How-To Tutorials - News

According to information obtained by Bloomberg, The White House is reportedly making a draft executive order against online platform bias in Social Media firms. Per this draft, federal antitrust and law enforcement agencies are instructed to investigate into the practices of Google, Facebook, and other social media companies. The existence of the draft was first reported by Capital Forum. Federal law enforcers are required to investigate primarily against two violations. First, if an online platform has acted in violation of the antitrust laws. Second, to remove anti-competitive spirit among online platforms and address online platform bias. Per the sources by Capital Forum, the draft is written in two parts. The first part is a policy statement stating that online platforms are central to the flow of information and commerce and need to be held accountable through competition. The second part instructs agencies to investigate bias and anticompetitive conduct in online platforms where they have the authority. In case of lack of authorization, they are required to report concerns or issues to the Federal Trade Commission or the Department of Justice. No online platforms are mentioned by name in the draft. It’s unclear when or if the White House will decide to issue the order. Donald Trump and the White House have always been vocal about the prevalent bias in Social media platforms. In August, Trump tweeted about Social Media discriminating against Republican and Conservative voices. Source: Twitter He also went on to claim that Google search results for “Trump News” reports fake news. He accused the search engines’ algorithms of being rigged. However, that allegation having not been backed by evidence, let Google slam Trump’s accusations, asserting that its search engine algorithms do not favor any political ideology. Earlier this month, Facebook COO Sheryl Sandberg and Twitter CEO Jack Dorsey faced the Senate Select Intelligence Committee, to discuss foreign interference through social media platforms in US elections. Google, Facebook, and Twitter also released a Testimony ahead of appearing before the committee. As reported by the Wall Street Journal, Google CEO Sundar Pichai also plans to meet privately with top Republican lawmakers this Friday to discuss a variety of topics, including the company's alleged political bias in search results. This meeting is organized by the House Majority Leader, Kevin McCarthy. Pichai said on Tuesday that “I look forward to meeting with members on both sides of the aisle, answering a wide range of questions, and explaining our approach." Google is also facing public scrutiny over a report that it intends to launch a censored search engine in China. Google’s custom search engine would link Chinese users’ search queries to their personal phone numbers, thus making it easier for the government to track their searches. About a thousand Google employees frustrated with a series of controversies involving Google have signed a letter to demand transparency on building the alleged search engine. Google’s new Privacy Chief officer proposes a new framework for Security Regulation. Amazon is the next target on EU’s antitrust hitlist. Mark Zuckerberg publishes Facebook manifesto for safeguarding against political interference.  

Researchers at the University of California, Los Angeles (UCLA) have developed a 3D-printed all-optical deep learning architecture called Diffractive Deep Neural Network (D2NN). D2NN is a deep learning neural network physically formed by multiple layers of diffractive surfaces that work in collaboration to optically perform an arbitrary function. While the inference/prediction of the physical network is all-optical, the learning part that leads to its design is done through a computer. How does D2NN work? A computer-simulated design was created first, then the researchers with the help of a 3D printer created very thin polymer wafers. The uneven surface of the wafers helped diffract light coming from the object in different directions. The layers are composed of tens of thousands of artificial neurons or tiny pixels from which the light travels through. These layers together, form an “optical network” that shapes how incoming light travels through them. The network is able to identify an object because the light coming from the object is diffracted mostly toward a single pixel that is assigned to that type of object. The network was then trained using a computer to identify the objects in front of it by learning the pattern of diffracted light each object produced as the light from that object passes through the device. What are its advantages? Scalable: It can easily be scaled up using numerous high-throughput and large-area 3D fabrication methods, such as, soft-lithography, additive manufacturing, and wide-field optical components and detection systems. Easily reconfigurable: D2NN can be easily improved by additional 3D printed layers or replacing some of the existing layers with newly trained ones. Lightening speed: Once the device is trained, it works at the speed of light. Efficient: No energy is consumed to run the device. Cost-effective: The device can be reproduced for less than $50, making it very cost-effective. What are the areas it can be used in? Image analysis Feature detection Object classification Can also enable new microscope or camera designs that can perform unique imaging tasks This new AI device could find applications in the area of medical technologies, data intensive tasks, robotics, security, and or any application where image and video data are essential. Refer to UCLA’s official news article to know more in detail. Also, you can refer to this paper  All-optical machine learning using diffractive deep neural Networks. OpenAI builds reinforcement learning based system giving robots human like dexterity Datasets and deep learning methodologies to extend image-based applications to videos AutoAugment: Google’s research initiative to improve deep learning performance

Humans have been excellent players in most of the gameplays be it indoor or outdoors. However, over the recent years we have been increasingly coming across machines that are playing and winning popular board games Go and Chess against humans using machine learning algorithms. If you think machines are only good at solving the black and whites, you are wrong. The recent achievement of a machine trying to solve a complex game (a Rubik’s cube) is DeepCube. Rubik cube is a challenging piece of puzzle that’s captivated everyone since childhood. Solving it is a brag-worthy accomplishment for most adults. A group of UC Irvine researchers have now developed a new algorithm (used by DeepCube) known as Autodidactic Iteration, which can solve a Rubik’s cube with no human assistance. The Erno Rubik’s cube conundrum Rubik’s cube, a popular three-dimensional puzzle was developed by Erno Rubik in the year 1974. Rubik worked for a month to figure out the first algorithm to solve the cube. Researchers at the UC Irvine state that “Since then, the Rubik’s Cube has gained worldwide popularity and many human-oriented algorithms for solving it have been discovered. These algorithms are simple to memorize and teach humans how to solve the cube in a structured, step-by-step manner.” After the cube became popular among mathematicians and computer scientists, questions around how to solve the cube with least possible turns became mainstream. In 2014, it was proved that the least number of steps to solve the cube puzzle was 26. More recently, computer scientists have tried to find ways for machines to solve the Rubik’s cube. As a first step, they tried and tested ways to use the same successful approach tried in the games Go and Chess. However, this approach did not work well for the Rubik’s cube. The approach: Rubik vs Chess and Go Algorithms used in Go and Chess are fed with rules of the game and then they play against themselves. The deep learning machine here is rewarded based on its performance at every step it takes. Reward process is considered as important as it helps the machine to distinguish between a good and a bad move. Following this, the machine starts playing well i.e it learns how to play well. On the other hand, the rewards in the case of Rubik’s cube are nearly hard to determine. This is because there are random turns in the cube and it is hard to judge whether the new configuration is any closer to a solution. The random turns can be unlimited and hence earning an end-state reward is very rare. Both Chess and Go have a large search space but each move