Tech News - Data

A team of researchers, namely, Ismail Akrout, Amal Feriani, and Mohamed Akrout, published a paper, titled ‘Hacking Google reCAPTCHA v3 using Reinforcement Learning’, last month. In the paper, researchers present a Reinforcement Learning (RL) method that can easily bypass Google reCAPTCHA v3. Google’s reCAPTCHA system is used for detection of bots from humans and is the most used defense mechanism. It’s used to protect the sites from automated agents and bots, attacks and spams. Google’s reCAPTCHA v3 makes use of machine learning to return a risk assessment score between 0.0 and 1.0. This score is used to characterize the trustability of the user. If a score is close to 1.0 then that means the user is human, if not, then it’s a bot. Method Used The problem has been formulated as a grid world in which the agents can learn the movement of the mouse and click on the reCAPTCHA button to receive a high score. The performance of the agent is studied on varying the cell size of the world. The paper shows that the performance drops when the agent takes big steps toward the goal. Finally, a divide and conquer strategy is used to defeat the reCAPTCHA system for any grid resolution. Researchers have produced a plausible formalization of the problem as a Markov Decision Process (MDP) that can be solved using advanced RL algorithms. Then, a new environment is introduced that simulates the user experience with websites that have reCAPTCHA system enabled. Finally, it is analyzed how RL agents learn or fail to defeat Google reCAPTCHA.   In order to pass the reCAPTCHA test, a human user is required to move the mouse starting from an initial position then perform a sequence of steps until the user reaches the reCAPTCHA check-box and clicks on it. Based on how the interaction goes, the reCAPTCHA system rewards the user with a score.   As shown in the figure, the point where the mouse is the starting point and goal is the position of reCAPTCHA. A grid is constructed where all the pixels between these two points is a possible position for the mouse. It is assumed in the paper that a normal user will not necessarily move the mouse pixel by pixel, hence, cell size is defined that refers to the number of pixels between these two consecutive positions.                                         Agent’s mouse movement After this, a browser page will be opened at each episode with the user mouse at a random position. The agent then takes in a sequence of actions until it reaches the reCAPTCHA or the time limit. Once the episode is complete, the user will receive a feedback of the reCAPTCHA algorithm as any normal human user would. Results Researchers trained a Reinforce agent on a grid world of a specific size. The results presented in the paper are success rates across different 1000 runs. For the experiment to be successful, the agent would have to defeat the reCAPTCHA and obtain a score of 0.9. As per the results of the experiment, the discount factor achieved was 0.99, thereby, successfully defeating the reCAPTCHA. “Our proposed method achieves a success rate of 97.4% on a 100 × 100 grid and 96.7% on a 1000 × 1000 screen resolution”, states the researchers. For more information, check out the official research paper. Google researchers propose building service robots with reinforcement learning to help people with mobility impairment Facebook researchers show random methods without any training can outperform modern sentence embeddings models for sentence classification Researchers release unCaptcha2, a tool that uses Google’s speech-to-text API to bypass the reCAPTCHA audio challenge

In a paper published last month, Google developers explained the design, API, and implementation of TensorFlow.js, the JavaScript implementation of TensorFlow. TensorFlow.js was first introduced at the TensorFlow Dev Summit 2018. It is basically the successor of deeplearn.js, which was released in August 2017, and is now named as TensorFlow.js Core. Google’s motivation behind creating TensorFlow.js was to bring machine learning in the hands of web developers who generally do not have much experience with machine learning. It also aims at allowing experienced ML users and teaching enthusiasts to easily migrate their work to JS. The TensorFlow.js architecture TensorFlow.js, as the name suggests, is based on TensorFlow, with a few exceptions specific to the JS environment. This library comes with the following two sets of APIs: The Ops API facilitates lower-level linear algebra operations such as matrix, multiplication, tensor addition, and so on. The Layers API, similar to the Keras API, provide developers high-level model building blocks and best practices with emphasis on neural networks. Source: TensorFlow.js TensorFlow.js backends In order to support device-specific kernel implementations, TensorFlow.js has a concept of backends. Currently it supports three backends: the browser, WebGL, and Node.js. The two new rising web standards, WebAssembly and WebGPU, will also be supported as a backend by TensorFlow.js in the future. To utilize the GPU for fast parallelized computations, TensorFlow.js relies on WebGL, a cross-platform web standard that provides low-level 3D graphics APIs. Among the three TensorFlow.js backends, the WebGL backend has the highest complexity. With the introduction of Node.js and event-driven programming, the use of JS in server-side applications has grown over time. Server-side JS has full access to the filesystem, native operating system kernel, and existing C and C++ libraries. In order to support the server-side use cases of machine learning in JavaScript, TensorFlow.js comes with a Node.js backend that binds to the official TensorFlow C API using the N-API. As a fallback, TensorFlow.js provides a slower CPU implementation in plain JS. This fallback can run in any execution environment and is automatically used when the environment has no access to WebGL or the TensorFlow binary. Current applications of TensorFlow.js Since its launch, TensorFlow.js have seen its applications in various domains. Here are some of the interesting examples the paper lists: Gestural Interfaces TensorFlow.js is being used in applications that take gestural inputs with the help of webcam. Developers are using this library to build applications that translate sign language to speech translation, enable individuals with limited motor ability control a web browser with their face, and perform real-time facial recognition and pose-detection. Research dissemination The library has facilitated ML researchers to make their algorithms more accessible to others. For instance, the Magenta.js library, developed by the Magenta team, provides in-browser access to generative music models. Porting to the web with TensorFlow.js has increased the visibility of their work with their audience, namely musicians. Desktop and production applications In addition to web development, JavaScript has been used to develop desktop and production applications. Node Clinic, an open source performance profiling tool, recently integrated a TensorFlow.js model to separate CPU usage spikes by the user from those caused by Node.js internals. Another example is, Mood.gg Desktop, which is a desktop application powered by Electron, a popular JavaScript framework for writing cross-platform desktop apps. With the help of TensorFlow.js, Mood.gg detects which character the user is playing in the game called Overwatch, by looking at the user’s screen. It then plays a custom soundtrack from a music streaming site that matches with the playing style of the in-game character. Read the paper, Tensorflow.js: Machine Learning for the Web and Beyond, for more details. TensorFlow.js 0.11.1 releases! Emoji Scavenger Hunt showcases TensorFlow.js 16 JavaScript frameworks developers should learn in 2019

