Tech News - Data

At the ongoing Microsoft Connect(); 2017, Microsoft has unveiled their latest innovations in AI development platforms. The Connect(); conference this year is all about developing new tools and cloud services that help developers seize the growing opportunity around artificial intelligence and machine learning. Microsoft has made two major announcements to capture the AI market. Visual Studio Tools for AI Microsoft has announced new tools for its Visual Studio IDE specific for building AI applications. Visual Studio Tools for AI is currently in the beta stage and is an extension to the Visual Studio 2017. It allows developers, data scientists, and machine learning engineers to embed deep learning models into applications. They also have built-in support for popular machine learning frameworks such as Microsoft Cognitive Toolkit (CNTK), Google TensorFlow, Caffe2, and MXNet. It also comes packed with features such as custom metrics, history tracking, enterprise-ready collaboration, and data science reproducibility and auditing. Visual Studio Tools for AI allows interactive debugging of deep learning applications with built-in features like syntax highlighting, IntelliSense and text auto formatting. Training of AI models on the cloud is also possible using the integration with Azure Machine Learning. This integration also allows deploying a model into production. Visualization and monitoring of AI models is available using TensorBoard, which is an integrated open tool and can be run both locally and in remote VMs. Azure IoT Edge Microsoft sees IoT as a mission-critical business asset. With this in mind, they have developed a product for IoT solutions. Termed as Azure IoT Edge, it enables developers to run cloud intelligence on the edge of IoT devices. Azure IoT Edge can operate on Windows and Linux as well as on multiple hardware architectures (x64 and ARM). Developers can work on languages such as C#, C and Python to deploy models on Azure IoT Edge. The Azure IoT edge is a bundle of multiple components. With AI Toolkit, developers can start building AI applications. With Azure Machine learning, AI applications can be created, deployed, and managed with the toolkit on any framework. Azure Machine Learning also includes a set of pre-built AI models for common tasks. In addition, using the Azure IoT Hub, developers can deploy Edge modules on multiple IoT Edge devices. Using a combination of Azure Machine Learning, Azure Stream Analytics, Azure Functions, and any third-party code, a complex data pipeline can be created to build and test container-based workloads. This pipeline can be managed using the Azure IoT Hub. The customer reviews on Azure IoT edge have been positive up till now. Here’s what Matt Boujonnier, Analytics Application Architect at Schneider Electric says: "Azure IoT Edge provided an easy way to package and deploy our Machine Learning applications. Traditionally, machine learning is something that has only run in the cloud, but for many IoT scenarios that isn’t good enough, because you want to run your application as close as possible to any events. Now we have the flexibility to run it in the cloud or at the edge—wherever we need it to be." With the launch of these two new tools, Microsoft is catching up quickly with the likes of Google and IBM to capture the AI market and providing developers with an intelligent edge.

YouTube updates its policies to ban dangerous pranks and challenges that can be harmful to the victim of a prank or encourages people to partake in dangerous behavior. Pranks and challenges have been around on YouTube for a long time. Many of the pranks are entertaining and harmless, some challenges potentially unsafe like an extreme food eating challenge. Recently, the “Bird Box Challenge” has been popular inspired after the Netflix movie Bird Box. The challenge is to perform difficult tasks, like driving a car, blindfolded. This challenge has received media coverage not for the entertainment value but for the dangers involved. It has caused many accidents where people take this challenge. What is banned on YouTube? In the light of this challenge being harmful and dangerous to lives, YouTube bans certain content by updating its policies page. Primarily, it has banned three kinds of pranks: Challenges that can cause serious danger to life or cause death Pranks that lead the victims to believe that they’re in serious physical danger Any pranks that cause severe emotional distress in children They state in their policies page: “YouTube is home to many beloved viral challenges and pranks, but we need to make sure what’s funny doesn’t cross the line into also being harmful or dangerous.” What are the terms? Other than the points listed above there is no clear or exhaustive list of the kind of activities that are banned. The YouTube moderators may take a call to remove a video. In the next two months, YouTube will be removing any existing content that falls into this radar, however, content creators will not receive a strike. Going forward, any new content that may have objectionable content as per their policies will get the channel a ‘strike’. Three strikes in the span of three months will lead to the channel’s termination. Questionable content includes custom thumbnails or external links that display pornographic, graphic violent, malware, or spam content. So now you are less likely to see videos on driving blindfolded or eating tide pods. Google Chrome announces an update on its Autoplay policy and its existing YouTube video annotations Is the YouTube algorithm’s promoting of #AlternativeFacts like Flat Earth having a real-world impact? Worldwide Outage: YouTube, Facebook, and Google Cloud goes down affecting thousands of users