can be evaluated and rewarded accordingly. This isn’t the case for Rubik’s cube! UC Irvine researchers have found a way for machines to create its own set of rewards in the Autodidactic Iteration method for DeepCube. Autodidactic Iteration: Solving the Rubik’s Cube without human Knowledge DeepCube’s Autodidactic Iteration (ADI) is a form of deep learning known as deep reinforcement learning (DRL). It combines classic reinforcement learning, deep learning, and Monte Carlo Tree Search (MCTS). When DeepCube gets an unsolved cube, it decides whether the specific move is an improvement on the existing configuration. To do this, it must be able to evaluate the move. The algorithm, Autodidactic iteration starts with the finished cube and works backwards to find a configuration that is similar to the proposed move. Although this process is imperfect, deep learning helps the system figure out which moves are generally better than others. Researchers trained a network using ADI for 2,000,000 iterations. They further reported, “The network witnessed approximately 8 billion cubes, including repeats, and it trained for a period of 44 hours. Our training machine was a 32-core Intel Xeon E5-2620 server with three NVIDIA Titan XP GPUs.” After training, the network uses a standard search tree to hunt for suggested moves for each configuration. The researchers in their paper said, “Our algorithm is able to solve 100% of randomly scrambled cubes while achieving a median solve length of 30 moves — less than or equal to solvers that employ human domain knowledge.” Researchers also wrote, “DeepCube is able to teach itself how to reason in order to solve a complex environment with only one reward state using pure reinforcement learning.” Furthermore, this approach will have a potential to provide approximate solutions to a broad class of combinatorial optimization problems. To explore Deep Reinforcement Learning check out our latest releases, Hands-On Reinforcement Learning with Python and Deep Reinforcement Learning Hands-On. How greedy algorithms work Creating a reference generator for a job portal using Breadth First Search (BFS) algorithm Anatomy of an automated machine learning algorithm (AutoML)    

Genetics company 23andMe, which uses machine learning algorithms for human genome analysis, has entered into a four year collaboration with pharmaceutical giant GlaxoSmithKline. They will now share their 5 million client genetic data with GSK to advance research into treatments of diseases. This collaboration will be used to identify novel drug targets, tackle new subsets of disease and enable rapid progression of clinical programs. The 12 years old firm has already published more than 100 scientific papers based on its customers' data. All activities within the collaboration will initially be co-funded, with either company having certain rights to reduce its funding share. "The goal of the collaboration is to gather insights and discover novel drug targets driving disease progression and develop therapies," GlaxoSmithKline said in a press release. GSK is also reported to have invested $300 million in 23andMe. During the four year collaboration GSK will use 23andMe’s database and statistical analytics for drug target discovery. This collaboration will be used to design GSK’s LRRK2 inhibitor, which is in development for the potential treatment for Parkinson’s disease. 23andMe’s database of consented customers who have a LRRK2 variant status will be used to accelerate the progress of this programme. Together, GSK and 23andMe will target and recruit patients with defined LRRK2 mutations in order to reach clinical proof of concept. 23andMe have made it quite clear that participating in this program is voluntary and requires clients to affirmatively consent to participate. However not everyone is clear of how this would work. First, the company has specified that any research involving customer data that has already been performed or published prior to receipt of withdrawal request will not be reversed. This may have a negative effect as people are generally not aware of all the privacy policies and generally don’t read the Terms of Service. Moreover, as Peter Pitts, president of the Center for Medicine in the Public Interest, notes, “If a person's DNA is used in research, that person should be compensated. Customers shouldn’t be paying for the privilege of 23andMe working with a for-profit company in a for-profit research project.” Both the companies have sworn to provide maximum data protection for their employees. In a blog post, they note, “The continued protection of customers’ data and privacy is the highest priority for both GSK and 23andMe. Both companies have stringent security protections in place when it comes to collecting, storing and transferring information about research participants.” You can read more about the news, on a blog by 23andMe founder, Anne Wojcicki. 6 use cases of Machine Learning in Healthcare Healthcare Analytics: Logistic Regression to Reduce Patient Readmissions NIPS 2017 Special: How machine learning for genomics is bridging the gap between research and clinical trial success by Brendan Frey

Justin Sun, founder of the decentralized Internet platform, Tron announced the acquisition of BitTorrent, which is a popular file-sharing network. As reported by Techcrunch, the blockchain-based platform is said to have acquired BitTorrent for a sum of about $126 million. TRON foundation is a decentralized platform for sharing entertainment content, including music and games. It uses blockchain and peer-to-peer (p2p) network technology to exclude the need for a middleman between content producers and consumers such as Google and Amazon. BitTorrent, founded in the year 2004, is a popular peer-to-peer file sharing protocol with 100 million users. It also owns the popular, uTorrent client software and torrent client. BitTorrent is known to stream movies and music with great ease and is also fast and reliable. Moreover, it has changed how and why we watch things online. With the BitTorrent acquisition, Justin wants to make Tron the largest decentralized ecosystem in the world. While that’s an exciting prospect for both tech users, users had questions if Tron would charge them via cryptocurrency for the services offered. BitTorrent, in their blog, stated that “it has no plans to change what we do or charge for the services we provide. We have no plans to enable mining of cryptocurrency now or in the future." However, Tron’s plans for BitTorrent are still under the hood. ‘TRON + BitTorrent: The world’s largest decentralized ecosystem’ In an official letter by the Tron foundation, it stated that the firm would continue BitTorrent’s protocol legacy post integrating it within the Tron ecosystem. https://twitter.com/BitTorrent/status/1021629735258841088 The letter also states that, “With the integration of BitTorrent, TRON aims to liberate the Internet from the stranglehold of large corporations, give data rights back to the individual, and reignite the early 21st century vision of a free, transparent, decentralized network to connect the world, because the internet belongs to the people.” Sun in his letter also mentioned BitTorrent as the genesis of the decentralization movement. Tron’s developers, entrepreneurs, and the community consider BitTorrent as the original pioneers of decentralization technology. Sun stated, "We believe that joining the TRON network will further enhance BitTorrent and accelerate our mission of creating an Internet of options, not rules." Due to this acquisition, BitTorrent may lose its primary illegal user base. However, it still continues to demonstrate its legal uses and will further continue to evolve with TRON’s ecosystem. It will also take control of its two popular Torrent applications, BitTorrent and μTorrent clients, which will be free to download, and supported by ads. This merger is a happy turning point for BitTorrent. BitTorrent was in a total mess some years back and had not raised any money since 2008 following which they fired its dual CEOs. Given its commitment to the notion of a decentralized internet, BitTorrent still attempted to function as a business, with its app or service. But these strategies did not work out well. However, TRON’s acquisition has turned the tables for BitTorrent recently. It could be the story of Cinderella meeting Prince Charming of this decade. Read more about BitTorrent’s acquisition on Techcrunch. Top 15 Cryptocurrency Trading Bots Crypto-ML, a machine learning powered cryptocurrency platform Can Cryptocurrency establish a new economic world order?