On Monday, a group of Amazon employees sent out an internal email to the We Won’t Build it mailing list, calling on Amazon to stop working with Palantir. Palantir is a data analytics company, founded by Peter Thiel, one of President Trump’s most vocal supporters in Silicon Valley, has a strong association with the Immigration and Customs Enforcement (ICE). https://twitter.com/WeWontBuildIt/status/1155872860742664194 Last year in June, an alliance of more than 500 Amazon employees had signed a petition addressing to CEO Jeff Bezos and AWS head Andy Jassy to abandon its contracts with government agencies. It seems that those protests are ramping up again. The email sent to employee mailing lists within Amazon Web Services demanded that Palantir to be removed from Amazon’s cloud for violating its terms of service. It also called on Amazon to take a stand against ICE by making a statement establishing its position against immigration raids, deportations and camps for migrants at the border. They have also demanded to stop selling its facial recognition tech to the government agencies. https://twitter.com/WeWontBuildIt/status/1155872862055485441 In May, Amazon shareholders had rejected the proposal to ban the sale of its facial recognition tech to government. With this they had also rejected eleven other proposals made by employees including a climate resolution, salary transparency and other issues. "The world is watching the abuses in ICE's concentration camps unfold. We know that our company should, and can do better,” the email read. The protests broke out at Amazon’s AWS Summit, held in New York, last week on Thursday. As Amazon CTO Werner Vogels gave a presentation, a group led by a man identified in a tweet as a tech worker interrupted to protest Amazon ties with ICE. https://twitter.com/altochulo/status/1149305189800775680 https://twitter.com/MaketheRoadNY/status/1149306940377448449 Vogels was caught off guard by the protests but continued on about the specifics of AWS, according to ZDNet. “I’m more than willing to have a conversation, but maybe they should let me finish first,” Vogels said amidst protesters, whose audio was cut off on Amazon’s official livestream of the event, per ZDNet. “We’ll all get our voices heard,” he said before returning to his planned speech. According to Business Insider reports, Palantir has a $51 million contract with ICE, which entails providing software to gather data on undocumented immigrant’s employment information, phone records, immigration history and similar information. Its software is hosted in the AWS cloud. The email states that Palantir enables ICE to violate the rights of others and working with such a company is harmful to Amazon’s reputation. The employees also state that their protest is in the spirit of similar actions at companies including Wayfair, Microsoft and Salesforce where workers have protested against their employers to cut ties with ICE and US Customs and Border Protection (CBP). Amazon has been facing increasing pressure from its employees. Last week workers had protested on Amazon Prime day demanding a safe working conditions and fair wages. Amazon, which typically takes a cursory view of such employee outcry, has so far given no indication that it will reconsider providing services to Palantir and other law enforcement agencies. Instead the company argued that the government should determine what constitutes “acceptable use” of technology of the type it sells. “As we’ve said many times and continue to believe strongly, companies and government organizations need to use existing and new technology responsibly and lawfully,” Amazon said to BuzzFeed News. “There is clearly a need for more clarity from governments on what is acceptable use of AI and ramifications for its misuse, and we’ve provided a proposed legislative framework for this. We remain eager for the government to provide this additional clarity and legislation, and will continue to offer our ideas and specific suggestions.” Other tech worker groups like Google Walkout For Real Change, Ban Google for Pride stand in solidarity with Amazon workers on this protest. https://twitter.com/GoogleWalkout/status/1155976287803998210 https://twitter.com/NoPrideForGoog/status/1155906615930806276 #TechWontBuildIt: Entropic maintainer calls for a ban on Palantir employees contributing to the project and asks other open source communities to take a stand on ethical grounds Amazon workers protest on its Prime day, demand a safe work environment and fair wages Amazon shareholders reject proposals to ban sale of facial recognition tech to govt and to conduct independent review of its human and civil rights impact