Qlik’s new CEO Mike Capone keynoted the recently held Qlik Qonnections 2018, with some interesting feature rollouts and announcements. He also shed light on the evolution of Qlik’s two premium products - Qlikview and Qlik Sense, and shared their roadmap for the coming year. Close to 4000 developers and Business Intelligence professionals were in attendance, and were very receptive to the positive announcements made in the keynote. Let us take a quick look at some of the important announcements: Qlik continues to be the market leader Capone began the keynote by sharing some of the interesting performance metrics over the past year, which have led to Qlik being listed as a ‘Leader’ in the Gartner Magic Quadrant 2017. One of the most impressive achievements among all is the impressive customer base that Qlik boasts of, including: 9 out of the 10 major banks 8 out of the 10 major insurance companies 11 out of the 15 major global investment and securities companies With an impressive retention rate of 94%, Qlik have also managed to add close to 4000 new customers over the last year and have also doubled the developer community to over 25,000 members. These numbers mean only one thing - Qlik will continue to dominate. Migration from Qlikview to Qlik Sense There has been a lot of talk (and confusion) of late about Qlik supposedly looking to transition its focus from Qlikview to Qlik Sense. In the keynote, Capone gave us all the much needed clarity on the licensing and migration options for those looking to move from Qlikview’s guided analytics features to Qlik Sense’s self-service analytics. These are some of the important announcements in this regard: Migration from Qlikview to Qlik Sense is optional: Acknowledging some of the loyal customers who don’t want to move away from QlikView, Capone said that the migration from Qlikview to Qlik Sense is optional. For those who do want to migrate, Qlik have assured that the transition will be made as smooth as possible, and that they would be making this a priority. Single license to use both Qlikview and Qlik Sense: Qlik have made it possible for customers to get the most out of their products without having to buy multiple licenses for multiple products. With just an additional maintenance fee, they will be able to enjoy the premium features of both the tools seamlessly. Qlik venturing into cognitive analytics One of the most notable announcements of this conference was incorporating aspects of Artificial Intelligence into the Business Intelligence capabilities of the Qlik products. Qlik are aiming to improving the core associative engine that works with the available data smartly. Not just that, they have also announced the Insight Advisor feature, to auto-generate the best possible visualizations and reports. Hybrid and multi-cloud support added Qlik’s vision going forward is quite simple and straightforward - to support deployment of their applications and services in a hybrid-cloud or multi-cloud environment. Going forward, users will be able to move their Qlik Sense applications that run using a microservices-based architecture on Linux, in either public or private clouds. They will also be able to self-manage these applications with the support features provided by Qlik. New tools for Qlik developers Qonnections 2018 saw 2 important announcements made to make the lives of Qlik developers easier. Along with Qlik Branch - a platform to collaborate on projects and share innovations and new developments, Qlik also announced a new platform for developers called Qlik Core. This new platform will allow Qlik developers to leverage the offerings of IoT, edge analytics and more to design and drive innovative business models and strategies. Qlik Core is currently in the beta stage, and is expected to be generally available very soon. Interesting times ahead for Qlik In recent times, Qlik has faced stiff competition from other popular Business Intelligence tools such as Tableau, Spotfire, Microsoft’s very own Power BI - apart from the freely available tools which are easily available to customers for fast, effective business intelligence. With all the tools delivering on a similar promise and not coming out with any groundbreaking blue ocean features, it will be interesting to see how Qlik’s new offerings will fare against these sharks. The recent restructuring of the Qlik management and the downsizing happening over the past few years can make one wonder if they are struggling to keep up. However, the announcements in Qonnections 2018 indicate the company is indeed moving in a positive direction with their products, and should restore the public faith and dispel any doubts Qlik’s customers may have. How Qlik Sense is driving self-service Business Intelligence Overview of a Qlik Sense® Application’s Life Cycle QlikView Tips and Tricks

On Monday, Deezer, a French online music streaming service, released Spleeter which is a music separation engine.  It comes in the form of a Python Library based on Tensorflow. Stating the reason behind Spleeter, the researchers state, “We release Spleeter to help the Music Information Retrieval (MIR) community leverage the power of source separation in various MIR tasks, such as vocal lyrics analysis from audio, music transcription, any type of multilabel classification or vocal melody extraction.” Spleeter comes with pre-trained models for 2, 4 and 5 track separation. These include: Vocals (singing voice) / accompaniment separation (2 stems) Vocals / drums / bass / other separation (4 stems) Vocals / drums / bass / piano / other separation (5 stems) It can also train source separation models or fine-tune pre-trained ones with Tensorflow if you have a dataset of isolated sources. Deezer benchmarked Spleeter against Open-Unmix another open-source model recently released and reported slightly better performances with increased speed. It can perform separation of audio files to 4 stems 100x faster than real-time when running on a GPU. You can use Spleeter straight from the command line as well as directly in your own development pipeline as a Python library. It can be installed with Conda, with pip or be used with Docker. Spleeter creators mention a number of potential applications of source separation engine including remixes, upmixing, active listening, educational purposes, and pre-processing for other tasks such as transcription. Spleeter received mostly positive feedback on Twitter, as people experimented to separate vocals from music. https://twitter.com/lokijota/status/1191580903518228480 https://twitter.com/bertboerland/status/1191110395370586113 https://twitter.com/CholericCleric/status/1190822694469734401 Wavy.org also ran several songs through the two-stem filter and evaluated them in a blog post. They tried a variety of soundtracks across multiple genres. The performance of audio was much better than expected, however, vocals sometimes felt robotically autotuned. The amount of bleed was shockingly low relative to other solutions and surpassed any available free tool and rival commercial plugins and services. https://twitter.com/waxpancake/status/1191435104788238336 Spleeter will be presented and live-demoed at the 2019 ISMIR conference in Delft. For more details refer to the official announcement. DeepMind AI’s AlphaStar achieves Grandmaster level in StarCraft II with 99.8% efficiency. Google AI introduces Snap, a microkernel approach to ‘Host Networking’ Firefox 70 released with better security, CSS, and JavaScript improvements

