in-store wifi provides them with detailed customer analytics and enables to track various shopping patterns. Cloud4Wi, offers Volare guest Wi-Fi, which provides free wi-fi service to customers within the retail store. It provides a faster and easier login option, to connect to the wifi. It also collects customer’s data for retailers to provides unique and selective marketing list. 4. Workforce tools Unison among the staff members within the work environment creates positivity in store. To increase communication between the staff, workforce tools are put to use. These are various messaging applications and work-planner platforms that help in maintaining a rapport among the staff members. It helps empower employees to maintain their work-life, check overtime details, attendance, and more. Branch, a tool to improve workforce productivity, helps internal messaging networks and also notifies employees about their shift timing, and other details. 5. Omnichannel retail analytics Omnichannel retail enables customer with an interactive and seamless shopping experience across platforms. Additionally,  with the data collected from different digital channels, retailers get an overview of customer’s shopping journey and the choices they made over time. Omnichannel analytics also assists them to showcase personalized shopping ads based on customer’s social media habits. Intel offers solutions for Omnichannel analytics which helps retailers increase customer loyalty and generate substantial revenue growth. 6. Dressing Room Technology The mirror within the trial room knows it all! Retailers can attract maximum customer traffic with the mirror technology. It is an interactive, touch screen mirror that allows customers to request new items and adjust the lights in the trial room. The mirror can also sense products that the customer brings in, using the RFID technology, and recommends similar products. It also assists them in saving products to their online accounts-- in case they decide to purchase them later--or digitally seek assistance from the store associate. Oak Labs, has created one such mirror which transforms customer trial room experience while bridging the gap between technology and retail. 7. Pop-ups and kiosks Pop-ups are mini-outlets for large retail formats, set up to sell a seasonal product. Whereas kiosks are temporary alternatives for retailers, to attract a high number of footfalls in store. Both pop-ups and kiosks benefit shoppers with the choice of self-service. They get an option to shop from the store’s physical as well as online product offering. They not only enable secure purchase but also deliver orders to your doorstep. Such techniques attract customers to choose retail shopping over online shopping. Withme, a startup firm that offers a platform to set up POP ups for retail outlets and brands.   8. Inventory management Managing the inventory is a major task for a store manager - to place the right product in the right place at the right time. Predictive analytics helps optimize inventory management for proper allocation, and replenishment process. It also equips retailers to markdown the inventory for clearance to reload a new batch. Celect, an inventory management startup helps retailers to analyze customer preferences and simultaneously map future demand for the product. It also helps in extraction of existing data from the inventory to gain meaningful insights. Such insights can then be taken into account for the faster sale of inventory and to get a detailed retail analytics based sales report. 9.  Smart receipts and ratings Retailers continuously aim to provide better quality service to the customer. Receiving a 5-star rating for their service in return is like a cherry on the cake.  For higher customer engagement, retailers offer smart receipts, which helps retailers collect customer email addresses to send promotional offers or festive sale discounts. Retailers also provide customers with personalized offerings and incentives in order to attract customer revisitation. To know how well retailers have fared in providing services, they set up a digital kiosk at the checkout area, where in-store customers can rate retailers based on the shopping experience. Startup firms such as TruRating aid retailers with a rating mechanism for shoppers at the checkout. FlexReceipts helps retailers to set up smart receipt application for the customers. 10. Shopping cart tech Retailers can now provide a next-gen shopping cart to their customers. A technology that can guide customer’s in-store shopping journey with a tablet-equipped shopping cart. The tablet uses machine vision to keep a track of the shelves, as the cart moves within the store. It also displays digital-ads to promote each product, the shopping cart passes through. Focal Systems build powerful technical assistance for retailers, which can give tough competition to their online counterparts. Online shopping is convenient but more often than not we still crave for the look and feel of a product and the immersive shopping experience especially during holidays and festive occasions. And that’s the USP of a Brick and Mortar shop. Offline retailers who know their data and know how to leverage retail analytics using advances in machine learning and retail tech stand a chance to provide their customers with a shopping experience superior to their online counterparts.

Today, machine learning in Python has become almost synonymous with scikit-learn. The "Big Bang" moment for scikit-learn was in 2007 when a gentleman named David Cournapeau decided to write this project as part of Google Summer of Code 2007. Let's take a moment to thank him. Matthieu Brucher later came on board and developed it further as part of his thesis. From that point on, sklearn never looked back. In 2010, the prestigious French research organization INRIA took ownership of the project with great developers like Gael Varoquaux, Alexandre Gramfort et al. starting work on it. Here's the oldest pull request I could find in sklearn’s repository. The title says "we're getting there"! Starting from there to today where sklearn receives funding and support from Google, Telecom ParisTech and Columbia University among others, it surely must’ve been quite a journey. Sklearn is an open source library which uses the BSD license. It is widely used in industry as well as in academia. It is built on Numpy, Scipy and Matplotlib while also having wrappers around various popular libraries such LIBSVM. Sklearn can be used “out of the box” after installation. Can I trust scikit-learn? Scikit-learn, or sklearn, is a very active open source project having brilliant maintainers. It is used worldwide by top companies such as Spotify, booking.com and the like. That it is open source where anyone can contribute might make you question the integrity of the code, but from the little experience I have contributing to sklearn, let me tell you only very high-quality code gets merged. All pull requests have to be affirmed by at least two core maintainers of the project. Every code goes through multiple iterations. While this can be time-consuming for all the parties involved, such regulations ensure sklearn’s compliance with the industry standard at all times. You don’t just build a library that’s been awarded the “best open source library” overnight! How can I use scikit-learn? Sklearn