According to research conducted by T. W. Hughes, M. Minkov, Y. Shi, and S. Fan, artificial neural networks can be directly trained on an optical chip. The research, titled “Training of photonic neural networks through in situ backpropagation and gradient measurement” demonstrates that an optical circuit has all the capabilities to perform the critical functions of an electronics-based artificial neural network. This makes performing complex tasks like speech or image recognition less expensive, faster and more energy efficient. According to research team leader, Shanhui Fan of Stanford University "Using an optical chip to perform neural network computations more efficiently than is possible with digital computers could allow more complex problems to be solved”. During the research, the training step on optical ANNs was performed using a traditional digital computer. The final settings were then imported into the optical circuit. But, according to Optica (the Optical Society journal for high impact research at Stanford),. there is a more direct method for training these networks. This involves making use of an optical analog within the ‘backpropagation' algorithm. Tyler W. Hughes, the first author of the research paper, states that "using a physical device rather than a computer model for training makes the process more accurate”.  He also mentions that “because the training step is a very computationally expensive part of the implementation of the neural network, performing this step optically is key to improving the computational efficiency, speed and power consumption of artificial networks." Neural network processing is usually performed with the help of a traditional computer. But now, for neural network computing, researchers are interested in Optics-based devices as computations performed on these devices use much less energy compared to electronic devices. In New York researchers designed an optical chip that imitates the way, conventional computers train neural networks. This then provides a way of implementing an all-optical neural network. According to Hughes, the ANN is like a black box with a number of knobs. During the training stage, each knob is turned ever so slightly so the system can be tested to see how the algorithm’s performance changes. He says, “Our method not only helps predict which direction to turn the knobs but also how much you should turn each knob to get you closer to the desired performance”. How does the new training protocol work? This new training method uses optical circuits which have tunable beam splitters. You can adjust these spitters by altering the settings of optical phase shifters. First, you feed a laser which is encoded with information that needs to be processed through the optical circuit. Once the laser exits the device, the difference against the expected outcome is calculated. This information that is collected then generates a new light signal through the optical network in the opposite direction. Researchers also showed that neural network performance changes with respect to each beam splitter's setting. You can also change the phase shifter settings based on this information. The whole process is repeated until the desired outcome is produced by the neural network. This training technique has been further tested by researchers using optical simulations. In these tests, the optical implementation performed similarly to a conventional computer. The researchers are planning to further optimize the system in order to come out with a practical application using a neural network. How Deep Neural Networks can improve Speech Recognition and generation Recurrent neural networks and the LSTM architecture  

Google AI Quantum team announced two releases at the First International Workshop on Quantum Software and Quantum Machine Learning(QSML) yesterday. Firstly the public alpha release of Cirq, an open source framework for NISQ computers. The second release is OpenFermion-Cirq, an example of a Cirq-based application enabling near-term algorithms. Noisy Intermediate Scale Quantum (NISQ) computers are devices including ~50 - 100 qubits and high fidelity quantum gates enhance the quantum algorithms such that they can understand the power that these machines uphold. However, quantum algorithms for the quantum computers have their limitations such as A poor mapping between the algorithms and the machines Also, some quantum processors have complex geometric constraints These and other nuances inevitably lead to wasted resources and faulty computations. Cirq comes as a great help for researchers here. It is focussed on near-term questions, which help researchers to understand whether NISQ quantum computers are capable of solving computational problems of practical importance. It is licensed under Apache 2 and is free to be either embedded or modified within any commercial or open source package. Cirq highlights With Cirq, researchers can write quantum algorithms for specific quantum processors. It provides a fine-tuned user control over quantum circuits by, specifying gate behavior using native gates, placing these gates appropriately on the device, and scheduling the timing of these gates within the constraints of the quantum hardware. Other features of Cirq include: Allows users to leverage the most out of NISQ architectures with optimized data structures to write and compile the quantum circuits. Supports running of the algorithms locally on a simulator Designed to easily integrate with future quantum hardware or larger simulators via the cloud. OpenFermion-Cirq highlights Google AI Quantum team also released OpenFermion-Cirq, which is an example of a CIrq-based application that enables the near-term algorithms.  OpenFermion is a platform for developing quantum algorithms for chemistry problems. OpenFermion-Cirq extends the functionality of OpenFermion by providing routines and tools for using Cirq for compiling and composing circuits for quantum simulation algorithms. An instance of the OpenFermion-Cirq is, it can be used to easily build quantum variational algorithms for simulating properties of molecules and complex materials. While building Cirq, the Google AI Quantum team worked with early testers to gain feedback and insight into algorithm design for NISQ computers. Following are some instances of Cirq work resulting from the early adopters: Zapata Computing: simulation of a quantum autoencoder (example code, video tutorial) QC Ware: QAOA implementation and integration into QC Ware’s AQUA platform (example code, video tutorial) Quantum Benchmark: integration of True-Q software tools for assessing and extending hardware capabilities (video tutorial) Heisenberg Quantum Simulations: simulating the Anderson Model Cambridge Quantum Computing: integration of proprietary quantum compiler t|ket> (video tutorial) NASA: architecture-aware compiler based on temporal-planning for QAOA (slides) and simulator of quantum computers (slides) The team also announced that it is using Cirq to create circuits that run on Google’s Bristlecone processor. Their future plans include making the Bristlecone processor available in cloud with Cirq as the interface for users to write programs for this processor. To know more about both the releases, check out the GitHub repositories of each Cirq and OpenFermion-Cirq. Q# 101: Getting to know the basics of Microsoft’s new quantum computing language Google Bristlecone: A New Quantum processor by Google’s Quantum AI lab Quantum A.I. : An intelligent mix of Quantum+A.I.