In a research published in the Nature journal on Monday, a team of neuroscientists from the University of California, San Francisco, introduced a neural decoder that can synthesize natural-sounding speech based on brain activity. This research was led by Gopala Anumanchipalli, a speech scientist, and Josh Chartier, a bioengineering graduate student in the Chang lab. It is being developed in the laboratory of Edward Chang, a Neurological Surgery professor at University of California. Why is this neural decoder being introduced? There are many cases of people losing their voice because of stroke, traumatic brain injury, or neurodegenerative diseases such as Parkinson’s disease, multiple sclerosis, and amyotrophic lateral sclerosis. Currently,assistive devices that track very small eye or facial muscle movements to enable people with severe speech disabilities express their thoughts by writing them letter-by-letter, do exist. However, generating text or synthesized speech with such devices is often time consuming, laborious, and error-prone. Another limitation these devices have is that they only permit generating a maximum of 10 words per minute, compared to the 100 to 150 words per minute of natural speech. This research shows that it is possible to generate a synthesized version of a person’s voice that can be controlled by their brain activity. The researchers believe that in future, this device could be used to enable individuals with severe speech disability to have fluent communication. It could even reproduce some of the “musicality” of the human voice that expresses the speaker’s emotions and personality. “For the first time, this study demonstrates that we can generate entire spoken sentences based on an individual’s brain activity,” said Chang. “This is an exhilarating proof of principle that with technology that is already within reach, we should be able to build a device that is clinically viable in patients with speech loss.” How does this system work? This research is based on another study by Josh Chartier and Gopala K. Anumanchipalli, which shows how the speech centers in our brain choreograph the movements of the lips, jaw, tongue, and other vocal tract components to produce fluent speech. In this new study, Anumanchipalli and Chartier asked five patients being treated at the UCSF Epilepsy Center to read several sentences aloud. These patients had electrodes implanted into their brains to map the source of their seizures in preparation for neurosurgery. Simultaneously, the researchers recorded activity from a brain region known to be involved in language production. The researchers used the audio recordings of volunteer’s voice to understand the vocal tract movements needed to produce those sounds. With this detailed map of sound to anatomy in hand, the scientists created a realistic virtual vocal tract for each volunteer that could be controlled by their brain activity. The system comprised of two neural networks: A decoder for transforming brain activity patterns produced during speech into movements of the virtual vocal tract. A synthesizer for converting these vocal tract movements into a synthetic approximation of the volunteer’s voice. Here’s a video depicting the working of this system: https://www.youtube.com/watch?v=kbX9FLJ6WKw&feature=youtu.be The researchers observed that the synthetic speech produced by this system was much better as compared to the synthetic speech directly decoded from the volunteer’s brain activity. The generated sentences were also understandable to hundreds of human listeners in crowdsourced transcription tests conducted on the Amazon Mechanical Turk platform. The system is still in its early stages. Explaining its limitations, Chartier said, “We still have a ways to go to perfectly mimic spoken language. We’re quite good at synthesizing slower speech sounds like ‘sh’ and ‘z’ as well as maintaining the rhythms and intonations of speech and the speaker’s gender and identity, but some of the more abrupt sounds like ‘b’s and ‘p’s get a bit fuzzy. Still, the levels of accuracy we produced here would be an amazing improvement in real-time communication compared to what’s currently available.” Read the full report on UCSF’s official website. OpenAI introduces MuseNet: A deep neural network for generating musical compositions Interpretation of Functional APIs in Deep Neural Networks by Rowel Atienza Google open-sources GPipe, a pipeline parallelism Library to scale up Deep Neural Network training  