can be used for a wide variety of use-cases ranging from image classification to music recommendation to classical data modeling. Scikit-learn in various industries: In the Image classification domain, Sklearn’s implementation of K-Means along with PCA has been used for handwritten digit classification very successfully in the past. Sklearn has also been used for facial/ faces recognition using SVM with PCA. Image segmentation tasks such as detecting Red Blood Corpuscles or segmenting the popular Lena image into sections can be done using sklearn. A lot of us here use Spotify or Netflix and are awestruck by their recommendations. Recommendation engines started off with the collaborative filtering algorithm. It basically says “if people like me like something, I’ll also most probably like that.” To find out users with similar tastes, a KNN algorithm can be used which is available in sklearn. You can find a good demonstration of how it is used for music recommendation here. Classical data modeling can be bolstered using sklearn. Most people generally start their kaggle competitive journeys with the titanic challenge. One of the better tutorials out there on starting out is by dataquest and generally acts as a good introduction on how to use pandas and sklearn (a lethal combination!) for data science. It uses the robust Logistic Regression, Random Forest and the Ensembling modules to guide the user. You will be able to experience the user-friendliness of sklearn first hand while completing this tutorial. Sklearn has made machine learning literally a matter of importing a package. Sklearn also helps in Anomaly detection for highly imbalanced datasets (99.9% to 0.1% in credit card fraud detection) through a host of tools like EllipticEnvelope and OneClassSVM. In this regard, the recently merged IsolationForest algorithm especially works well in higher dimensional sets and has very high performance. Other than that, sklearn has implementations of some widely used algorithms such as linear regression, decision trees, SVM and Multi Layer Perceptrons (Neural Networks) to name a few. It has around 39 models in the “linear models” module itself! Happy scrolling here! Most of these algorithms can run very fast compared to raw python code since they are implemented in Cython and use Numpy and Scipy (which in-turn use C) for low-level computations. How is sklearn different from TensorFlow/MLllib? TensorFlow is a popular library to implement deep learning algorithms (since it can utilize GPUs). But while it can also be used to implement machine learning algorithms, the process can be arduous. For implementing logistic regression in TensorFlow, you will first have to “build” the logistic regression algorithm using a computational graph approach. Scikit-learn, on the other hand, provides the same algorithm out of the box however with the limitation that it has to be done in memory. Here's a good example of how LogisticRegression is done in Tensorflow. Apache Spark’s MLlib, on the other hand, consists of algorithms which can be used out of the box just like in Sklearn, however, it is generally used when the ML task is to be performed in a distributed setting. If your dataset fits into RAM, Sklearn would be a better choice for the task. If the dataset is massive, most people generally prototype on a small subset of the dataset locally using Sklearn. Once prototyping and experimentation are done, they deploy in the cluster using MLlib. Some sklearn must-knows Scikit-learn can be used for three different kinds of problems in machine learning namely supervised learning, unsupervised learning and reinforcement learning (ahem AlphaGo). Unsupervised learning happens when one doesn’t have ‘y’ labels in their dataset. Dimensionality reduction and clustering are typical examples. Scikit-learn has implementations of variations of the Principal Component Analysis such as SparsePCA, KernelPCA, and IncrementalPCA among others. Supervised learning covers problems such as spam detection, rent prediction etc. In these problems, the ‘y’ tag for the dataset is present. Models such as Linear regression, random forest, adaboost etc. are implemented in sklearn. From sklearn.linear_models import LogisticRegression Clf = LogisticRegression().fit(train_X, train_y) Preds = Clf.predict(test_X) Model evaluation and analysis Cross-validation, grid search for parameter selection and prediction evaluation can be done using the Model Selection and Metrics module which implements functions such as cross_val_score and f1_score respectively among others. They can be used as such: Import numpy as np From model_selection import cross_val_score From sklearn.metrics import f1_score Cross_val_avg = np.mean(cross_val_score(clf, train_X, train_y, scoring=’f1’)) # tune your parameters for better cross_val_score # for model results on a certain classification problem F_measure = f1_score(test_y, preds) Model Saving Simply pickle your model using pickle.save and it is ready to be distributed and deployed! Hence a whole machine learning pipeline can be built easily using sklearn. Finishing Remarks There are many good books out there talking about machine learning, but in context to Python,  Sebastian Raschka`s  (one of the core developers on sklearn) recently released his book titled “ Python Machine Learning” and it’s in great demand. Another great blog you could follow is Erik Bernhardsson’s blog. Along with writing about machine learning, he also discusses software development and other interesting ideas. Do subscribe to the scikit-learn mailing list as well. There are some very interesting questions posted there and a lot of learnings to take home. The machine learning subreddit also collates information from a lot of different sources and is thus a good place to find useful information. Scikit-learn has revolutionized the machine learning world by making it accessible to everyone. Machine learning is not like black magic anymore. If you use scikit-learn and like it, do consider contributing to sklearn. There is a huge clutter of open issues and PRs on the sklearn GitHub page. Scikit-learn needs contributors! Have a look at this page to start contributing. Contributing to a library is easily the best way to learn it! [author title="About the Author"]Devashish Deshpande started his foray into data science and machine learning in 2015 with an online course when the question of how machines can learn started intriguing him. He pursued more online courses as well as courses in data science during his undergrad. In order to gain practical knowledge he started contributing to open source projects beginning with a small pull request in Scikit-Learn. He then did a summer project with Gensim and delivered workshops and talks at PyCon France and India in 2016. Currently, Devashish works in the data science team at belong.co, India. Here's the link to his GitHub profile.[/author]