Microsoft's Brad Smith took the unprecedented move last week of calling for government to regulate facial recognition technology. In an industry that has resisted government intervention, it was a bold yet humble step. It was a way of saying "we can't deal with this on our own." There will certainly be people who disagree with Brad Smith. For some the entrepreneurial spirit that is central to tech and startup culture will only be stifled by regulation. But let's be realistic about where we are at the moment - the technology industry has never faced such a crisis of confidence and met with substantial public cynicism. Perhaps government regulation is precisely what we need to move forward. Here are 4 reasons why government should regulate technology.  Regulation can restore accountability and rebuild trust in tech We've said it a lot in 2018, but there really is a significant trust deficit in technology at the moment. From Cambridge Analytica scandal to AI bias, software has been making headlines in a way it never has before. This only cultivates a culture of cynicism across the public. And with talk of automation and job losses, it paints a dark picture of the future. It's no wonder that TV series like Black Mirror have such a hold over the public imagination. Of course, when used properly, technology should simply help solve problems - whether that's better consumer tech or improved diagnoses in healthcare. The problem arises when we find that there our problem-solving innovations have unintended consequences. By regulating, government can begin to think through some of these unintended consequences. But more importantly, trust can only be rebuilt once there is some degree of accountability within the industry. Think back to Zuckerberg's Congressional hearing earlier this year - while the Facebook chief may have been sweating, the real takeaway was that his power and influence was ultimately untouchable. Whatever mistakes he's made were just part and parcel of moving fast and breaking things. An apology and a humble shrug might normally pass, but with regulation, things begin to get serious. Misusing user data? We've got a law for that. Potentially earning money from people who want to undermine western democracy? We've got a law for that. Read next: Is Facebook planning to spy on you through your mobile’s microphones? Government regulation will make the conversation around the uses and abuses of technology more public Too much conversation about how and why we build technology is happening in the wrong places. Well, not the wrong places, just not enough places. The biggest decisions about technology are largely made by some of the biggest companies on the planet. All the dreams about a new democratized and open world are all but gone, as the innovations around which we build our lives come from a handful of organizations that have both financial and cultural clout. As Brad Smith argues, tech companies like Microsoft, Google, and Amazon are not the place to be having conversations about the ethical implications of certain technologies. He argues that while it's important for private companies to take more responsibility, it's an "inadequate substitute for decision making by the public and its representatives in a democratic republic." He notes that the commercial dynamics are always going to twist conversations. Companies, after all, are answerable to shareholders - only governments are accountable to the public. By regulating, the decisions we make (or don't make) about technology immediately enter into public discourse about the kind of societies we want to live in. Citizens can be better protected by tech regulation... At present, technology often advances in spite of, not because of, people. For all the talk of human-centered design, putting the customer first, every company that builds software is interested in one thing: making money. AI in particular can be dangerous for citizens For example, according to a ProPublica investigation, AI has been used to predict future crimes in the justice system. That's frightening in itself, of course, but it's particularly terrifying when you consider that criminality was falsely predicted at twice the times for black people as white people. Even in the context of social media filters, in which machine learning serves content based on a user's behavior and profile presents dangers to citizens. It gives rise to fake news and dubious political campaigning, making citizens more vulnerable to extreme - and false - ideas. By properly regulating this technology we should immediately have more transparency over how these systems work. This transparency would not only lead to more accountability in how they are built, it also ensures that changes can be made when necessary. Read next: A quick look at E.U.’s pending antitrust case against Google’s Android ...Software engineers need protection too One group haven't really been talked about when it comes to government regulation - the people actually building the software. This a big problem. If we're talking about the ethics of AI, software engineers building software are left in a vulnerable position. This is because the lines of accountability are blurred. Without a government framework that supports ethical software decision making, engineers are left in limbo. With more support for software engineers from government, they can be more confident in challenging decisions from their employers. We need to have a debate about who's responsible for the ethics of code that's written into applications today - is it the engineer? The product manager? Or the organization itself? That isn't going to be easy to answer, but some government regulation or guidance would be a good place to begin. Regulation can bridge the gap between entrepreneurs, engineers and lawmakers Times change. Years ago, technology was deployed by lawmakers as a means of control, production or exploration. That's why the military was involved with many of the innovations of the mid-twentieth century. Today, the gap couldn't be bigger. Lawmakers barely understand encryption, let alone how algorithms work. But there is also naivety in the business world too. With a little more political nous and even critical thinking, perhaps Mark Zuckerberg could have predicted the Cambridge Analytica scandal. Maybe Elon Musk would be a little more humble in the face of a coordinated rescue mission. There's clearly a problem - on the one hand, some people don't know what's already possible. For others, it's impossible to consider that something that is possible could have unintended consequences. By regulating technology, everyone will have to get to know one another. Government will need to delve deeper into the field, and entrepreneurs and engineers will need to learn more about how regulation may affect them. To some extent, this will have to be the first thing we do - develop a shared language. It might also be the hardest thing to do, too.

This is a great news for all Silicon Valley Fans. The amazing Not Hotdog A.I. app shown on season 4’s 4th episode, has been nominated for a Primetime Emmy Award. The Emmys has placed Silicon Valley and the app in the category “Outstanding Creative Achievement In Interactive Media Within a Scripted Program” among other popular shows. Other nominations include 13 Reasons Why for “Talk To The Reasons”, a website that lets you chat with the characters. Rick and Morty, for “Virtual Rick-ality”, a virtual reality game. Mr. Robot, for "Ecoin", a fictional Global Digital Currency. And Westworld for “Chaos Takes Control Interactive Experience”, an online experience for promoting the show’s second season. Within a day of its launch, the ‘Not Hotdog’ application was trending on the App Store and on Twitter, grabbing the #1 spot on both Hacker News & Product Hunt, and won a Webby for Best Use of Machine Learning. The app uses state-of-the-art deep learning, with a mix of React Native, Tensorflow & Keras. It has averaged 99.9% crash-free users with a 4.5+/5 rating on the app stores. The ‘Not Hotdog’ app does what the name suggests. It identifies hotdogs — and not hot dogs. It is available for both Android and iOS devices whose description reads “What would you say if I told you there is an app on the market that tell you if you have a hotdog or not a hotdog. It is very good and I do not want to work on it any more. You can hire someone else.” How the Not Hotdog app is built The creator Tim Anglade uses sophisticated neural architecture for the Silicon Valley A.I. app that runs directly on your phone and trained it with Tensorflow, Keras & Nvidia GPUs. Of course, the use case is not very useful, but the overall app is a substantial example of deep learning and edge computing in pop culture.  The app provides better privacy as images never leave a user’s device. Consequently, users are provided with a faster experience and offline availability as processing doesn’t go to the cloud. Using a no cloud-based AI approach means that the company can run the app at zero cost, providing significant savings, even under a load of millions of users. What is amazing about the app is that it was built by a single creator with limited resources ( a single laptop and GPU, using hand-curated data). This talks lengths of how much can be achieved even with a limited amount of time and resources, by non-technical companies, individual developers, and hobbyists alike. The initial prototype of the app was built using Google Cloud Platform’s Vision API, and React Native. React Native is a good choice as it supports many devices. The Google Cloud’s Vision API, however, was quickly abandoned. Instead, what was brought into the picture was Edge Computing.  