Earlier this week, the Bank of England’s chief economist, Andy Haldane, gave a warning that the UK needs a skills revolution as up to 15 million jobs in Britain are at stake. This is apparently due to a “third machine age” where Artificial Intelligence is making a huge number of jobs that were previously the preserve of humans outdated. Haldane says that this potential "Fourth Industrial Revolution" could cause disruptions on a "much greater scale" than the damage experienced during the first three Industrial Revolutions. This is because the first three industrial revolutions were mainly about machines replacing humans doing manual tasks.  But, the fourth Industrial revolution will be different. As Haldane told the BBC Radio 4’s Today programme, “the 20th-century machines have substituted not just for manual human tasks, but cognitive ones too -- human skills machines could reproduce, at lower cost, has both widened and deepened”. With robots becoming more intelligent, there will be deeper degrees of hollowing-out of jobs in this revolution than in the past. The Bank of England has classified jobs into three categories –jobs with a high (greater than 66%), medium (33-66%) and low (less than 33%) chances of automation. Administrative, clerical and production jobs are at the highest risk of getting replaced by Robots. Whereas, jobs focussing on human interaction, face-to-face conversation, and negotiation are less likely to suffer. Probability of automation by occupation This “hollowing out” poses risk not only for low-paid jobs but will also affect the mid-level jobs. Meanwhile, the UK’s Artificial Intelligence Council Chair, Tabitha Goldstaub, mentioned that the “challenge will be ensuring that people are prepared for the cultural and economic shifts” with focus on creating "the new jobs of the future" in order to avoid mass replacement by robots. Haldane echoed Goldstaub’s sentiments and told the BBC that “we will need even greater numbers of new jobs to be created in the future if we are not to suffer this longer-term feature called technological unemployment”. Every cloud has a silver lining Although the automation of these tasks can lead to mass unemployment, Goldstaub is positive. She says “there are great opportunities ahead as well as significant challenges”. Challenge being bracing the UK workforce for the coming change. Whereas, the silver lining, according to Goldstaub is that “there is a hopeful view -- that a lot of these jobs (existing) are boring, mundane, unsafe, drudgery - there could be -- liberation from -- these jobs and a move towards a brighter world.” OpenAI builds reinforcement learning based system giving robots human like dexterity OpenAI Five bots beat a team of former pros at Dota 2 What if robots get you a job! Enter Helena, the first artificial intelligence recruiter  

Yesterday, Facebook announced that Cadence, Esperanto, Intel, Marvell, and Qualcomm Technologies Inc, have committed to support their Glow compiler in future silicon products. Facebook, with this partnership aims to build a hardware ecosystem for machine learning. With Glow, their partners will be able to rapidly design and optimize new silicon products for AI and ML and help Facebook scale their platform. They are also planning to expand this ecosystem by adding more partners in 2018. What is Glow? Glow is a machine learning compiler which is used to speed up the performance of deep learning frameworks on different hardware platforms. The name “Glow” comes from Graph-Lowering, which is the main method that the compiler uses for generating efficient code. This compiler is designed to allow state-of-the-art compiler optimizations and code generation of neural network graphs. With Glow, hardware developers and researchers can focus on building next generation hardware accelerators that can be supported by deep learning frameworks like PyTorch. Hardware accelerators for ML solve a range of distinct problems. Some focus on inference, while others focus on training. How it works? Glow accepts a computation graph from deep learning frameworks such as, PyTorch and TensorFlow and generates highly optimized code for machine learning accelerators. To do so, it lowers the traditional neural network dataflow graph into a two-phase strongly-typed intermediate representation: Source: Facebook High-level intermediate representation allows the optimizer to perform domain-specific optimizations. Lower-level intermediate representation, an instruction-based address-only representation allows the compiler to perform memory-related optimizations, such as instruction scheduling, static memory allocation, and copy elimination. The optimizer then performs machine-specific code generation to take advantage of specialized hardware features. Glow supports a high number of input operators as well as a large number of hardware targets with the help of its lowering phase, which eliminates the need to implement all operators on all targets. The lowering phase reduces the input space and allows new hardware backends to focus on a small number of linear algebra primitives. You can read more about Facebook’s goals for Glow in its official announcement. If you are interesting in knowing how it works in more detail, check out this research paper and also its GitHub repository. Facebook launches LogDevice: An open source distributed data store designed for logs Google’s new What-if tool to analyze Machine Learning models and assess fairness without any coding Facebook introduces Rosetta, a scalable OCR system that understands text on images using Faster-RCNN and CNN

Machine Learning was one of the most talked about topic at the Amazon’s re:invent this year. In order to make machine learning models accessible to everyday users, regardless of their expertise level, Amazon Web services launched an end-to-end machine learning service – Sagemaker. Amazon Sagemaker allows data scientists, developers, and machine learning experts to quickly build, train, and deploy machine learning models at scale. The below image shows the process adopted by Sagemaker to aid developers in building ML models. Source: aws.amazon.com Model Building Amazon SageMaker makes it easy to build ML models by easy training and selection of best algorithms and frameworks for a particular model. Amazon Sagemaker has zero-setup hosted Jupyter notebooks which makes it easy to explore, connect, and visualize the training data stored on Amazon S3. These notebook IDEs are runnable on either general instance types or GPU powered instances. Model Training ML models can be trained by a single click in the Amazon SageMaker console. For training the data, Sagemaker also has a provision for moving training data from Amazon RDS, Amazon DynamoDB, and Amazon Redshift into S3. Amazon Sagemaker is preconfigured to run TensorFlow and Apache MXNet. However, developers can use their own frameworks and also create their own training with Docker containers. Model Tuning and Hosting Amazon Sagemaker has a model hosting service with HTTPs endpoints. These endpoints can invoke real-time inferences, support traffic, and simultaneously allow A/B Testing. Amazon Sagemaker can automatically tune models to achieve high accuracy. This makes the training process faster and easier. Sagemaker can automate the underlying infrastructure and allows developers to easily scale to train models at petabyte scale. Model Deployment After training and tuning come the deployment phase. Sagemaker deploys the models on an auto-scaling cluster of Amazon EC2 instances, for running predictions on new data. These high-performance instances are spread across multiple availability zones. According to the official product page, Amazon Sagemaker has multiple use cases. One of them being Ad targeting, where Amazon Sagemaker can be used with other AWS services to help build, train, and deploy ML models for targeting online ads, optimize return on ad spend, customer segmentation, etc. Another interesting use case of Sagemaker is how it can train recommender systems within its serverless, distributed environment which can be hosted easily in low-latency, auto-scaling endpoint systems. Sagemaker can also be used for building highly efficient Industrial IoT and ML models to predict machine failure or for maintenance scheduling. As of now, Amazon Sagemaker is free for developers for the first two months. Each month developers are provided with 250 hours of t2.medium notebook usage, 50 hours of m4.xlarge usage for training, and 125 hours of m4.xlarge usage for hosting. After the free period, the pricing would vary by region and customers would be billed per-second for instance usage, per-GB of storage, and per-GB of Data transfer into and out of the service. AWS Sagemaker provides an end-to-end solution for the development of machine learning applications. The ease and flexibility offered by AWS Sagemaker could be harnessed by developers to solve several business-related problems.