In the digital disruption era, Artificial Intelligence in Fintech is viewed as an emerging technology forming the sole premise for revolution in the sector. Tech giants positioned in the Fortune’s 500 technology list such as Apple, Microsoft, Facebook are putting resources in product innovations and technology automation. Businesses are investing hard to bring agility, better quality and high end functionality for driving their revenue growth by multi digits. Widely used AI-powered applications such as Virtual Assistants, Chatbots, Algorithmic Trading and Purchase Recommendation systems are fueling up the businesses with low marginal costs, growing revenues and providing a better customer experience. According to a survey, by National Business Research Institute, more than 62% of the companies will deploy AI powered fintech solutions in their applications to identify new opportunities and areas to scale the business higher. What has led the disruption? The Financial sector is experiencing a faster technological evolution right from providing personalized financial services, executing smart operations to simplify the complex and repetitive process. Use of machine learning and predictive analytics has enabled financial companies to provide smart suggestions on buying and selling stocks, bonds and commodities. Insurance companies are accelerating in automating their loan applications, thereby saving umpteen number of hours. Leading Investment Bank, Goldman Sachs automated their stock trading business replacing their trading professionals with computer engineers. Black Rock, one of the world’s largest asset management company facilitates high net worth investors with automated advice platform superseding highly paid wall street professionals. Applications such as algorithmic trading, personal chatbots, fraud prevention & detection, stock recommendations, and credit risk assessment are the ones finding their merit in banking and financial services companies.   Let us understand the changing scenarios with next-gen technologies: Fraud Prevention & Detection Fraud prevention is tackled by the firms using an anomaly detection API. The API is designed using machine learning & deep learning mechanism. It helps identify and report any suspicious or fraudulent activity taking place among-st billions of transactions on a daily basis. Fintech companies are infusing huge capital to handle cyber-crime, resulting into a global market spends of more than 400 billion dollars annually. Multi-national giants such as MasterCard, Sun Financial, Goldman Sachs, and Bank of England use AI-powered systems to safeguard and prevent money laundering, banking frauds and illegal transactions. Danske Bank, a renowned Nordic-based financial service provider, deployed AI engines in their operations helping them investigate millions of online banking transactions in less than a second. With this, cost of fraud investigation and delivering faster actionable insights reduced drastically. AI Powered Chatbots Chatbots are automated customer support chat applications powered by Natural Language Processing (NLP). They help deliver quick, engaging, personalized, and effective conversation to the end user. With an upsurge in the number of investors and varied investment options, customers seek financial guidance, profitable investment options and query resolution, faster and in real-time. Large number of banks such as Barclays, Bank of America, JPMorgan Chase are widely using AI-supported digital Chatbots to automate their client support, delivering effective customer experience with smarter financial decisions. Bank of America, the largest bank in US launched Erica, a Chatbot which guides customers with investment option notification, easy bill payments, and weekly update on their mortgage score.  MasterCard offers a chatbot to their customers which not only allows them to review their bank balance or transaction history but also facilitates seamless payments worldwide. Credit Risk Management For money lenders, the most common business risk is the credit risk and that piles up largely due to inaccurate credit risk assessment of borrowers. If you are unaware of the term credit risk, it is simply a risk associated with a borrower defaulting to repay the loan amount. AI backed Credit Risk evaluation tools developed using predictive analytics and advanced machine learning techniques has enabled bankers and financial service providers to simplify the borrower’s credit evaluation thereby transforming the labor intensive scorecard assessment method. Wells Fargo, an American international banking company adopted AI technology in executing mortgage verification and loan processing. It resulted in lower market exposure risk of their lending assets. With this, the team was able to establish smarter and faster credit risk management functionality. It resulted in analysis of millions of structured and unstructured data points for investigation thereby proving AI as an extremely valuable asset for credit security and assessment. Algorithmic Trading More than half a dozen US citizens own individual stocks, mutual funds, and exchange-traded mutual funds. Also, a good number of users trade on a daily basis, making it imperative for major broking and financial trading companies to offer AI powered algorithmic trading platform. The platform enables customers with strategic execution of trades offering significant returns. The algorithms analyse hundreds of millions of data pointers and draw down a decisive trading pattern enabling traders to book higher profits every microsecond of the trading hour. France-based international bank BNP Paribas deployed algorithmic trading which aids their customers in executing trades strategically and provides graphical representation of stock market liquidity. With the help of this, customers are able to determine the most appropriate ways of executing trade under various market conditions. The advances in automated trading has assisted users with suggestions and rich insights, helping humans to take better decisions. How do we see the Future of AI in Financial sector? The influence of AI in fintech has marked disruption in almost each and every financial institution, right from investment banks to retail banking, to small credit unions. Data science and machine learning practitioners are endeavoring to position AI as an essential part of the banking ecosystem. Financial companies are synergizing with data analytics and fintech professionals to orient AI as the primary interface for interaction with their customers. However, the sector commonly faces challenges in adoption of emerging technologies, making it inevitable for AI too. The foremost challenge companies face is availability of massive data which is clean and rich to train machine learning algorithms. The next hurdle in line would be the reliability and accuracy of the data insights provided by the AI mechanized solution. With dynamic market situation, businesses could experience decline in efficacy of their models causing serious harm to the company. Hence, they need to be smarter and cannot solely trust the AI technology in achieving the business mission. Absence of emotional intelligence in Chatbots is another area of concern resulting in an unsatisfactory customer service experience. While there may be other roadblocks, the rising investment in AI technology would definitely assist financial companies in overcoming such challenges and developing competitive intelligence in their product offerings. Predicting the near future, adoption of cutting edge technologies such as machine learning and predictive analytics will boost higher customer engagement, exceptional banking experience, lesser frauds and higher operating margins for banks, financial institutions and Insurance companies.