It enabled training the neural network directly on the laptop, to be exported and embedded directly into the mobile app, making the neural network execution phase run directly inside the user’s phone. How TensorFlow powers the Not Hotdog app After React Native, the second part of their tech stack was TensorFlow. They used the TensorFlow’s Transfer Learning script, to retrain the Inception architecture which helps in dealing with a more specific image problem. Transfer Learning helped them get better results much faster, and with less data compared to training from scratch. Inception turned out too big to be retrained. So, at the suggestion of Jeremy P. Howard, they explored and settled down on SqueezeNet.  It provided explicit positioning as a solution for embedded deep learning, and the availability of a pre-trained Keras model on GitHub. The final architecture was largely based on Google’s MobileNets paper, which provided their neural architecture with Inception-like accuracy on simple problems, with only almost 4M parameters. YouTube has a $25 million plan to counter fake news and misinformation Microsoft’s Brad Smith calls for facial recognition technology to be regulated Too weird for Wall Street: Broadcom’s value drops after purchasing CA Technologies

After the back-to-back release of Tensorflow 1.9 release candidates, rc-0, rc-1, and rc-2, the final version TensorFlow 1.9 is out and generally available. Key highlights of this version include support for gradient boosted trees estimators, new keras layers to speed up GRU and LSTM implementations and tfe.Network deprecation. It also includes improved functions for supporting data loading, text processing and pre-made estimators. Tensorflow 1.9 major features and improvements As mentioned in Tensorflow 1.9 rc-2, new Keras-based get started page and programmers guide page in the tf.Keras have been updated. The tf.Keras has been updated to Keras 2.1.6 API. One should try the newly added  tf.keras.layers.CuDNNGRU, used for a faster GRU implementation and tf.keras.layers.CuDNNLSTM layers, which allows faster LSTM implementation. Both these layers are backed by cuDNN( NVIDIA CUDA Deep Neural Network library (cuDNN)). Gradient boosted trees estimators, a non-parametric statistical learning technique for  classification and regression, are now supported by core feature columns and losses. Also, the python interface for the TFLite Optimizing Converter has been expanded, and the command line interface (AKA: toco, tflite_convert) is once again included in the standard pip installation. The distributions.Bijector API in the TF version 1.9 also supports broadcasting for Bijectors with the new API changes. Tensorflow 1.9 also includes improved data-loading and text processing with tf.decode_compressed, tf.string_strip, and Tf.strings.regex_full_match. It also has an added experimental support for new pre-made estimators like tf.contrib.estimator.BaselineEstimator, tf.contrib.estimator.RNNClassifier, tf.contrib.estimator.RNNEstimator. This version includes two breaking changes. Firstly for opening up empty variable scopes one can replace variable_scope('', ...) by variable_scope(tf.get_variable_scope(), ...), which is used to get the current scope of the variable. And the second breakthrough change is, headers used for building custom ops have been moved to a different file path. From site-packages/external to site-packages/tensorflow/include/external. Some bug fixes and other changes include: The tfe.Network has been deprecated Layered variable names have changed in the following conditions: Using tf.keras.layers with custom variable scopes. Using tf.layers in a subclassed tf.keras.Model class. Added the ability to pause recording operations for gradient computation via tf.GradientTape.stop_recording in the Eager execution and updated its documentation and introductory notebooks. Fixed an issue in which the TensorBoard Debugger Plugin, which could not handle total source file size exceeding gRPC message size limit (4 MB). Added GCS Configuration Ops and complex128 support to FFT, FFT2D, FFT3D, IFFT, IFFT2D, and IFFT3D. Conv3D, Conv3DBackpropInput, Conv3DBackpropFilter now supports arbitrary. Prevents tf.gradients() from backpropagating through integer tensors. LinearOperator[1D,2D,3D]Circulant added to tensorflow.linalg. To know more about the other changes, visit TensorFlow 1.9 release notes on GitHub. Create a TensorFlow LSTM that writes stories [Tutorial] Build and train an RNN chatbot using TensorFlow [Tutorial] Use TensorFlow and NLP to detect duplicate Quora questions [Tutorial]

It won’t be wrong to say that the first day of the ongoing Google I/O 2018 conference was largely dominated by Artificial Intelligence. CEO Sundar Pichai didn’t hide his excitement in giving us a sneak-peek of a whole host of features across various products driven by AI, which Google plan to roll out to the masses in the coming days. One of the biggest announcements was the unveiling of the next-gen silicon chip - the Tensor Processing Unit 3.0. The TPU has been central to Google’s AI market dominance strategy since its inception in 2016, and the latest iteration of this custom-made silicon chip promises to deliver faster and more powerful machine learning capabilities. What’s new in TPU 3.0? TPU is Google’s premium AI hardware offering for its cloud platform, with the objective of making it easier to run machine learning systems in a fast, cheap and easy manner. In his I/O 2018 keynote, Sundar Pichai declared that TPU 3.0 will be 8x more powerful than its predecessor. [embed]https://www.youtube.com/watch?v=ogfYd705cRs&t=5750[/embed] A TPU 3.0 pod is expected to crunch numbers at approximately 100 petaflops, as compared to 11.5 petaflops delivered by TPU 2.0. Pichai did not comment about the precision of the processing in these benchmarks - something which can make a lot of difference in real-world applications. Not a lot of other TPU 3.0 features were disclosed. The chip is still only for Google’s use for powering their applications. It also powers the Google Cloud Platform to handle customers’ workloads. High performance - but at a cost? An important takeaway from Pichai’s announcement is that TPU 3.0 is expected to be power-hungry - so much so that Google’s data centers deploying the chips now require liquid cooling to take care of the heat dissipation problem. This is not necessarily a good thing, as the need for dedicated cooling systems will only increase as Google scale up their infrastructure. A few analysts and experts, including Patrick Moorhead, tech analyst and founder of Moor Insights & Strategy, have raised concerns about this on twitter. [embed]https://twitter.com/PatrickMoorhead/status/993905641390391296[/embed] [embed]https://twitter.com/ryanshrout/status/993906310197506048[/embed] [embed]https://twitter.com/LostInBrittany/status/993904650888724480[/embed] The TPU is keeping up with Google’s growing AI needs The evolution of the Tensor Processing Unit has been rather interesting. When TPU was initially released way back in 2016, it was just a simple math accelerator used for training models, supporting barely close to 8 software instructions. However, Google needed more computing power to keep up with their neural networks to power their applications on the cloud. TPU 2.0 supported single-precision floating calculations and added 8GB of HBM (High Bandwidth Memory) for faster, more improved performance. With 3.0, TPU have stepped up another notch, delivering the power and performance needed to process data and run their AI models effectively. The significant increase in the processing capability from 11 petaflops to more than 100 petaflops is a clear step in this direction. Optimized for Tensorflow - the most popular machine learning framework out there - it is clear that TPU 3.0 will have an important role to play as Google look to infuse AI into all their major offerings. A proof of this is some of the smart features that were announced in the conference - the smart compose option in Gmail, improved Google Assistant, Gboard, Google Duplex, and more. TPU 3.0 was needed, with the competition getting serious It comes as no surprise to anyone that almost all the major tech giants are investing in cloud-ready AI technology. These companies are specifically investing in hardware to make machine learning faster and more efficient, to make sense of the data at scale, and give intelligent predictions which are used to improve their operations. There are quite a few examples to demonstrate this. Facebook’s infrastructure is being optimized for the Caffe2 and Pytorch frameworks, designed to process the massive information it handles on a day to day basis. Intel have come up with their neural network processors in a bid to redefine AI. It is also common knowledge that even the cloud giants like Amazon want to build an efficient cloud infrastructure powered by Artificial Intelligence. Just a few days back, Microsoft previewed their Project Brainwave in the Build 2018 conference, claiming super-fast Artificial Intelligence capabilities which rivaled Google’s very own TPU. We can safely infer that Google needed a TPU 3.0 like hardware to join the elite list of prime enablers of Artificial Intelligence in the cloud, empowering efficient data management and processing. Check out our coverage of Google I/O 2018, for some exciting announcements on other Google products in store for the developers and Android fans. Read also AI chip wars: Is Brainwave Microsoft’s Answer to Google’s TPU? How machine learning as a service is transforming cloud The Deep Learning Framework Showdown: TensorFlow vs CNTK