A paper on a new simple network architecture, the Transformer, based solely on attention mechanisms The NIPS 2017 accepted paper, Attention Is All You Need, introduces Transformer, a model architecture relying entirely on an attention mechanism to draw global dependencies between input and output. This paper is authored by professionals from the Google research team including Ashish Vaswani, Noam Shazeer, Niki Parmar,  Jakob Uszkoreit, Llion Jones, Aidan N. Gomez, Łukasz Kaiser, and Illia Polosukhin. The Transformer – Attention is all you need What problem is the paper attempting to solve? Recurrent neural networks (RNN), long short-term memory networks(LSTM) and gated RNNs are the popularly approaches used for Sequence Modelling tasks such as machine translation and language modeling. However, RNN/CNN handle sequences word-by-word in a sequential fashion. This sequentiality is an obstacle toward parallelization of the process. Moreover, when such sequences are too long, the model is prone to forgetting the content of distant positions in sequence or mix it with following positions’ content. Recent works have achieved significant improvements in computational efficiency and model performance through factorization tricks and conditional computation. But they are not enough to eliminate the fundamental constraint of sequential computation. Attention mechanisms are one of the solutions to overcome the problem of model forgetting. This is because they allow dependency modelling without considering their distance in the input or output sequences. Due to this feature, they have become an integral part of sequence modeling and transduction models. However, in most cases attention mechanisms are used in conjunction with a recurrent network. Paper summary The Transformer proposed in this paper is a model architecture which relies entirely on an attention mechanism to draw global dependencies between input and output. The Transformer allows for significantly more parallelization and tremendously improves translation quality after being trained for as little as twelve hours on eight P100 GPUs. Neural sequence transduction models generally have an encoder-decoder structure. The encoder maps an input sequence of symbol representations to a sequence of continuous representations. The decoder then generates an output sequence of symbols, one element at a time. The Transformer follows this overall architecture using stacked self-attention and point-wise, fully connected layers for both the encoder and decoder. The authors are motivated to use self-attention because of three criteria.   One is that the total computational complexity per layer. Another is the amount of computation that can be parallelized, as measured by the minimum number of sequential operations required. The third is the path length between long-range dependencies in the network. The Transformer uses two different types of attention functions: Scaled Dot-Product Attention, computes the attention function on a set of queries simultaneously, packed together into a matrix. Multi-head attention, allows the model to jointly attend to information from different representation subspaces at different positions. A self-attention layer connects all positions with a constant number of sequentially executed operations, whereas a recurrent layer requires O(n) sequential operations. In terms of computational complexity, self-attention layers are faster than recurrent layers when the sequence length is smaller than the representation dimensionality, which is often the case with machine translations. Key Takeaways This work introduces Transformer, a novel sequence transduction model based entirely on attention mechanism. It replaces the recurrent layers most commonly used in encoder-decoder architectures with multi-headed self-attention. Transformer can be trained significantly faster than architectures based on recurrent or convolutional layers for translation tasks. On both WMT 2014 English-to-German and WMT 2014 English-to-French translation tasks, the model achieves a new state of the art.  In the former task the model outperforms all previously reported ensembles. Future Goals Transformer has only been applied to transduction model tasks as of yet. In the near future, the authors plan to use it for other problems involving input and output modalities other than text. They plan to apply attention mechanisms to efficiently handle large inputs and outputs such as images, audio and video. The Transformer architecture from this paper has gained major traction since its release because of major improvements in translation quality and other NLP tasks. Recently, the NLP research group at Harvard have released a post which presents an annotated version of the paper in the form of a line-by-line implementation. It is accompanied with 400 lines of library code, written in PyTorch in the form of a notebook, accessible from github or on Google Colab with free GPUs.  