We have come a long way from what ecommerce looked like two decades ago. From a non-existent entity, it has grown into a world-devouring business model that is a real threat to the traditional retail industry. It has moved from a basic static web page with limited product listings to a full grown virtual marketplace where anyone can buy or sell anything from anywhere at anytime at the click of a button. At the heart of this transformation are two things: customer experience and technology. This is what Jeff Bezos, founder & CEO of Amazon, one of the world’s largest ecommerce sites believes: “We see our customers as invited guests to a party, and we are the hosts. It's our job every day to make every important aspect of the customer experience a little bit better.” Now with the advent of AI, the retail space especially e-commerce is undergoing another major transformation that will redefine customer experiences and thereby once again change the dynamics of the industry. So, how is AI-powered ecommerce actually changing the way shoppers shop? AI-powered ecommerce makes search easy, accessible and intuitive Looking for something? Type it! Say it!...Searching for a product you can’t name? No worries. Just show a picture. "A lot of the future of search is going to be about pictures instead of keywords." - Ben Silbermann, CEO of Pinterest We take that statement with a pinch of salt. But we are reasonably confident that a lot of product search is going to be non-text based. Though text searches are common, voice and image searches in e-commerce are now gaining traction. AI makes it possible for the customer to move beyond simple text-based product search and search more easily and intuitively through voice and visual product searches. This also makes search more accessible. It uses Natural Language Processing to understand the customer’s natural language, be it in text or speech to provide more relevant search results. Visual product searches are made possible through a combination of computer vision, image recognition, and reverse image search algorithms.   Amazon Echo, a home-automated speaker has a voice assistant Alexa that helps customers to buy products online by having simple conversations with Alexa. Slyce, uses a visual search feature, wherein the customer can scan a barcode, a catalog, and even a real image; just like Amazon’s in-app visual feature. Clarifai helps developers to build applications that detect images and videos and searches related content. AI-powered ecommerce makes personalized product recommendations   When you search for a product, the AI underneath recommends further options based on your search history or depending on what other users who have similar tastes found interesting. Recommendations engines employ one or a combination of the three types of recommendation algorithms: content-based filtering, collaborative filtering, and complementary products. The relevance and accuracy of the results produced depend on various factors such as the type of recommendation engine used, the quantity and quality of data used to train the system, the data storage and retrieval strategies used amongst others. For instance, Amazon uses DSSTNE (Deep Scalable Sparse Tensor Network Engine, pronounced as Destiny) to make customized product recommendations to their customers. The customer data collected and stored is used by DSSTNE to train and generate predictions for customers. The data processing itself takes place on CPU clusters whereas the training and predictions take place on GPUs to ensure speed and scalability. Virtual Assistants as your personal shopping assistants   Now, what if we said you can have all the benefits we have discussed above without having to do a lot of work yourself? In other words, what if you had a personal shopping assistant who knows your preferences, handles all the boring aspects of shopping (searching, comparing prices, going through customers reviews, tracking orders etc.) and brought you products that were just right with the best deals? Mona, one such personal shopper, can do all of the above and more. It uses a combination of artificial intelligence and big data to do this. Virtual assistants can either be fully AI driven or a combination of AI-human collaboration. Chatbots also assist shoppers but within a more limited scope. They can help resolve customer queries with zero downtime and also assist in simple tasks such as notify the customer of price changes, place and track orders etc. Dominos has a facebook messenger Bot that enables customers to order food. Metail, an AI-powered ecommerce website, take in your body measurements. With this, you can actually see how a clothing would look on you. Botpress helps developers to build their own chatbots consuming lesser time. Maximizing CLV (customer lifetime value) with AI-powered CRM AI-powered ecommerce in CRM aims to help businesses predict CLV and sell the right product to the right customer at the right time, every time leveraging the machine learning and predictive capabilities of AI. It also helps businesses provide the right level of customer service and engagement. In other words, by combining the predictive capabilities and automated 1-1 personalization, an AI backed CRM can maximize CLV for every customer!    Salesforce Einstein, IBM Watson are some of the frontrunners in this space. IBM Watson, with its cognitive touch, helps ecommerce sites analyze their mountain of customer data and glean useful insights to predict a lot of things like what customers are looking for, the brands that are popular, and so on.  It can also help with dynamic pricing of products by predicting when to discount and when to increase the price based on analyzing demand and competitions’ pricing tactics. It is clear that AI not only has the potential to transform e-commerce as we know it but that it has already become central to the way leading ecommerce platforms such as Amazon are functioning. Intelligent e-commerce is here and now. The near future of ecommerce is omnicommerce driven by the marriage between AI and robotics to usher in the ultimate customer experience - one that is beyond our current imagination.

Mixed Reality has become a disruptive force that is bridging the gap between reality and imagination. With AI, it is now poised to change the world as we see it! The global mixed reality market is expected to reach USD 6.86 billion by 2024. Mixed Reality has found application in not just the obvious gaming and entertainment industries but also has great potential in business and other industries ranging from manufacturing, travel, and medicine to advertising. Maybe that is why the biggest names in tech are battling it out to capture the MR market with their devices - Microsoft HoloLens, GoogleGlass 2.0, Meta2 handsets to name a few. Incorporating Artificial Intelligence is their next step towards MR market domination. So what’s all the hype about MR and how can AI take it to the next level? Through the looking glass: Understanding Mixed Reality Mixed reality is essentially a fantastic concoction of virtual reality (a virtual world with virtual objects) and augmented reality (the real world with digital information). This means virtual objects are overlaid in the real world and mixed reality enables the person who is experiencing the MR environment to perceive virtual objects as ”real ones”. While in augmented reality it is easy to break the illusion and recognize that the objects are not real (Hello Pokemon Go!), in Mixed Reality, it is harder to break the illusion as virtual objects behave like real-world objects. So when you lean in close or interact with a virtual object in MR, it gets closer in a way a real object would. The MR experience is made possible with mixed