Mark Zuckerberg yesterday (April 10 2018) testified in front of congress. That's a pretty big deal. Congress has been waiting some time for the chance to grill the Facebook chief, with "Zuck" resisting. So the fact that he finally had his day in D.C. indicates the level of pressure currently on him. Some have lamented the fact that senators were given so little time to respond to Zuckerberg - there was no time to really get deep into the issues at hand. However, although it's true that there was a lot that was superficial about the event, if you looked closely, there was plenty to take away from it. Here are the 5 of the most important things we learned from Mark Zuckerberg's testimony in front of Congress. Policy makers don't really understand that much about tech The most shocking thing to come out of Zuckerberg's testimony was unsurprising; the fact that some of the most powerful people in the U.S. don't really understand the technology that's being discussed. More importantly this is technology they're going to have to be making decisions on. One Senator brought printouts of Facebook pages and asked Zuckerberg if these were examples of Russian propaganda groups. Another was confused about Facebook's business model - how could it run a free service and still make money? Those are just two pretty funny examples, but the senators' lack of understanding could be forgiven due to their age. However, there surely isn't any excuse for 45 year old Senator Brian Schatz to misunderstand the relationship between Whatsapp and Facebook. https://twitter.com/pdmcleod/status/983809717116993537 Chris Cillizza argued on CNN that "the senate's tech illiteracy saved Zuckerberg". He explained: The problem was that once Zuckerberg responded - and he largely stuck to a very strict script in doing so - the lack of tech knowledge among those asking him questions was exposed. The result? Zuckerberg was rarely pressed, rarely forced off his talking points, almost never made to answer for the very real questions his platform faces. This lack of knowledge led to proceedings being less than satisfactory for onlookers. Until this knowledge gap is tackled, it's always going to be a challenge for political institutions to keep up with technological innovators. Ultimately, that's what makes regulation hard. Zuckerberg is still held up as the gatekeeper of tech in 2018 Zuckerberg is still held up as a gatekeeper or oracle of modern technology. That is probably a consequence of the point above. Because there's such a knowledge gap within the institutions that govern and regulate, it's more manageable for them to look to a figurehead. That, of course, goes both ways - on the one hand Zuckerberg is a fountain of knowledge, someone who can solve these problems. On the other hand is part of a Silicon Valley axis of evil, nefariously plotting the downfall of democracy and how to read your WhatsApp messages. Most people know that neither is true. The key point, though, is that however you feel about Zuckerberg, he is not the man you need to ask about regulation. This is something that Zephy Teachout argues on the Guardian. "We shouldn’t be begging for Facebook’s endorsement of laws, or for Mark Zuckerberg’s promises of self-regulation" she writes. In fact, one of the interesting subplots of the hearing was the fact that Zuckerberg didn't actually know that much. For example, a lot has been made of how extensive his notes were. And yes, you certainly would expect someone facing a panel of Senators in Washington to be well-briefed. But it nevertheless underlines an important point - the fact that Facebook is a complex and multi-faceted organization that far exceeds the knowledge of its founder and CEO. In turn, this tells you something about technology that's often lost within the discourse: the fact that its hard to consider what's happening at a superficial or abstract level without completely missing the point. There's a lot you could say about Zuckerberg's notes. One of the most interesting was the point around GDPR. The note is very prescriptive: it says "Don't say we already do what GDPR requires." Many have noted that this throws up a lot of issues, not least how Facebook plan to tackle GDPR in just over a month if they haven't moved on it already. But it's the suggestion that Zuckerberg was completely unaware of the situation that is most remarkable here. He doesn't even know where his company is on one of the most important pieces of data legislation for decades. Facebook is incredibly naive If senators were often naive - or plain ignorant - on matters of technology - during the hearing, there was plenty of evidence to indicate that Zuckerberg is just as naive. The GDPR issue mentioned above is just one example. But there are other problems too. You can't, for example, get much more naive than thinking that Cambridge Analytica had deleted the data that Facebook had passed to it. Zuckerberg's initial explanation was that he didn't realize that Cambridge Analytica was "not an app developer or advertiser", but he corrected this saying that his team told him they were an advertiser back in 2015, which meant they did have reason to act on it but chose not to. Zuckerberg apologized for this mistake, but it's really difficult to see how this would happen. There almost appears to be a culture of naivety within Facebook, whereby the organization generally, and Zuckerberg specifically, don't fully understand the nature of the platform it has built and what it could be used for. It's only now, with Zuckerberg talking about an "arms race" with Russia that this naivety is disappearing. But its clear there was an organizational blindspot that has got us to where we are today. Facebook still thinks AI can solve all of its problems The fact that Facebook believes AI is the solution to so many of its problems is indicative of this ingrained naivety. When talking to Congress about the 'arms race' with Russian intelligence, and the wider problem of hate speech, Zuckerberg signaled that the solution lies in the continued development of better AI systems. However, he conceded that building systems actually capable of detecting such speech could be 5 to 10 years away. This is a problem. It's proving a real challenge for Facebook to keep up with the 'misuse' of its platform. Foreign Policy reports that: "...just last week, the company took down another 70 Facebook accounts, 138 Facebook pages, and 65 Instagram accounts controlled by Russia’s Internet Research Agency, a baker’s dozen of whose executives and operatives have been indicted by Special Counsel Robert Mueller for their role in Russia’s campaign to propel Trump into the White House." However, the more AI comes to be deployed on Facebook, the more that the company is going to have to rethink how it describes itself. By using algorithms to regulate the way the platform is used, there comes to be an implicit editorializing of content. That's not necessarily a bad thing, but it does mean we again return to this final problem... There's still confusion about the difference between a platform and a publisher Central to every issue that was raised in Zuckerberg's testimony was the fact that Facebook remains confused about whether it is a platform or a publisher. Or, more specifically, the extent to which it is responsible for the content on the platform. It's hard to single out Zuckerberg here because everyone seems to be confused on this point. But it's interesting that he seems to have never really thought about the problem. That does seem to be changing, however. In his testimony, Zuckerberg said that "Facebook was responsible" for the content on its platforms. This statement marks a big change from the typical line used by every social media platform - that platforms are just platforms, they bear no responsibility for what is published on them. However, just when you think Zuckerberg is making a definitive statement, he steps back. He went on to say that "I agree that we are responsible for the content, but we don't produce the content." This statement hints that he still wants to keep the distinction between platform and publisher. Unfortunately for Zuckerberg, that might be too late. Read Next OpenAI charter puts safety, standards, and transparency first ‘If tech is building the future, let’s make that future inclusive and representative of all of society’ – An interview with Charlotte Jee What your organization needs to know about GDPR 20 lessons on bias in machine learning systems by Kate Crawford at NIPS 2017