The Skymind team has announced a milestone release of Eclipse Deeplearning4j (DL4J), an open-source library for deep learning. DL4J 1.0.0-alpha has some breakthrough changes which will ease development of deep learning applications using Java and Scala. From a developer’s perspective, the roadmap provides an exciting opportunity to perform complex numerical computations with the major updates done to each module of Deeplearning4j. DL4J is a distributed neural network library in Java and Scala which allows distributed training on Hadoop and Spark. It provides powerful data processing that enables efficient use of CPUs and GPUs. With new features, bug fixes and optimizations in the toolkit, Deeplearning4j provides excellent capabilities to perform advanced deep learning tasks. Here are some of the significant changes available in DL4J 1.0.0-alpha: Deeplearning4j: New changes made to the framework Enhanced and new layers added to the DL4J toolkit. Lots of new API changes to optimize the training, building, and deploying neural network models in the production environment. A considerable amount of bug fixes and optimizations are done to the DL4J toolkit. Keras 2 import support Now you can import Keras 2 models into DL4J, while still keeping backward compatibility for Keras 1. The older module DL4J-keras and Model API in DL4J version 0.9.1 is removed. In order to import Keras models, the only entry point you can use is KerasModelImport.   Refer DL4J-Keras import support to know more about the complete list of updates. ND4J: New features A powerful library used for scientific and numerical computing for the JVM: Hundreds of new operations and features added to ease scientific computing, an essential building block for deep learning tasks. Added NVIDIA CUDA support for 9.0/9.1. They are continuing support for CUDA 8.0, however dropping support for CUDA 7.5. New API changes are done to the ND4J library. ND4J: SameDiff There is a new Alpha release of SameDiff, which is an auto-differentiation engine for ND4J. It supports two execution modes for serialized graphs: Java-driven execution, and Native execution. It also supports import of TensorFlow and ONNX graphs for inference purposes. You can know all the other new features at SameDiff release notes. DataVec: New features An effective ETL library for getting data into the pipeline, so neural networks can understand: Added new features and bug fixes to perform efficient and powerful ETL processes. New API changes incorporated in the DataVec library. Arbiter: New features A package for efficient optimization of neural networks to obtain good performance: New Workspace support added to carry out hyperparameter optimization of machine learning models. New layers and API changes have been done to the tool. Bug fixes and improvements for optimized tuning performances. A complete list of changes is available on Arbiter release notes. RL4J: New features A reinforcement learning framework integrated with deeplearning4j for the JVM: Added support for LSTM layers to asynchronous advantage actor-critic (A3C) models. Now you can use the latest version of VizDoom since MDP for Doom is updated. Lots of fixes and improvements implemented in the RL4J framework. ScalNet A scala wrapper for DL4J resembling a Keras like API for deep learning: New ScalNet Scala API is released which is very much similar to Keras API. It supports Keras based sequential models. The project module closely resembles both DL4J model-import module and Keras. Refer ScalNet release notes, if you like to know more. ND4S: N-Dimensional Arrays for Scala An open-source Scala bindings for ND4J: ND4S now has Scala 2.12 support Possible issues with the DL4J 1.0.0-alpha release Since this is an alpha release, you may encounter performance related issues and other possible issues (when compared to DL4J version 0.9.1). This will be addressed and rectified in the next release. Support for training a Keras model in DL4J is still very limited. This issue will be handled in the next release. To know more, you can refer Keras import bug report. Major new operations added in ND4J still do not use GPU yet. The same applies to the new auto-differentiation engine for ND4J. We can expect more improvements and new features on DL4J 1.0.0 roadmap. For the full list of updates, you can refer the release notes.  Check out other popular posts: Top 6 Java Machine Learning/Deep Learning frameworks you can’t miss Top 10 Deep learning frameworks    