reality devices that are typically lightweight and wearable. They are generally equipped with front-mounted cameras to recognize the distinctive features of the real world (such as objects and walls) and blend them with the virtual reality as seen through the headset. They also include a processor for processing the information relayed by an array of sensors embedded in the headset. These processors run the algorithms used for pattern recognition on a cloud-based server. These devices then use a projector for displaying virtual images in real environments which are finally reflected to the eye with the help of beam-splitting technology. All this sounds magical already, what can AI do for MR to top this? Curiouser and curiouser: The AI-powered Mixed Reality Mixed Reality and Artificial Intelligence are two powerful technology tools. The convergence of the two means a seamless immersive experience for users that blends the virtual and physical reality. Mixed Reality devices already enable interaction of virtual holograms in the physical environment and thereby combine virtual worlds with reality. But most MR devices require a large number of calculations and adjustments to accurately determine the position of a virtual object in a real-world scenario. They then apply rules and logic to those objects to make them behave like real-world objects. As these computations happen on the cloud, the results have perceivable time lag which comes in the way of giving the user a truly immersive experience. Also, user mobility is restricted due to current device limitations. Recently there has been a rise of the AI coprocessor in Mixed Reality devices. The announcement of Microsoft’s HoloLens 2 project, an upgrade to the existing MR device which now includes an AI coprocessor is a case in point. By using AI chips for computing, the above calculations, for example, MR devices will deliver high precision results faster. It means algorithms and calculations can run instantaneously without the need for data to be sent to/from a cloud. Having the data locally on your headset will eliminate time lag, thereby creating more real-time immersive experiences. In other words, as the visual data is analyzed directly on the device and computationally-exhaustive tasks are performed close to the data source, the enhanced processing speed results in quicker performance. Since the data remains on your headset always, fewer computations are needed to be performed on the cloud, hence the data is more secure. Using an AI chip also allows flexible implementation of deep neural networks. They help in automating complex calculations such as depth perception estimations and generally provide a better understanding of the environment to the MR devices. Generative models from Deep Learning can be used to generate believable virtual characters (avatars) in the real world. Images can also be more intelligently compressed with AI techniques. This would enable faster transmission over wireless networks. Motion capture techniques are now employing AI functionalities such as phase-functioned neural networks and self-teaching AI. They use machine learning techniques to combine a vast library of stored movements and combine and fit them into new characters. By using AI-powered Mixed Reality devices, the plan is to provide a more realistic experience which is fast and provides more mobility. The ultimate goal is to build AI-powered mixed reality devices that are intelligent and self-learning. Follow the white rabbit: Applications of AI-power Mixed Reality  Let us look at various sectors where Artificially Intelligent Mixed Reality has started finding traction. Gaming and Entertainment In the field of gaming, procedural content generation techniques allow automatic generation of Mixed Reality games (as opposed to manual creation by game designers) by encoding elements such as individual structures, enemies etc with their relationships. Artificial Intelligence enhances PCG algorithms in object identification and in recognizing other relationships between the real and virtual objects. Deep learning techniques can be used for tasks like super resolution, photo to texture mapping, and texture multiplication. Healthcare and Surgical Procedures AI-powered mixed reality tech has also found its use in the field of Healthcare and surgical operations. Scopis has announced a mixed reality surgical navigation system that uses the Microsoft HoloLens for spinal surgery applications. It employs image recognition and manipulation techniques which allow the surgeon to see both the patient and a superimposed image of the pedicle screws (used for vertebrae fixation surgeries) for surgical procedures. Retail Retail is another sector which is under the spell of this AI infused MR tech. DigitalBridge, a mixed-reality company uses mixed reality, artificial intelligence, and deep learning to create a platform that allows consumers to virtually try on home decor products before buying them. Image and Video Manipulation AI algorithms and MR techniques can also enrich video and image manipulation. As we speak, Microsoft is readying the release of Microsoft Remix 3D service. This software adds “mixed reality” digital images and animations to videos. It keeps the digital content in the same position in relation to the real objects using image recognition, computer vision, and AI algorithms. Military and Defence AI-powered Mixed Reality is also finding use in the defense sector, where MR training simulations controlled by artificially intelligent software combine real people and physical environments with a virtual setup. Construction and Homebuilding Builders can visualize their options in life-sized models with MR devices. With AI, they can leave virtual messages or videos at key locations to keep other technicians and architects up to date when they’re away. Using MR and AI techniques, an architect can call a remote expert into the virtual environment if need be, and virtual assistants can be utilized for further assistance. COINS is a construction and homebuilding organization which uses AI-powered Mixed Reality devices. They are collaborating with Sketchup for virtual messaging and 3D modeling and Microsoft for HoloLens and Skype Chatbot assistance. Industrial Farming Machine learning algorithms can be used to study a sensor-enabled field of crop and record the growth, requirements and anticipate future needs. An MR device can then provide a means to interact with the plants and analyze present conditions all the while adjusting future needs. Infosys’ Plant.IO is one such digital farm which when combined with the power of an MR device can overlay virtual objects over a real-world scenario. Conclusion Through these examples, we can see the rapid adoption of AI-powered Mixed Reality recipes across diverse fields enabled by the rise of AI chips and the employment of more exhaustive computations with complex machine learning and deep learning algorithms. The next milestone is to see the rise of a Mixed Reality environment which is completely immersive and untethered. This would be made possible by the adoption of more complex AI techniques and advances made in the field of AI both in hardware and software. Instead of voice or search based commands in the MR environments, AI techniques will be used to harness eye and body gestures. As MR devices become smaller and mobile, AI-powered mixed reality will also give rise to intelligent application development incorporating mixed reality through the phone’s camera lens. AI technologies would thus help expand the scope of MR not only as an interesting tool, with applications in gaming and entertainment, but also as a practical and useful approach for how we see, interact and learn from our environment.