Patreon, a popular crowdfunding platform published a post yesterday in defense of its removal of Sargon of Akkad or Carl Benjamin, an English YouTuber famous for his anti-feminist content, last week, over the concerns of him violating its policies on hate speech. Patreon has been receiving backlash ever since from the users and patrons of the website who are calling for a boycott. “Patreon does not and will not condone hate speech in any of its forms. We stand by our policies against hate speech. We believe it’s essential for Patreon to have strong policies against hate speech to build a safe community for our creators and their patrons”, says the Patreon team. Patreon mentioned that it reviews the creations posted by the content creators on other platforms that are funded via Patreon. Since Benjamin is quite popular for his collaborations with other creators, Patreon’s community guidelines, which strictly prohibits hate speech also get applied to those collaborations. According to Patreon’s community guidelines, “Hate speech includes serious attacks, or even negative generalizations, of people based on their race [and] sexual orientation.” Benjamin in one of his interviews on another YouTuber’s channel used racial slurs linked with “negative generalizations of behavior” quite contrasting to how people of those races actually act, to insult others. Apart from using racial slurs, he also used sexual orientation related slurs which violates Patreon’s community guidelines. However, a lot of people are not happy with Patreon’s decision. For instance, Sam Harris, a popular American author, podcast host, and neuroscientist, who had one of the top-grossing accounts (with nearly 9,000 paying patrons at the end of November) on Patreon deleted his account earlier this week, accusing the platform of “political bias”. He wrote “the crowdfunding site Patreon has banned several prominent content creators from its platform. While the company insists that each was in violation of its terms of service, these recent expulsions seem more readily explained by political bias. I consider it no longer tenable to expose any part of my podcast funding to the whims of Patreon’s ‘Trust and Safety” committee’”.     https://twitter.com/SamHarrisOrg/status/1074504882210562048 Apart from banning Carl Benjamin, Patreon also banned Milo Yiannopoulos, a British public speaker and YouTuber with over 839,286 subscribers earlier this month over his association with the Proud Boys, which Patreon has classified as a hate group. https://twitter.com/Patreon/status/1070446085787668480 James Allsup, an alt-right political commentator, and associate of Yiannopoulus', was also banned from Patreon last month for their association with hate groups. Amidst this controversy, some of the top Patreon creators such as Jordan Peterson, a popular Canadian clinical psychologist whose YouTube channel has over 1.6 M subscribers and Dave Rubin, an American libertarian political commentator announced their plans of starting an alternative to Patreon, earlier this week. Peterson said that the new platform will work on a subscriber model similar to Patreon’s, only with few additional features. https://www.youtube.com/watch?v=GWz1RDVoqw4 “We understand some people don’t believe in the concept of hate speech and don’t agree with Patreon removing creators on the grounds of violating our Community Guidelines for using hate speech. We have a different view,” says the Patreon team. Emmanuel Macron teams up with Facebook in a bid to fight hate speech on social media Twitter takes action towards dehumanizing speech with its new policy How IRA hacked American democracy using social media and meme warfare to promote disinformation and polarization: A new report to Senate Intelligence Committee

Google researchers have build a tool called JAX, a domain-specific tracing JIT compiler, which generates high-performance accelerator code from pure Python and Numpy machine learning programs. It combines Autograd and XLA for high-performance machine learning research. At its core, it is an extensible system for transforming numerical functions. Autograd helps JAX automatically differentiate native Python and Numpy code. It can handle a large subset of Python features such as loops, branches, recursion, and closures. It comes with support for reverse-mode (backpropagation) and forward-mode differentiation, and these two can be composed arbitrarily in any order. XLA or Accelerated Linear Algebra is a linear algebra compiler used for optimizing TensorFlow computations. To run the NumPy programs on GPUs and TPUs, JAX uses XLA. The library calls are compiled and executed just-in-time. JAX also allows compiling your own Python functions just-in-time into XLA-optimized kernels using a one-function API, jit. How JAX works? The basic function of JAX is specializing and translating high-level Python and NumPy functions into a representation that can be transformed and then lifted back into a Python function. It traces Python functions by monitoring all the basic operations applied to its input to produce output and then records these operations and the data-flow between them in a directed acyclic graph (DAG). For tracing the functions, it wraps primitive operations and when they’re called they add themselves to a list of operations performed along with their inputs and outputs. In order to keep track of the data flow between these primitive operations, the values being tracked are wrapped in the Tracer class instances. The team is working towards expanding this project and provide support for cloud TPU, multi-GPU, and multi-TPU. In future, it will come with full NumPy coverage and some SciPy coverage, and more. As this is still a research project, we can expect bugs and is not recommended to be used in production. To read more in detail and contribute to this project, head over to GitHub. Google AdaNet, a TensorFlow-based AutoML framework Graph Nets – DeepMind’s library for graph networks in Tensorflow and Sonnet Dopamine: A Tensorflow-based framework for flexible and reproducible Reinforcement Learning research by Google