Until recently, the first thing that came to our mind with Amazon Web Services was that of an infrastructure provider. But things are changing, rightly so in tune with times. The AWS is now into an all out mode to scale up the artificial intelligence ladder, gradually shifting focus towards machine learning, deep learning and data science. Last week it went serverless, and now the cloud leader has added yet another function to its repertoire: AWS IoT Analytics. AWS IoT Analytics provides advanced data analysis of data collected from your IoT devices. It is a fully managed service of AWS IoT, which can be used to cleanse, process, enrich, store, and analyze IoT data at scale. Amazon calls it “the easiest way to run analytics on IoT data.” Announced closely on the heels of re:Invent 2017, the AWS IoT Analytics has been designed specifically for common IoT use cases like predictive maintenance, asset usage patterns, and failure profiling. The platform captures data from devices connected to AWS IoT Core, and filters, transforms, and enriches it before storing it in a time-series database for analysis. “You can set up the service to collect only the data you need from your devices, apply mathematical transforms to process the data, and enrich the data with device-specific metadata such as device type and location before storing the processed data. Then, you can use IoT Analytics to run ad hoc queries using the built-in SQL query engine, or perform more complex processing and analytics like statistical inference and time series analysis,” Amazon said in its release. The new service feature integrates with Amazon Quicksight for visualization of your data and brings the power of machine learning through integration with Jupyter Notebooks. Benefits of AWS IoT Analytics Helps with predictive analysis of data by providing access to pre-built analytical functions Provides ability to visualize analytical output from service Provides tools to clean up data Can help identify patterns in the gathered data Getting Started: Common IoT Analytics Concepts Channel: archives the raw, unprocessed messages and collects data from MQTT topics. Pipeline: consumes messages from channels and allows message processing. Activities: perform transformations on your messages including filtering attributes and invoking lambda functions advanced processing. Data Store: Used as a queryable repository for processed messages. Provide ability to have multiple datastores for messages coming from different devices or locations or filtered by message attributes. Data Set: Data retrieval view from a data store, can be generated by a recurring schedule. This is how it looks like Source: aws.amazon.com First, you create a channel to receive incoming messages. For this, select the Channels menu option and click the Create a channel button (as shown above). It creates a new form where you have to name your channel and give the channel a MQTT topic filter, from which this channel will ingest messages. Your channel is then created once you click the Create Channel button. Once your Channel is created, set up a Data Store to receive and store the messages received on the Channel from your IoT device. Multiple Data Stores can be created for complex solutions. Now that you have your Channel and Data Store stored, connect the two using a Pipeline (in manner something similar to how we created a Channel and Data Store) for the processing and transformation of messages. Additional attributes can be added to create a more robust pipeline, if need be. To use AWS IoT Analytics, all we need now is an IoT rule that sends data to a channel.  Choosing the Analyze menu option will bring up the screens to Create a data set. And this is how you set up advanced data analytics for AWS IoT: Source: aws.amazon.com In addition to the ability to collect, visualize, process, query and store large amounts of data generated from AWS IoT connected devices, Amazon said the AWS IoT Analytics service can be used in so many other possibilities such as the AWS Command Line Interface (AWS CLI), the AWS IoT API, language-specific AWS SDKs, and AWS IoT Device SDKs. To learn more about AWS IoT Analytics and to register for the preview, visit the product page.

Xilinx Inc., has reportedly won orders from Microsoft Corp.’s Azure cloud unit to account for half of the co-processors currently used on Azure servers to handle machine-learning workloads. This will replace the chips made by Intel Corp, according to people familiar with Microsoft's’ plans as reported by Bloomberg Microsoft’s decision comes with effect to add another chip supplier in order to serve more customers interested in machine learning. To date, this domain was run by Intel’s Altera division. Now that Xilinx has bagged the deal, does this mean Intel will no longer serve Microsoft? Bloomberg reported Microsoft’s confirmation that it will continue its relationship with Intel in its current offerings. A Microsoft spokesperson added that “There has been no change of sourcing for existing infrastructure and offerings”. Sources familiar with the arrangement also commented on how Xilinx chips will have to achieve performance goals to determine the scope of their deployment. Cloud vendors these days are investing heavily in research and development centering around the machine learning field. The past few years has seen an increase in need of flexible chips that can be configured to run machine-learning services. Companies like Microsoft, Google and Amazon are massive buyers of server chips and are always looking for alternatives to standard processors to increase the efficiency of their data centres. Holger Mueller, an analyst with Constellation Research Inc., told SiliconANGLE that “Programmable chips are key to the success of infrastructure-as-a-service providers as they allow them to utilize existing CPU capacity better, They’re also key enablers for next-generation application technologies like machine learning and artificial intelligence.” Earlier this year, Xilinx CEO Victor Peng made clear his plans to focus on data center customers, saying “data center is an area of rapid technology adoption where customers can quickly take advantage of the orders of magnitude performance and performance per-watt improvement that Xilinx technology enables in applications like artificial intelligence (AI) inference, video and image processing, and genomics” Last month, Xilinx made headlines with the announcement of a new breed of computer chips designed specifically for AI inference. These chips combine FPGAs with two higher-performance Arm processors, plus a dedicated AI compute engine and relates to the application of deep learning models in consumer and cloud environments. The chips promise higher throughput, lower latency and greater power efficiency than existing hardware. Looks like Xilinx is taking noticeable steps to make itself seen in the AI market. Head over to Bloomberg for the complete coverage of this news. Microsoft Ignite 2018: New Azure announcements you need to know Azure DevOps outage root cause analysis starring greedy threads and rogue scale units Microsoft invests in Grab; together aim to conquer the Southeast Asian on-demand services market with Azure’s Intelligent Cloud    