We are nearing the end of 2017. But with each ending chapter, we have remarkable achievements to be thankful for. Similarly, for the data science community, this year was filled with a number of new technologies, tools, version updates etc. 2017 saw blockbuster releases such as PyTorch, TensorFlow 1.0 and Caffe 2, among many others. We invite data scientists, machine learning experts, and other data science professionals to come together on this Thanksgiving Day, and thank the organizations, which made our interactions with AI easier, faster, better and generally more fun. Let us recall our blessings in 2017, one month at a time... [dropcap]Jan[/dropcap] Thank you, Facebook and friends for handing us PyTorch Hola 2017! While the world was still in the New Year mood, a brand new deep learning framework was released. Facebook along with a few other partners launched PyTorch. PyTorch came as an improvement to the popular Torch framework. It now supported the Python language over the less popular Lua. As PyTorch worked just like Python, it was easier to debug and create unique extensions. Another notable change was the adoption of a Dynamic Computational Graph, used to create graphs on the fly with high speed and flexibility. [dropcap]Feb[/dropcap] Thanks Google for TensorFlow 1.0 The month of February brought Data Scientist’s a Valentine's gift with the release of TensorFlow 1.0. Announced at the first annual TensorFlow Developer Summit, TensorFlow 1.0 was faster, more flexible, and production-ready. Here’s what the TensorFlow box of chocolate contained: Fully compatibility with Keras Experimental APIs for Java and Go New Android demos for object and image detection, localization, and stylization A brand new Tensorflow debugger An introductory glance of  XLA--a domain-specific compiler for TensorFlow graphs [dropcap]Mar[/dropcap] We thank Francois Chollet for making Keras 2 a production ready API Congratulations! Keras 2 is here. This was a great news for Data science developers as Keras 2, a high- level neural network API allowed faster prototyping. It provided support both CNNs (Convolutional Neural Networks) as well as RNNs (Recurrent Neural Networks). Keras has an API designed specifically for humans. Hence, a user-friendly API. It also allowed easy creation of modules, which meant it is perfect for carrying out an advanced research. Developers can now code in  Python, a compact, easy to debug language. [dropcap]Apr[/dropcap] We like Facebook for brewing us Caffe 2 Data scientists were greeted by a fresh aroma of coffee, this April, as Facebook released the second version of it’s popular deep learning framework, Caffe. Caffe 2 came up as a easy to use deep learning framework to build DL applications and leverage community contributions of new models and algorithms. Caffe 2 was fresh with a first-class support for large-scale distributed training, new hardware support, mobile deployment, and the flexibility for future high-level computational approaches. It also provided easy methods to convert DL models built in original Caffe to the new Caffe version. Caffe 2 also came with over 400 different operators--the basic units of computation in Caffe 2. [dropcap]May[/dropcap] Thank you, Amazon for supporting Apache MXNet on AWS and Google for your TPU The month of May brought in some exciting launches from the two tech-giants, Amazon and Google. Amazon Web Services’ brought Apache MXNet on board and Google’s Second generation TPU chips were announced. Apache MXNet, which is now available on AWS allowed developers to build Machine learning applications which can train quickly and run anywhere, which means it is a scalable approach for developers. Next up, was Google’s  second generation TPU (Tensor Processing Unit) chips, designed to speed up machine learning tasks. These chips were supposed to be (and are) more capable of CPUs and even GPUs. [dropcap]Jun[/dropcap] We thank Microsoft for CNTK v2 The mid of the month arrived with Microsoft’s announcement of the version 2 of its Cognitive Toolkit. The new Cognitive Toolkit was now enterprise-ready, had production-grade AI and allowed users to create, train, and evaluate their own neural networks scalable to multiple GPUs. It also included the Keras API support, faster model compressions, Java bindings, and Spark support. It also featured a number of new tools to run trained models on low-powered devices such as smartphones. [dropcap]Jul[/dropcap] Thank you, Elastic.co for bringing ML to Elastic Stack July made machine learning generally available for the Elastic Stack users with its version 5.5. With ML, the anomaly detection of the Elasticsearch time series data was made possible. This allows users to analyze the root cause of the problems in the workflow and thus reduce false positives. To know about the changes or highlights of this version visit here. [dropcap]Aug[/dropcap] Thank you, Google for your Deeplearn.js August announced the arrival of Google’s Deeplearn.js, an initiative that allowed Machine Learning models to run entirely in a browser. Deeplearn.js was an open source WebGL- accelerated JS library. It offered an interactive client-side platform which helped developers carry out rapid prototyping and visualizations. Developers were now able to use hardware accelerator such as the GPU via the webGL and perform faster computations with 2D and 3D graphics. Deeplearn.js also allowed TensorFlow model’s capabilities to be imported on the browser. Surely something to thank for! [dropcap]Sep[/dropcap] Thanks, Splunk and SQL for your upgrades September surprises came with the release of Splunk 7.0, which helps in getting Machine learning to the masses with an added Machine Learning Toolkit, which is scalable, extensible, and accessible. It includes an added native support for metrics which speed up query processing performance by 200x. Other features include seamless event annotations, improved visualization, faster data model acceleration, a cloud-based self-service application. September also brought along the release of MySQL 8.0 which included a first-class support for Unicode 9.0. Other features included are An extended support for native JSOn data Inclusion of windows functions and recursive SQL syntax for queries that were previously impossible or difficult to write Added document-store functionality So, big thanks to the Splunk and SQL upgrades. [dropcap]Oct[/dropcap] Thank you, Oracle for the Autonomous Database Cloud and Microsoft for SQL Server 2017 As Fall arrived, Oracle unveiled the World’s first Autonomous Database Cloud. It provided full automation associated with tuning, patching, updating and maintaining the database. It was self scaling i.e., it instantly resized compute and storage without downtime with low manual administration costs. It was also self repairing and guaranteed 99.995 percent reliability and availability. That’s a lot of reduction in workload! Next, developers were greeted with the release of SQL Server 2017 which was a major step towards making SQL Server a platform. It included multiple enhancements in Database Engine such as adaptive query processing, Automatic database tuning, graph database capabilities, New Availability Groups, Database Tuning Advisor (DTA) etc. It also had a new Scale Out feature in SQL Server 2017 Integration Services (SSIS) and SQL Server Machine Learning Services to reflect support for Python language. [dropcap]Nov[/dropcap] A humble thank you to Google for TensorFlow Lite and Elastic.co for Elasticsearch 6.0 Just a month more for the year to end!! The Data science community has had a busy November with too many releases to keep an eye on with Microsoft Connect(); to spill the beans. So, November, thank you for TensorFlow Lite and Elastic 6. Talking about TensorFlow Lite, a lightweight product  for mobile and embedded devices, it is designed to be: Lightweight: It allows inference of the on-device machine learning models that too with a small binary size, allowing faster initialization/ startup. Speed: The model loading time is dramatically improved, with an accelerated hardware support. Cross-platform: It includes a runtime tailormade to run on various platforms–starting with Android and iOS. And now for Elasticsearch 6.0, which is made generally available. With features such as easy upgrades, Index sorting, better Shard recovery, support for Sparse doc values.There are other new features spread out across the Elastic stack, comprised of Kibana, Beats and Logstash. These are, Elasticsearch’s solutions for visualization and dashboards, data ingestion and log storage. [dropcap]Dec[/dropcap] Thanks in advance Apache for Hadoop 3.0 Christmas gifts may arrive for Data Scientists in the form of General Availability of Hadoop 3.0. The new version is expected to include support for Erasure Encoding in HDFS, version 2 of the YARN Timeline Service, Shaded Client Jars, Support for More than 2 NameNodes, MapReduce Task-Level Native Optimization, support for Opportunistic Containers and Distributed Scheduling to name a few. It would also include a rewritten version of Hadoop shell scripts with bug fixes, improved compatibility and many changes in some existing installation procedures. Pheww! That was a large list of tools for Data Scientists and developers to thank for this year. Whether it be new frameworks, libraries or a new set of software, each one of them is unique and helpful to create data-driven applications. Hopefully, you have used some of them in your projects. If not, be sure to give them a try, because 2018 is all set to overload you with new, and even more amazing tools, frameworks, libraries, and releases.