Yesterday, Benjamin Groessing, a web consultant and developer at byteq, released remove.bg, a tool built on python, ruby and deep learning. This tool automatically removes the background of any image within 5 seconds. It uses various custom algorithms for the processing of the image. https://twitter.com/hammer_flo_/status/1074914463726350336 It is a free service and users don’t have to manually select the background/foreground layers to separate them. One can simply select an image and instantly download the resulting image with the background removed. Features of remove.bg Personal and professional use Remove.bg can be used by graphic designer, photographer or selfie lover for removing backgrounds. Saves time and money It saves time as it is automated and it is free of cost. 100% Automatic Apart from the image file, this release doesn’t require inputs such as selecting pixels, marking persons, etc. How does remove.bg work? https://twitter.com/begroe/status/1074645152487129088 Remove.bg uses AI technology for detecting foreground layers and separating them from the background. It uses additional algorithms for improving fine details and preventing color contamination. The AI detects persons as foreground and everything else as background. So, it only works if there is at least one person in the image. Users can upload images of any resolution but for performance reasons, the output image has been limited to 500 × 500 pixels. Privacy in remove.bg User images are uploaded through a secure SSL/TLS-encrypted connection. These images are processed and the result is temporarily stored till the time a user can download them. After which, approximately an hour later, these image files get deleted. Privacy message on the official website of remove.bg states, “We do not share your images or use them for any other purpose than removing the background and letting you download the result.” What can be expected from the next release? The next set of releases might support other kinds of images such as product images. The team at Remove.bg might also release an easy-to-use API. Users are very excited about this release and the technology used behind it. Many users are comparing it with the portrait mode on iPhone X. Though it is not that fast but users are still liking it. https://twitter.com/Baconbrix/status/1074805036264316928 https://twitter.com/hammer_flo_/status/1074914463726350336 But how strong is remove.bg with regards to privacy is a bigger question. Though the website gives a privacy note at the end but it will take more to win the user’s trust. The images uploaded to remove.bg’ cloud might be at risk. How strong is the security and what preventive measures have they taken? These are few of the questions that might bother many. To have a look at the ongoing discussion on remove.bg, check out Benjamin Groessing’s AMA twitter thread. Facebook open-sources PyText, a PyTorch based NLP modeling framework Deep Learning Indaba presents the state of Natural Language Processing in 2018 NYU and AWS introduce Deep Graph Library (DGL), a python package to build neural network graphs

In a GitHub pull request last week, the NumPy community decided to remove support for Python 2.7. Python 3.4 support will also be dropped with this pull request. So now, to use NumPy 1.17 and newer versions, you will need Python 3.5 or later. NumPy has been supporting both Python versions since 2010. This move doesn't come as a surprise with the Python core team itself dropping support for Python 2 in 2020. The NumPy team had mentioned that this move comes in “Python 2 is an increasing burden on our limited resources”. The discussion to drop Python 2 support in NumPy started almost a year ago. Running pip install numpy on Python 2 will still install the last working version. But here on now, it may not contain the latest features as released for Python 3.5 or higher. However, NumPy on Python 2 will still be supported until December 31, 2019. After January 1, 2020, it may not contain the newest bug fixes. The Twitter audience sees this as a welcome move: https://twitter.com/TarasNovak/status/1073262599750459392 https://twitter.com/esc___/status/1073193736178462720 A comment on Hacker News reads: “Let's hope this move helps with the transitioning to Python 3. I'm not a Python programmer myself, but I'm tired of things getting hairy on Linux dependencies written in Python. It almost seems like I always got to have a Python 2 and a Python 3 version of some packages so my system doesn't break.” Another one reads: “I've said it before, I'll say it again. I don't care for everything-is-unicode-by-default. You can take my Python 2 when you pry it from my cold dead hands.” Some researchers who use NumPy and SciPy stick Python 2, this move from the NumPy team will help in getting everyone to work on a single version. One single supported version will sure help with the fragmentation. Often, Python developers find themselves in a situation where they have one version installed and a specific module is available/works properly in another version. Some also argue about stability, that Python 2 has greater stability and x or y feature. But the general sentiment is more supportive of adopting Python 3. Introducing numpywren, a system for linear algebra built on a serverless architecture NumPy 1.15.0 release is out! Implementing matrix operations using SciPy and NumPy  

Last week, a Jupyter Community Workshop on dashboarding was held in Paris. At the workshop, several contributors came together to build the Voila package, the details of which QuantStack shared yesterday. Voila serves live Jupyter notebooks as standalone web applications providing a neat way to share your work results with colleagues. Why do we need Voila? Jupyter notebooks allow you to do something called “literature programming” in which human-friendly explanations are accompanied with code blocks. It allows scientists, researchers, and other practitioners of scientific computing to add theory behind their code including mathematical equations. However, Jupyter notebooks may prove to be a little bit problematic when you plan to communicate your results with other non-technical stakeholders. They might get put-off by the code blocks and also the need for running the notebook to see the results. It also does not have any mechanism to prevent arbitrary code execution by the end user. How Voila works? Voila addresses all the aforementioned queries by converting your Jupyter notebook to a standalone web application. After connecting to a notebook URL, Voila launches the kernel for that notebook and runs all the cells. Once the execution is complete, it does not shut down the kernel. The notebook gets converted to HTML and is served to the user. This rendered HTML includes JavaScript that is responsible for initiating a websocket connection with the Jupyter kernel. Here’s a diagram depicting how it works: Source: Jupyter Blog Following are the features Voila provides: Renders Jupyter interactive widgets: It supports Jupyter widget libraries including bqplot, ipyleafet, ipyvolume, ipympl, ipysheet, plotly, and ipywebrtc. Prevents arbitrary code execution: It does not allow arbitrary code execution by consumers of dashboards. A language-agnostic dashboarding system: Voila is built upon Jupyter standard protocols and file formats enabling it to work with any Jupyter kernel (C++, Python, Julia). Includes custom template system for better extensibility: It provides a flexible template system to produce rich application layouts. Many Twitter users applauded this new way of creating live and interactive dashboards from Jupyter notebooks: https://twitter.com/philsheard/status/1138745404772818944 https://twitter.com/andfanilo/status/1138835776828071936 https://twitter.com/ToluwaniJohnson/status/1138866411261124608 Some users also compared it with another dashboarding solution called Panel. The main difference between Panel and Voila is that Panel supports Bokeh widgets whereas Voila is framework and language agnostic. “Panel can use a Bokeh server but does not require it; it is equally happy communicating over Bokeh Server's or Jupyter's communication channels. Panel doesn't currently support using ipywidgets, nor does Voila currently support Bokeh plots or widgets, but the maintainers of both Panel and Voila have recently worked out mechanisms for using Panel or Bokeh objects in ipywidgets or using ipywidgets in Panels, which should be ready soon,” a Hacker News user commented. To read more in detail about Voila, check out the official announcement on the Jupyter Blog. JupyterHub 1.0 releases with named servers, support for TLS encryption and more Introducing Jupytext: Jupyter notebooks as Markdown documents, Julia, Python or R scripts JupyterLab v0.32.0 releases