Continuing from where we left our previous post, we are back with a quick roundup of top research papers on Machine Translation, Predictive Modelling, Image-to-Image Translation, and Recommendation Systems from NIPS 2017. Machine Translation In layman terms, Machine translation (MT) is the process by which a computer software translates a text from one natural language to another. This year at NIPS, a large number of presentations focused on innovative ways of improving translations. Here are our top picks. Value Networks: Improving beam search for better Translation Microsoft has ventured into translation tasks with the introduction of Value Networks in their paper “Decoding with Value Networks for Neural Machine Translation”. Their prediction network improves beam search which is a shortcoming of Neural Machine Translation (NMT). This new methodology inspired by the success of AlphaGo, takes the source sentence x, the currently available decoding output y1, ··· , yt1 and a candidate word w at step t as inputs, using which it predicts the long-term value (e.g., BLEU score) of the partial target sentence if it is completed by the NMT(Neural Machine Translational) model. Experiments show that this approach significantly improves the translation accuracy of several translation tasks. CoVe: Contextualizing Word Vectors for Machine Translation Salesforce researchers have used a new approach to contextualize word vectors in their paper “Learned in Translation: Contextualized Word Vectors”. A wide variety of common NLP tasks namely sentiment analysis, question classification, entailment, and question answering use only supervised word and character vectors to contextualize Word vectors. The paper uses a deep LSTM encoder from an attentional sequence-to-sequence model trained for machine translation. Their research portrays that adding these context vectors (CoVe) improves performance over using only unsupervised word and character vectors. For fine-grained sentiment analysis and entailment also, CoVe improves the performance of the baseline models to the state-of-the-art. Predictive Modelling A lot of research showcased at NIPS was focussed around improving the predictive capabilities of Neural Networks. Here is a quick look at the top presentations. Deep Ensembles for Predictive Uncertainty Estimation Bayesian Solutions are most frequently used in quantifying predictive uncertainty in Neural networks. However, these solutions can at times be computationally intensive. They also require significant modifications to the training pipeline. DeepMind researchers have proposed an alternative to Bayesian NNs in their paper “Simple and scalable predictive uncertainty estimation using deep ensembles”. Their proposed method is easy to implement, readily parallelizable requires very little hyperparameter tuning, and yields high-quality predictive uncertainty estimates. VAIN: Scaling Multi-agent Predictive Modelling Multi-agent predictive modeling predicts the behavior of large physical or social systems by an interaction between various agents. However, most approaches come at a prohibitive cost. For instance, Interaction Networks (INs) were not able to scale with the number of interactions in the system (typically quadratic or higher order in the number of agents). Facebook researchers have introduced VAIN, which is a simple attentional mechanism for multi-agent predictive modeling that scales linearly with the number of agents. They can achieve similar accuracy but at a much lower cost. You can read more about the mechanism in their paper “VAIN: Attentional Multi-agent Predictive Modeling” PredRNN: RNNs for Predictive Learning with ST-LSTM Another paper titled “PredRNN: Recurrent Neural Networks for Predictive Learning using Spatiotemporal LSTMs” showcased a new predictive recurrent neural network.  This architecture is based on the idea that spatiotemporal predictive learning should memorize both spatial appearances and temporal variations in a unified memory pool. The core of this RNN is a new Spatiotemporal LSTM (ST-LSTM) unit that extracts and memorizes spatial and temporal representations simultaneously. Memory states are allowed to zigzag in two directions: across stacked RNN layers vertically and through all RNN states horizontally. PredRNN is a more general framework, that can be easily extended to other predictive learning tasks by integrating with other architectures. It achieved state-of-the-art prediction performance on three video prediction datasets. Recommendation Systems New researches were presented by Google and Microsoft to address the cold-start problem and to build robust and powerful of Recommendation systems. Off-Policy Evaluation For Slate Recommendation Microsoft researchers have studied and evaluated policies that recommend an ordered set of items in their paper “Off-Policy Evaluation For Slate Recommendation”. General recommendation approaches require large amounts of logged data to evaluate whole-page metrics that depend on multiple recommended items, which happens when showing ranked lists. The number of these possible lists is called as slates. Microsoft researchers have developed a technique for evaluating page-level metrics of such policies offline using logged past data, reducing the need for online A/B tests. Their method models the observed quality of the recommended set as an additive decomposition across items. It fits many realistic measures of quality and shows exponential savings in the amount of required data compared with other off-policy evaluation approaches. Meta-Learning on Cold-Start Recommendations Matrix Factorization techniques for product recommendations, although efficient, suffer from serious cold-start problems. The cold start problem concerns with the recommendations for users with no or few past history i.e new users. Providing recommendations to such users becomes a difficult problem for recommendation models because their learning and predictive ability are limited. Google researchers have come up with a meta-learning strategy to address item cold-start when new items arrive continuously. Their paper “A Meta-Learning Perspective on Cold-Start Recommendations for Items” has two deep neural network architectures that implement this meta-learning strategy. The first architecture learns a linear classifier whose weights are determined by the item history while the second architecture learns a neural network whose biases are instead adjusted. On evaluating this technique on the real-world problem of Tweet recommendation, the proposed techniques significantly beat the MF baseline. Image-to-Image Translation NIPS 2017 exhibited a new image-to-image translation system, a model to hide images within images, and use of feature transforms to improve universal style. Unsupervised Image-to-Image Translation Researchers at Nvidia have proposed an unsupervised image-to-image translation framework based on Coupled GANs. Unsupervised image-to-image translation learns a joint distribution of images in different domains by using images from the marginal distributions in individual domains. However, there exists an infinite set of joint distributions that can arrive from the given marginal distributions. So, one could infer nothing about the joint distribution from the marginal distributions, without additional assumptions. Their paper “Unsupervised Image-to-Image Translation Networks ” uses a shared-latent space assumption to address this issue. Their method presents high-quality image translation results on various challenging unsupervised image translation tasks, such as street scene image translation, animal image translation, and face image translation. Deep Steganography Steganography is commonly used to unobtrusively hide a small message within the noisy regions of a larger image. Google researchers in their paper “Hiding Images in Plain Sight: Deep Steganography” have demonstrated the successful application of deep learning to hiding images. They have placed a full-size color image within another image of the same size. They have also trained Deep neural networks to create the hiding and revealing processes and are designed to specifically work as a pair. Their approach compresses and distributes the secret image's representation across all of the available bits, instead of encoding the secret message within the least significant bits of the carrier image. This system is trained on images drawn randomly from the ImageNet database and works well on natural images. Improving Universal style transfer on images NIPS 2017 witnessed another paper aimed at improving the Universal Style Transfer. Universal style transfer is used for transferring arbitrary visual styles to content images. The paper “Universal Style Transfer via Feature Transforms” by Nvidia researchers highlight feature transforms, as a simple yet effective method to tackle the limitations of existing feed-forward methods for Universal Style Transfer, without training on any pre-defined styles. Existing feed-forward based methods are mainly limited by the inability of generalizing to unseen styles or compromised visual quality. The research paper embeds a pair of feature transforms, whitening and coloring, to an image reconstruction network. The whitening and coloring transform reflect a direct matching of feature covariance of the content image to a given style image. The algorithm can generate high-quality stylized images with comparisons to a number of recent methods. Key Takeaways from NIPS 2017 The Research papers covered in this and the previous post highlight that most organizations are at the forefront of machine learning and are actively exploring virtually all aspects of the field. Deep learning practices were also in trend. The conference was focussed on the current state and recent advances in Deep Learning. A lot of talks and presentations were about industry-ready neural networks suggesting a fast transition from research to industry. Researchers are also focusing on areas of language understanding, speech recognition, translation, visual processing, and prediction. Most of these techniques rely on using GANs as the backend. For live content coverage, you can visit NIPS’ Facebook page.