Amazon and Microsoft, the pioneer tech giants have collaborated their efforts to bring in a compelling, easy, and powerful deep learning interface known as Gluon. If you are into physics, you must be aware of the term Gluon. Gluon is a hypothetical particle believed to be exchanged between quarks in order to bind them. If we go by the literal meaning, Gluon, similar to a glue, works as a binding agent. Having gained inspiration from this, Amazon and Microsoft have glued in their efforts to bring deep learning to a wider developer audience with the launch of Gluon. It is a simple, efficient, and compact API for deep learning. Why is Gluon essential? Any Neural network, has three important phases: First, the manual coding, where the developer explains the specific behaviour of the network. Then is the training phase where the error of the output is calculated and subsequently, the weights are adjusted. This activity requires memory and is computationally exhaustive. After the training phase, the network is used to make predictions. The process of building up a neural network is labor-intensive as well as time consuming. These networks have to be trained to parse large and complex data sets and therefore they are  usually  constructed manually. Thus, making them difficult to debug and reuse. Also, manual construction requires expertise and advanced skill-sets which are possessed by experienced data scientists. However, the reach of machine learning technique is at every doorstep now. A large number of developers are looking for solutions that can help them build deep learning models with ease and practicality without compromising on the power. Gluon is a flexible and approachable way to train and construct neural networks. It comes across as a more concise and easy-to-use programming interface providing developers the ability to quickly prototype, build, and train deep learning models without sacrificing performance. The API plays around with MXNet to reduce the complexity of deep learning making it reachable to a large number of developers. How is it different? A few compelling advantages that makes Gluon stand out: An Easy to Use API A strong differentiating feature of Gluon is  that it provides interface, in the form of an API. Making it easier for the developers to grasp and develop DL models with the help of modular components. This functionality is simpler  to comprehend than the formal neural net definition methods. Data Structure Approach Deep learning models in Gluon can be defined, flexed, and modified in a way similar to a data structure. This ability makes it a  familiar interface especially for  developers who have just recently  stepped into the machine learning world. Dynamic networks can be easily managed with Gluon as it mixes the programming models from TensorFlow (symbolic representations) and PyTorch (imperative definitions of networks). Network Defining Ability Gluon provides the ability to define the network. Thus, the dynamic adjustment of the network is possible during the definition and the training process. This essentially means that the training algorithm and the neural model can inform one another. Due to this, developers can make use of standard programming structures to build, and can also use sophisticated algorithms and models to advance neural nets. High Speed Training Friendly APIs and flexible approaches are all great, but they shouldn't be incurred at the cost of training speed. Gluon is better than the manual approach as it can perform  all of the tasks without compromising on performance while providing abstractions without losing out on training speed. This is because, Gluon blends the formal definitions and specific details of the network under the hood of a concise API, allowing users to implement models, rather than doing tasks like compiler optimizations manually. Easy algorithmic implementations using Gluon Gluon supports a wide range of prebuilt and optimized components for building neural networks. Developers can build deep learning models using the MXNet framework in the Gluon interface. Gluon allows building neural nets from predefined layers. It  can also keep a note of when to record or not to record the computation graph. It can invoke highly optimized layers written in C++. Training of parallel data can also be accomplished easily. As compared to other interfaces, Gluon can run code, faster on both CPUs and GPUs.  Also, movement from one to multiple devices and initializing network parameters over them is pretty easy. Even for a simple problem like a linear regression model, Gluon can help in writing quick, and clean code. For linear regression, it eliminates the need of allocating parameters individually, implementing a stochastic gradient descent, or defining a loss function. Subsequently, it also reduces the workload required for multiclass logistic regression. On similar terms, Gluon can be used to transform the logic of a logistic regression model to a multilayer perceptron with a few additional lines of code. A convolutional neural network can also be designed easily and concisely. Limitations: On the flip side In spite of Gluon being easy, compact and efficient, it has certain limitations. Currently it  is available on Apache MXNet, and is awaiting a release in the Microsoft Cognitive Toolkit. However, not much has been known about other frameworks. For instance, it currently lacks support for the two most widely used deep learning frameworks, Caffe2 and TensorFlow. This could pose an issue for Gluon because most interfaces released, provide integration with multiple frameworks. Ultimately, it boils down to the project requirements including the model requirements and the difficulty associated with building networks from a particular tool. So, for a computer vision project people would prefer using Caffe. While TensorFlow is popular  among the developers because of the existing community, the complex nature of the platform, makes a digestible deep learning interface like Gluon highly appreciated.  Hence, each framework performs on its own tradeoffs. Conclusion Gluon comes as a boon, for both experienced data scientists and nascent developers alike. For developers, this interface, models like a data structure, providing more familiarity.  On the other side, for researchers and data scientists, it provides an interface to build prototypes quickly and easily for complex neural networks, without sacrificing training speeds. Overall, Gluon will accelerate the development of advanced neural networks and models, resulting in robust artificial intelligence based applications.