Google released version 4.0 of its OCR engine, Tesseract, yesterday. Tesseract 4.0 comes with a new neural net (LSTM) based OCR engine, updated build system, other improvements, and bug fixes. Tesseract is an OCR engine that offers support for unicode (a specification that supports all character set) and comes with an ability to recognize more than 100 languages out of the box. It can be trained to recognize other languages and is used for text detection on mobile devices, videos, and in Gmail image spam detection. Let’s have a look at what's new in Tesseract 4.0. New neural net (LSTM) based OCR engine The new OCR engine uses a neural network system based on LSTMs, with major accuracy gains. This consists of new training tools for the LSTM OCR engine. You can train a new model from scratch or by fine-tuning an existing model. Trained data including LSTM models and 123 languages have been added to the new OCR engine. Optional accelerated code paths have been added for the LSTM recognizer: Moreover, a new parameter lstm_choice_mode that allows including alternative symbol choices in the hOCR output has been added. Updated Build System Tesseract 4.0 uses semantic versioning and requires Leptonica 1.74.0 or a higher version. In case you want to build Tesseract from source code then a compiler with strong C++ 11 support is necessary. Unit tests have been added to the main repo. Tesseract's source tree has been reorganized in version 4.0. A new option has been added that lets you compile Tesseract without the code of the legacy OCR engine. Bug Fixes Issues in trainingdata rendering have been fixed. Damage caused to binary images when processing PDFs has been fixed. Issues in the OpenCL code have been fixed. OpenCL now works fine for the legacy Tesseract OCR engine but the performance hasn’t improved yet. Other Improvements Multi-page TIFF handling is improved in Tesseract 4.0. Improvements are made to PDF rendering. The version information and improved help texts have been added to the training tools. tessedit_pageseg_mode 1 has been removed from hocr, pdf, and tsv config files. The user has to now explicitly use --psm 1 if that is desired. For more information, check out the official release notes. Tesla v9 to incorporate neural networks for autopilot Neural Network Intelligence: Microsoft’s open source automated machine learning toolkit

For decades, Intel and AMD have remained bitter archrivals. Today, they find themselves teaming up to thwart a common enemy – Nvidia. As Intel revealed its partnership with Advanced Micro Devices (AMD) over a next-generation notebook chip, it was the first time the two chip giants collaborated since the ‘80s. The proposed chip for thin and lightweight laptops combines an Intel processor and an AMD graphics unit for complex video gaming. The new series of processors will be part of Intel's 8th-generation Core H-series mobile chips, expected to hit the market in the first quarter of 2018. What it means is that Intel’s high-performance x86 cores will get combined with AMD Radeon Graphics into the same processor package using Intel’s EMIB multi-die technology. That is not all. Intel is also bundling the design with built-in High Bandwidth Memory (HBM2) RAM. The new processor, Intel claims, reduces the usual silicon footprint by about 50%. And with a ‘semi-custom’ graphics processor from AMD, enthusiasts can look forward to discrete graphics-level performances for playing games, editing photos or videos, and other tasks that can leverage modern GPU technologies. What does AMD get? Having struggled to remain profitable in recent times, AMD has been losing share in the discrete notebook GPU market. The deal could bring additional revenues with increased market share. Most importantly, the laptops built with the new processors won’t be competing with AMD’s Ryzen chips (which are also designed for ultrathin laptops). AMD clarified on the difference: While the new Intel chips are designed for serious gamers, Ryzen chips (that are due out at the end of the year) can run games but are not specifically designed for that purpose. "Our collaboration with Intel expands the installed base for AMD Radeon GPUs and brings to market a differentiated solution for high-performance graphics,” Scott Herkelman, vice president and general manager of AMD's Radeon Technologies Group, said. "Together we are offering gamers and content creators the opportunity to have a thinner-and-lighter PC capable of delivering discrete performance-tier graphics experiences in AAA games and content creation applications.” While more information will be available in future, the first machines with the new technology are expected to release in the first quarter of 2018. Nvidia's stock fell on the news. While both AMD and Intel saw their shares surging. A rivalry that began when AMD reverse-engineered the Intel 8080 microchip in 1975 could still be far from over, but in graphics, the two have been rather cordial. Despite hating each other since formation, both decided to pick each other as lesser evil over Nvidia. This is why the Intel AMD laptop chip partnership has a definite future. Currently centered around laptop solutions, this could even stretch to desktops, who knows!