The ongoing 31st annual Conference on Neural Information Processing Systems (NIPS 2017) in Long Beach, California is scheduled from December 4-9, 2017. The 6-day conference is hosting a number of invited talks, demonstrations, tutorials, and paper presentations pertaining to the latest in machine learning, deep learning and AI research. This year the conference has grown larger than life with a record-high 3,240 papers, 678 selected ones, and a completely sold-out event. Top tech members from Google, Microsoft, IBM, DeepMind, Facebook, Amazon, are among other prominent players who enthusiastically participated this year. Here is a quick roundup of some of the top research papers till date. Generative Adversarial Networks Generative Adversarial Networks are a hot topic of research at the ongoing NIPS conference. GANs cast out an easy way to train the DL algorithms by slashing out the amount of data required in training with unlabelled data. Here are a few research papers on GANs. Regularization can stabilize training of GANs Microsoft researchers have proposed a new regularization approach to yield a stable GAN training procedure at low computational costs. Their new model overcomes the fundamental limitation of GANs occurring due to a dimensional mismatch between the model distribution and the true distribution. This results in their density ratio and the associated f-divergence to be undefined. Their paper “Stabilizing Training of Generative Adversarial Networks through Regularization” turns GAN models into reliable building blocks for deep learning. They have also used this for several datasets including image generation tasks. AdaGAN: Boosting GAN Performance Training GANs can at times be a hard task. They can also suffer from the problem of missing modes where the model is not able to produce examples in certain regions of the space. Google researchers have developed an iterative procedure called AdaGAN in their paper “AdaGAN: Boosting Generative Models”, an approach inspired by boosting algorithms, where many potentially weak individual predictors are greedily aggregated to form a strong composite predictor. It adds a new component into a mixture model at every step by running a GAN algorithm on a re-weighted sample. The paper also addresses the problem of missing modes. Houdini: Generating Adversarial Examples The generation of adversarial examples is considered as a critical milestone for evaluating and upgrading the robustness of learning in machines. Also, current methods are confined to classification tasks and are unable to alter the performance measure of the problem at hand. In order to tackle such an issue, Facebook researchers have come up with a research paper titled “Houdini: Fooling Deep Structured Prediction Models”, a novel and a flexible approach for generating adversarial examples distinctly tailormade for the final performance measure of the task taken into account (combinational and non-decomposable tasks). Stochastic hard-attention for Memory Addressing in GANs DeepMind researchers showcased a new method which uses stochastic hard-attention to retrieve memory content in generative models. Their paper titled “Variational memory addressing in generative models” was presented at the 2nd day of the conference and is an advancement over the popular differentiable soft-attention mechanism. Their new technique allows developers to apply variational inference to memory addressing. This leads to more precise memory lookups using target information, especially in models with large memory buffers and with many confounding entries in the memory. Image and Video Processing A lot of hype was also around developing sophisticated models and techniques for image and video processing. Here is a quick glance at the top presentations. Fader Networks: Image manipulation through disentanglement Facebook researchers have introduced Fader Networks, in their paper titled “Fader Networks: Manipulating Images by Sliding Attributes”. These fader networks use an encoder-decoder architecture to reconstruct images by disentangling their salient information and the values of particular attributes directly in a latent space. Disentanglement helps in manipulating these attributes to generate variations of pictures of faces while preserving their naturalness. This innovative approach results in much simpler training schemes and scales for manipulating multiple attributes jointly. Visual interaction networks for Video simulation Another paper titled “Visual interaction networks: Learning a physics simulator from video Tuesday” proposes a new neural-network model to learn physical objects without prior knowledge. Deepmind’s Visual Interaction Network is used for video analysis and is able to infer the states of multiple physical objects from just a few frames of video. It then uses these to predict object positions many steps into the future. It can also deduce the locations of invisible objects. Transfer, Reinforcement, and Continual Learning A lot of research is going on in the field of Transfer, Reinforcement, and Continual learning to make stable and powerful deep learning models. Here are a few research papers presented in this domain. Two new techniques for Transfer Learning Currently, a large set of input/output (I/O) examples are required for learning any underlying input-output mapping. By leveraging information based on the related tasks, the researchers at Microsoft have addressed the problem of data and computation efficiency of program induction. Their paper “Neural Program Meta-Induction” uses two approaches for cross-task knowledge transfer. First is Portfolio adaption, where a set of induction models is pretrained on a set of related tasks, and the best model is adapted towards the new task using transfer learning. The second one is Meta program induction, a k-shot learning approach which makes a model generalize itself to new tasks without requiring any additional training. Hybrid Reward Architecture to solve the problem of generalization in Reinforcement Learning A new paper from Microsoft “Hybrid Reward Architecture for Reinforcement Learning” highlights a new method to address the generalization problem faced by a typical deep RL method. Hybrid Reward Architecture (HRA) takes a decomposed reward function as the input and learns a separate value function for each component reward function. This is especially useful in domains where the optimal value function cannot easily be reduced to a low-dimensional representation. In the new approach, the overall value function is much smoother and can be easier approximated by a low-dimensional representation, enabling more effective learning. Gradient Episodic Memory to counter catastrophic forgetting in continual learning models Continual learning is all about improving the ability of models to solve sequential tasks without forgetting previously acquired knowledge. In the paper “Gradient Episodic Memory for Continual Learning”, Facebook researchers have proposed a set of metrics to evaluate models over a continuous series of data. These metrics characterize models by their test accuracy and the ability to transfer knowledge across tasks. They have also proposed a model for continual learning, called Gradient Episodic Memory (GEM) that reduces the problem of catastrophic forgetting. They have also experimented with variants of the MNIST and CIFAR-100 datasets to demonstrate the performance of GEM when compared to other methods. In our next post, we will cover a selection of papers presented so far at NIPS 2017 in the areas of Predictive Modelling, Machine Translation, and more. For live content coverage, you can visit NIPS’ Facebook page.