Machine learning has gained a lot of traction over the years because of the predictive solutions that it provides, including the development of intelligent, and reliable models. However, training the models is a laborious task because it takes time to curate the labeled data within the model and then to get the model ready. Reducing the time involved in training and labeling can be overcome by using the novel approach of Transfer Learning - a smarter and effective form of machine learning, where you can use the learnings of one scenario and apply that learning to a different but related problem. How exactly does Transfer Learning work? Transfer learning reduces the efforts to build a model from scratch by using the fundamental logic or base algorithms within one domain and applying it to another. For instance, in the real-world, the balancing logic learned while riding a bicycle can be transferred to learn driving other two-wheeled vehicles. Similarly, in the case of machine learning, transfer learning can be used to transfer the algorithmic logic from one ML model to the other. Let’s look into some of the possible use cases of transfer learning. [dropcap]1[/dropcap] Real-world Simulations Digital simulation is better than creating a physical prototype for real-world implementations. Training a robot in the real-world surroundings is both time and cost consuming. In order to minimize this, robots can now be trained using simulation and the knowledge acquired can be thus transferred onto a real-world robot. This is done using progressive networks, which are ideal for a simulation to the real world transfer of policies in robot control domains. These networks consist of essential features for learning numerous tasks in sequence while enabling transfer and are resistant to catastrophic forgetting--a tendency of Artificial Neural Networks(ANNs) to completely forget previously learned information, on learning a new information.   Another application of simulation can be seen while training self-driving cars, which are trained using simulations through video games. Udacity has open sourced its self-driving car simulator which allows training self-driving cars through GTA 5 and many other video games. However, not all features of a simulation are replicated successfully when they are brought into the real world, as the interactions in the real world are more complex.   [dropcap]2[/dropcap] Gaming The adoption of Artificial Intelligence has taken gaming to an altogether new level. DeepMind’s neural network program AlphaGo is a testament to this, as it successfully defeated a professional Go player. AlphaGo is a master in Go but fails when tasked to play other games. This is because its algorithm is tailored to play Go. So, the disadvantage of using ANNs in gaming is that they cannot master all games as a human brain does. In order to do this, AlphaGo has to totally forget Go and adapt itself to the new algorithms and techniques of the new game. With transfer Learning, the tactics learned in a game can be reapplied to play another game.   An example of how Transfer learning is implemented in gaming can be seen in MadRTS, a commercial Real Time Strategy games. MadRTS, is developed to carry out military simulations. MadRTS uses CARL(CAse-based Reinforcement Learner), a multi-tiered architecture which combines Case-based reasoning(CBR) and Reinforcement Learning(RL). CBR provides an approach to tackle unseen but related problems based on past experiences within each level of the game. RL algorithms, on the other hand, allow the model to carry out good approximations to a situation, based on the agent’s experience in its environment--also known as Markov’s Decision Process. These CBR/RL transfer learning agents are evaluated in order to perform effective learning on tasks given in MadRTS, and should be able to learn better across tasks by transferring experience. [dropcap]3[/dropcap] Image Classification Neural networks are experts in recognizing objects within an image as they are trained on huge datasets of labeled images, which is time-consuming. How transfer learning helps here is, it reduces the time to train the model by pre-training the model using ImageNet, which contains millions of images from different categories. Let’s assume that a convolutional neural network - for instance, a VGG-16 ConvNet - has to be trained to recognize images within a dataset. Firstly, it is pre-trained using ImageNet. Then, it is trained layer-wise starting by replacing the final layer with a softmax layer and training it until the training saturates. Further, the other dense layers are trained progressively. By the end of the training, the ConvNet model is successful in learning to detect images from the dataset provided. In cases where the dataset is not similar to the pre-trained model data, one can finetune weights in the higher layers of the ConvNet by backpropagation methods. The dense layers contain the logic for detecting the image, thus, tuning the higher layers won’t affect the base logic. The convolutional neural networks can be trained on Keras, using Tensorflow or as a backend. An example of Image Classification can be seen in the field of medical imaging, where the convolutional model is trained on ImageNet to solve kidney detection problem in ultrasound images. [dropcap]4[/dropcap] Zero Shot translation Zero shot translation is an extended part of supervised learning, where the goal of the model is, learning to predict novel values from values that are not present in the training dataset. The prominent working example of zero shot translation can be seen in Google’s Neural Translation model(GNMT), which allows for effective cross-lingual translations. Prior to Zero shot implementation, two discrete languages had to be translated using a pivot language. For instance, to translate Korean to Japanese, Korean had to be first translated into English and then English to Japanese. Here, English is the pivot language that acts as a medium to translate Korean to Japanese. This resulted in a translated language that was full of distortions created by the first language pair. Zero shot translation rips off the need for a pivot language. It uses available training data to learn the translational knowledge applied, to translate a new language pair. Another instance of Zero shot translation can be seen in Image2Emoji, which combines visuals and texts to predict unseen emoji icons in a zero shot approach. [dropcap]5[/dropcap] Sentiment Classification Businesses can know their customers better by implementing Sentiment Analysis, which helps them to understand emotions and polarity (negative or positive) underlying the feedback and the product reviews. Analyzing sentiments for a new text corpus is difficult to build up, as training the models to detect different emotions is difficult. A solution to this is Transfer Learning. This involves training the models on any one domain, twitter feeds for instance, and fine-tuning them to another domain you wish to perform Sentiment Analysis on; say movie reviews. Here, deep learning models are trained on twitter feeds by carrying out sentiment analysis of the text corpus and also detecting the polarity of each statement. Once the model is trained on understanding emotions through polarity of the twitter feeds, its underlying language model and learned representation is transferred onto the model assigned a task to analyze sentiments within movie reviews. Here, an RNN model is trained on logistic regression techniques carried out sentiment analysis on the twitter feeds. The word embeddings and the recurrent weights learned from the source domain (twitter feeds) are re-used in the target domain (movie reviews) to classify sentiments within the latter domain. Conclusion Transfer learning has brought in a new wave of learning in machines by reusing algorithms and the applied logic, thus speeding up their learning process. This directly results in a reduction in the capital investment and also the time invested to train a model. This is why many organizations are looking forward to replicating such a learning onto their machine learning models. Also, transfer learning has been carried out successfully in the field of Image processing, Simulations, Gaming, and so on. How transfer learning affects the learning curve of machines in other sectors in the future, is worth watching out for.