After the news of Amazon employees listening to your Echo audio recordings, now we have the non-shocker report of Google employees doing the same. The news was reported by Belgian public broadcaster, VRT NWS on Wednesday. Addressing this news, Google accepted in yesterday’s blog post that it does this to make its AI assistant smarter to understand user commands regardless of what their language is. In its privacy policies, the tech giant states, “Google collects data that's meant to make our services faster, smarter, more relevant, and more useful to you. Google Home learns over time to provide better and more personalized suggestions and answers.” Its privacy policies also have a mention that it shares information with its affiliates and other trusted businesses. What it does not explicitly say is that these recordings are shared with its employees too. Google hires language experts to transcribe audio clips recorded by Google’s AI assistant who can end up listening to sensitive information about users. Whenever you make a request to Google Home smart speaker or any other smart speaker for that matter, your speech is recorded. These audio recordings are sent to the servers of the companies that they use to train their speech recognition and natural language understanding systems. A small subset of these recordings, 0.2% in the case of Google, are sent to language experts around the globe who transcribe them as accurately as possible. Their work is not about analyzing what the user is saying, but, in fact, how they are saying it. This helps Google’s AI assistant to understand the nuances and accents of a particular language. The problem is these recordings often contain sensitive data. Google in the blog post claims that these audio snippets are analyzed in an anonymous fashion, which means that reviewers will not be able to identify the user they are listening to. “Audio snippets are not associated with user accounts as part of the review process, and reviewers are directed not to transcribe background conversations or other noises, and only to transcribe snippets that are directed to Google,” the tech giant said. Countering this claim, VRT NWS was able to identify people through personal addresses and other sensitive information in the recordings. “This is undeniably my own voice,” said one man. Another family was able to recognize the voice of their son and grandson in the recording. What is worse is that sometimes these smart speakers record the audio clips entirely by accident. Despite the companies claiming that these devices only start recording when they hear their “wake words” like “Okay Google”, there are many reports showing the devices often start recording by mistake. Out of the thousand or so recordings reviewed by VRT NWS, 153 were captured accidentally. Google in the blog post mentioned that it applies “a wide range of safeguards to protect user privacy throughout the entire review process.” It further accepted that these safeguards failed in the case of the Belgian contract worker who shared the audio recordings to VRT NWS, violating the company’s data security and privacy rules in the process. “We just learned that one of these language reviewers has violated our data security policies by leaking confidential Dutch audio data. Our Security and Privacy Response teams have been activated on this issue, are investigating, and we will take action. We are conducting a full review of our safeguards in this space to prevent misconduct like this from happening again,” the tech giant wrote. Companies not being upfront about the transcription process can cause legal trouble for them. Michael Veale, a technology privacy researcher at the Alan Turing Institute in London, told Wired that this practice of sharing personal information of users might not meet the standards set by the EU’s GDPR regulations. “You have to be very specific on what you’re implementing and how. I think Google hasn’t done that because it would look creepy,” he said. Read the entire story on VRT NWS’s official website. You can watch the full report on YouTube. https://youtu.be/x8M4q-KqLuo Amazon’s partnership with NHS to make Alexa offer medical advice raises privacy concerns and public backlash Amazon is being sued for recording children’s voices through Alexa without consent Amazon Alexa is HIPAA-compliant: bigger leap in the health care sector

StockX, an online marketplace for buying and selling sneakers, suffered a major data breach in May impacting 6.8 million customers. Records leaked included names, email addresses and hashed passwords. The full scale of this data breach came to light after an unnamed data breached seller contacted TechCrunch claiming information about the attack. Tech crunch then verified the claims by contacting people from a sample of 1,000 records using the information only they would know. StockX released a statement yesterday acknowledging that a data breach had indeed occurred. StockX says they were made aware of the breach on July 26 and immediately launched a forensic investigation and engaged experienced third-party data experts to assist. On getting evidence to suggest customer data may have been accessed by an unknown third party, they sent customers an email on August 3 to make them aware of the incident. This email surprisingly asked customers to reset their passwords citing system updates but said nothing about the data breach leaving users confused on what caused the alleged system update or why there was no prior warning. Later the same day, StockX confirmed that they had discovered a data security issue and confirmed that an unknown third-party was able to gain access to certain customer data, including customer name, email address, shipping address, username, hashed passwords, and purchase history. The hashes were encrypted using MD5 with salts. According to weleakinfo, this is a very weak hashing algorithm; at least 90% of all hashes can be cracked successfully. Users were infuriated that instead of being honest, StockX simply sent their customers an email asking them to reset their passwords. https://twitter.com/Asaud_7/status/1157843000170561536 https://twitter.com/kustoo/status/1157735133157314561 https://twitter.com/RunWithChappy/status/1157851839754383360 StockX released a system-wide security update, a full password reset of all customer passwords with an email to customers alerting them about resetting their passwords, a high-frequency credential rotation on all servers and devices and a lockdown of their cloud computing perimeter. However, they were a little too late in their ‘ongoing investigation’ as they mention on their blog. Techcrunch revealed that the seller had put the data for sale for $300 in a dark web listing and one person had already bought the data. StockX is also subject to EU’s General Data Protection Regulation considering it has a global customer base and can be potentially fined for the incident. https://twitter.com/ComplexSneakers/status/1157754866460221442 According to FTC, StockX is also not compliant with the US laws regarding a data breach. https://twitter.com/zruss/status/1157785830200619008 Following Capital One data breach, GitHub gets sued and AWS security questioned by a US Senator. British Airways set to face a record-breaking fine of £183m by the ICO over customer data breach. U.S. Senator introduces a bill that levies jail time and hefty fines for companies violating data breaches.