Last month, 2Hz introduced an app called krisp which was featured on the Nvidia website. It uses deep learning for noise suppression and is powered by krispNet Deep Neural Network. krispNet is trained to recognize and reduce background noise from real-time audio and yields clear human speech. 2Hz is a company which builds AI-powered voice processing technologies to improve voice quality in communications. What are the limitations in the current ways of noise suppression? Many edge devices from phones, laptops, to conferencing systems come with noise suppression technologies. Latest mobile phones come equipped with multiple microphones which helps suppress environmental noise when we talk. Generally, the first mic is placed on the front bottom of the phone to directly capture the user’s voice. The second mic is placed as far as possible from the first mic. After the surrounding sounds are captured by both these mics, the software effectively subtracts them from each other and yields an almost clean voice. The limitations of multiple mics design: Since multiple mics design requires a certain form factor, their application is only limited to certain use cases such as phones or headsets with sticky mics. These designs make the audio path complicated, requiring more hardware and code. Audio processing can only be done on the edge or device side, thus the underlying algorithm is not very sophisticated due to the low power and compute requirement. The traditional Digital Signal Processing (DSP) algorithms also work well only in certain use cases. Their main drawback is that they are not scalable to variety and variability of noises that exist in our everyday environment. This is why 2Hz has come up with a deep learning solution that uses a single microphone design and all the post processing is handled by a software. This allows hardware designs to be simpler and more efficient. How deep learning can be used in noise suppression? There are three steps involved in applying deep learning to noise suppression: Source: Nvidia Data collection: The first step is to build a dataset to train the network by combining distinct noises and clean voices to produce synthetic noisy speech. Training: Next, feed the synthetic noisy speech dataset to the DNN on input and the clean speech on the output. Inference: Finally, produce a mask which will filter out the noise giving you a clear human voice. What are the advantages of krispNet DNN? krispNet is trained with a very large amount of distinct background noises and clean human voices. It is able to optimize itself to recognize what’s background noise and separate it from a human speech by leaving only the latter. While inferencing, krispNet acts on real-time audio and removes background noise. krispNet DNN can also perform Packet Loss Concealment for audio and fill out missing voice chunks in voice calls by eliminating “chopping”. krispNet DNN can predict higher frequencies of a human voice and produce much richer voice audio than the original lower bitrate audio. Read more in detail about how we can use deep learning in noise suppression on the Nvidia blog. Samsung opens its AI based Bixby voice assistant to third-party developers Voice, natural language, and conversations: Are they the next web UI? How Deep Neural Networks can improve Speech Recognition and generation

DeepLab-v3+, Google’s latest and best performing Semantic Image Segmentation model is now open sourced! DeepLab is a state-of-the-art deep learning model for semantic image segmentation, with the goal to assign semantic labels (e.g., person, dog, cat and so on) to every pixel in the input image. Assigning these semantic labels sets a much stricter localization accuracy requirements than other visual entity recognition tasks such as image-level classification or bounding box-level detection. Examples of semantic image segmentation tasks include synthetic shallow depth-of-field effect shipped in the portrait mode of the Pixel 2 and Pixel 2 XL smartphones and mobile real-time video segmentation. DeepLab-v3+ is implemented in TensorFlow and has its models built on top of a powerful convolutional neural network (CNN) backbone architecture for the most accurate results, intended for server-side deployment. Source: Google Research blog Let’s have a look at some of the highlights of DeepLab v3: Google has extended DeepLab-v3 to include a simple yet effective decoder module to refine the segmentation results especially along object boundaries. In this encoder-decoder structure one can arbitrarily control the resolution of extracted encoder features by atrous convolution to trade-off precision and runtime. They has also shared their Tensorflow model training and evaluation code, along with models already pre-trained on the Pascal VOC 2012 and Cityscapes benchmark semantic segmentation tasks. This version also adopts two network backbones, MobileNetv2 and Xception. MobileNetv2 is a fast network structure designed for mobile devices. Xception is a powerful network structure intended for server-side deployment. You can read more about this announcement on the Google Research blog.