Say hello to IBM RXN, a free AI Tool in IBM Cloud for predicting chemical reactions Earlier this week, IBM launched an AI tool called IBM RXN in IBM cloud at the American Chemistry Society, Boston, for predicting chemical reactions in just seconds.  IBM RXN is an advanced AI model which is useful in daily research activities and experiments. IBM Research IBM presented a web-based app last year at the NIPS 2017 Conference, which is capable of relating organic chemistry to a language. It also applies state-of-the-art neural machine translation methods which take care of converting designing materials to generating products leveraging sequence-to-sequence (seq2seq) models. IBM RXN for Chemistry uses a system known as a simplified molecular-input line-entry system or SMILES. SMILES is used to represent a molecule as a sequence of characters. The model was trained using a combination of reaction datasets, equivalent to a total of 2 million reactions. SMILES in IBM RXN IBM RXN comprises of features such as Ketcher editor, pre-configured libraries, and challenge mode. Ketcher is a web-based chemical structure editor which is designed for chemists, lab scientists, and technicians. It involves selecting, modifying, and erasing the connected, and unconnected atom bonds with the help of a selection tool or shift key. There’s a cleanup tool which checks bond lengths, angles and spatial arrangement of atoms. It is also capable of checking the stereochemistry and structure layout with its advanced features. It is a simple data-driven tool which is trained without querying a database or any additional external information. Additionally, users can build projects and share them with friends or colleagues. There are Pre Configured libraries of molecules which enable adding reactants and reagents to your Ketcher board in just a few clicks. It also provides access to the most common molecules in organic chemistry via the installation of a library to your molecule set. You can also upload molecules to customize libraries. Enhancing the libraries with your own reaction outcomes or with molecules drawn on the Ketcher board is also possible. Finally, there is a challenge mode which puts your Organic Chemistry knowledge to test and helps with Organic grade preparation for class exams. IBM RXN is a completely free tool and available in the IBM cloud. For more information, check out the official IBM blog post. IBM’s DeepLocker: The Artificial Intelligence powered sneaky new breed of Malware Four IBM facial recognition patents in 2018, we found intriguing IBM unveils world’s fastest supercomputer with AI capabilities, Summit

Yesterday, researchers at Facebook introduced a machine learning system named, Rosetta for scalable optical character recognition (OCR). This model extracts text from more than a billion public Facebook and Instagram images and video frames. Then, this extracted text is fed into a text recognition model that has been trained on classifiers, which helps it understand the context of the text and the image together. Why Rosetta is introduced? Rosetta will help in the following scenarios: Provide a better user experience by giving users more relevant photo search results. Make Facebook more accessible for the visually impaired by incorporating the texts into screen readers. Help Facebook proactively identify inappropriate or harmful content. Help to improve the accuracy of classification of photos in News Feed to surface more personalized content. How it works? Rosetta consists of the following text extraction model: Source: Facebook Text extraction on an image is done in the following two steps: Text detection In this step, rectangular regions that potentially contain the text are detected. It performs text detection based on Faster R-CNN, a state-of-the-art object detection network. It uses Faster R-CNN but replaces ResNet convolutional body with a ShuffleNet-based architecture for efficiency reasons. The anchors in regional proposal network (RPN) are also modified to generate wider proposals, as text words are typically wider than the objects for which the RPN was designed. The whole detection system is trained jointly in a supervised, end-to-end manner. The model is bootstrapped with an in-house synthetic data set and then fine-tuned with human-annotated data sets so that it learns real-world characteristics. It is trained using the recently open-sourced Detectron framework powered by Caffe2. Text recognition The following image shows the architecture of the text recognition model: Source: Facebook In the second step, for each of the detected regions a convolutional neural network (CNN) is used to recognize and transcribe the word in the region. This model uses CNN based on the ResNet18 architecture, as this architecture is more accurate and computationally efficient. For training the model, finding what the text in an image says is considered as a sequence prediction problem. They input images containing the text to be recognized and the output generated is the sequence of characters in the word image. Treating the model as one of sequence prediction allows the system to recognize words of arbitrary length and to recognize the words that weren’t seen during training. This two-step model provides several benefits, including decoupling the training process of detection and recognition models, recognition of words in parallel, and independently supporting text recognition for different languages. Rosetta has been widely adopted by various products and teams within Facebook and Instagram. It offers a cloud API for text extraction from images and processes a large volume of images uploaded to Facebook every day. In future, the team is planning to extend this system to extract text from videos more efficiently and also support a wide number of languages used on Facebook. To get a more in-depth idea of how Rosetta works, check out the researchers’ post at Facebook code blog and also read this paper: Rosetta: Large Scale System for Text Detection and Recognition in Images. Why learn machine learning as a non-techie? Is the machine learning process similar to how humans learn? Facebook launches a 6-part Machine Learning video series

Yesterday, many parts of the Internet faced an unprecedented outage as Verizon, the popular Internet transit provider accidentally rerouted IP packages after it wrongly accepted a network misconfiguration from a small ISP in Pennsylvania, USA. According to The Register, “systems around the planet were automatically updated, and connections destined for Facebook, Cloudflare, and others, ended up going through DQE and Allegheny, which buckled under the strain, causing traffic to disappear into a black hole”. According to Cloudflare, “What exacerbated the problem today was the involvement of a “BGP Optimizer” product from Noction. This product has a feature that splits up received IP prefixes into smaller, contributing parts (called more-specifics). For example, our own IPv4 route 104.20.0.0/20 was turned into 104.20.0.0/21 and 104.20.8.0/21”. Many Google users were unable to access the web using the Google browser. Some users say the Google Calendar went down too. Amazon users were also unable to use some services such as Amazon books, as users were unable to reach the site. Source: Downdetector Source:Downdetector Source:Downdetector Also, in another incident, on June 6, more than 70,000 BGP routes were leaked from Swiss colocation company Safe Host to China Telecom in Frankfurt, Germany, which then announced them on the global internet. “This resulted in a massive rerouting of internet traffic via China Telecom systems in Europe, disrupting connectivity for netizens: a lot of data that should have gone to European cellular networks was instead piped to China Telecom-controlled boxes”, The Register reports. BGP caused a lot of blunder in this outage The Internet is made up of networks called Autonomous Systems (AS), and each of these networks has a unique identifier, called an AS number. All these networks are interconnected using a  Border Gateway Protocol (BGP), which joins these networks together and enables traffic to travel from an ISP to a popular website at a far off location, for example. Source: Cloudflare With the help of BGP, networks exchange route information that can either be specific, similar to finding a specific city on your GPS, or very general, like pointing your GPS to a state. DQE Communications with an AS number AS33154, an Internet Service Provider in Pennsylvania was using a BGP optimizer in their network. It announced these specific routes to its customer, Allegheny Technologies Inc (AS396531), a steel company based in Pittsburgh. This entire routing information was sent to Verizon (AS701), who further accepted and passed this information to the world. “Verizon’s lack of filtering turned this into a major incident that affected many Internet services”, Cloudfare mentions. “What this means is that suddenly Verizon, Allegheny, and DQE had to deal with a stampede of Internet users trying to access those services through their network. None of these networks were suitably equipped to deal with this drastic increase in traffic, causing disruption in service” Job Snijders, an internet architect for NTT Communications, wrote in a network operators' mailing list, “While it is easy to point at the alleged BGP optimizer as the root cause, I do think we now have observed a cascading catastrophic failure both in process and technologies.” https://twitter.com/bgpmon/status/1143149817473847296 Cloudflare's CTO Graham-Cumming told El Reg's Richard Speed, "A customer of Verizon in the US started announcing essentially that a very large amount of the internet belonged to them. For reasons that are a bit hard to understand, Verizon decided to pass that on to the rest of the world." "but normally [a large ISP like Verizon] would filter it out if some small provider said they own the internet", he further added. “If Verizon had used RPKI, they would have seen that the advertised routes were not valid, and the routes could have been automatically dropped by the router”, Cloudflare said. https://twitter.com/eastdakota/status/1143182575680143361 https://twitter.com/atoonk/status/1143139749915320321 Rerouting is highly dangerous as criminals, hackers, or government-spies could be lurking around to grab such a free flow of data. However, this creates security distension among users as their data can be used for surveillance, disruption, and financial theft. Cloudflare was majorly affected by this outage, “It is unfortunate that while we tried both e-mail and phone calls to reach out to Verizon, at the time of writing this article (over 8 hours after the incident), we have not heard back from them, nor are we aware of them taking action to resolve the issue”, the company said in their blogpost. One of the users commented, “BGP needs a SERIOUS revamp with Security 101 in mind.....RPKI + ROA's is 100% needed and the ISPs need to stop being CHEAP. Either build it by Federal Requirement, at least in the Nation States that take their internet traffic as Citizen private data or do it as Internet 3.0 cause 2.0 flaked! Either way, "Path Validation" is another component of BGP that should be looked at but honestly, that is going to slow path selection down and to instrument it at a scale where the internet would benefit = not worth it and won't happen. SMH largest internet GAP = BGP "accidental" hijacks” Verizon in a statement to The Register said, "There was an intermittent disruption in internet service for some [Verizon] FiOS customers earlier this morning. Our engineers resolved the issue around 9 am ET." https://twitter.com/atoonk/status/1143145626516914176 To know more about this news in detail head over to CloudFlare’s blog. OpenSSH code gets an update to protect against side-channel attacks Red Badger Tech Director Viktor Charypar talks monorepos, lifelong learning, and the challenges facing open source software [Interview] Facebook signs on more than a dozen backers for its GlobalCoin cryptocurrency including Visa, Mastercard, PayPal and Uber

Popular Quantum Computing Canadian startup D-Wave Systems Inc. launched a free, and real-time online Quantum Application Environment (QAE), called Leap, yesterday. What makes Leap so unique is the fact that it virtualizes quantum computing for almost anyone who has a computer and a broadband connection. Leap is the first cloud-based QAE that offers real-time access to a live quantum computer. It comes with open-source development tools, interactive demos, coding examples, educational resources, and knowledge base articles. Leap is designed for developers, researchers, and businesses. Its online community enables collaboration, thereby, helping Leap users write and run quantum applications. This accelerates the development of real-world applications. Major features of Leap Leap QAE provides free access to a D-Wave 2000Q quantum computer which lets you submit and run applications, helping receive solutions in seconds. It comes with an open-source Ocean software development kit (SDK). This comprises built-in templates for algorithms, along with an ability to develop new code using a familiar programming language, Python. Leap also provides Hands-on coding option. This consists of interactive examples in the form of Jupyter notebooks with live code, equations, visualizations, and narrative text to jumpstart the process of quantum application development. Apart from that, it offers learning resources, which includes comprehensive live demos and educational resources. This helps developers in writing applications for a quantum computer, quickly. It also offers Community support which includes community and technical forums enabling easy developer collaboration. Leap is an outcome of D-Wave’s continuous efforts to drive the real-world quantum application development forward. D‑Wave customers have built 100 early applications so far for problems such as airline schedules, election modeling, quantum chemistry simulation, automotive design, logistics, and more. A lot of them have built software tools that make it easy to develop new applications. These existing applications and tools, along with an access to a growing community, provide developers with a wealth of examples to learn from and build upon. “Our job is to sift through the sands of data to find the gold—information that will help our manufacturing customers increase equipment efficiency and reduce defects. With D‑Wave Leap, we are showing we can solve computationally difficult problems today, while also learning and preparing for new approaches to AI and machine learning that quantum computing will allow,” said Abhi Rampal, CEO of Solid State AI. Apart from D-Wave, Rigetti Computing, a California-based developer of quantum integrated circuits also launched Quantum Cloud Services last month, to bring together the best of classical and quantum computing on a single cloud platform. For more information, check out the official D-Wave blog post. Did quantum computing just take a quantum leap? A two-qubit chip by UK researchers makes controlled quantum entanglements possible Quantum Computing is poised to take a quantum leap with industries and governments on its side “The future is quantum” — Are you excited to write your first quantum computing code using Microsoft’s Q#?