Tech News - Data

Waymo, a former Google self-driving car project, announced yesterday, that it is making one of its 3D LIDAR (light detection and ranging) sensors, called Laser Bear Honeycomb, available to select partners. “Offering this LIDAR to partners helps spur the growth of applications outside of self-driving cars and also propels our business forward”, states the Waymo team on Medium. Waymo started developing its own set of sensors in 2011, including three different types of LIDARs . LIDAR refers to a remote sensing method that can track the distance using pulses of laser light. Waymo uses a medium-range LIDAR, located on top of the car. They have also developed a short-range and a long-range LIDAR. Waymo team states that these custom LIDARs are the ones that enabled Waymo to put its self-driving cars on road. Now, Waymo is set on expanding the realm of these sensors outside of self-driving, by including Robotics, security, agricultural technology, etc. Waymo team states that their Laser Bear Honeycomb is a best-in-class perimeter sensor. It’s the same short-range sensor that is used around the bumper of Waymo’s self-driving vehicles. Key features of Laser Bear Honeycomb LIDAR The Laser Bear Honeycomb LIDAR by Waymo comes with an outstanding set of features including a wider field of view, multiple returns per pulse, and minimum zero range. Wide field of view Most of the 3D LIDARs come with a vertical field of view (FOV) of just 30°. But, the Laser Bear Honeycomb LIDAR comes with a vertical FOV of 95°, along with a 360° horizontal FOV. What this means is that one Honeycomb is capable of performing the job of three other 3D sensors. Multiple returns per pulse On sending out a pulse of light, Laser Bear Honeycomb can see up to four different objects in the laser beams’ line of sight. For instance, it can spot the foliage in front of a tree branch as well as the tree branch itself, giving a more detailed view of the environment in turn. It can also uncover the objects that might otherwise get missed out. Minimum range of zero Laser Bear Honeycomb comes with a minimum range of zero. This means it can immediately track the objects that are in front of the sensor. It also comes with other capabilities such as near object detection and avoidance. For more information, check out the official Waymo blog post. Alphabet’s Waymo to launch the world’s first commercial self driving cars next month Anthony Levandowski announces Pronto AI and makes a coast-to-coast self-driving trip Aurora, a self-driving startup, secures $530 million in funding from Amazon, Sequoia, and T.Rowe Price among others

FaunaDB, a distributed OLTP (online transaction processing) database, released the official results of the test done on FaunaDB version 2.5.4 and 2.6.0 by Jepsen, an independent testing organization, yesterday. FaundDB passed the tests with flying colors, and is now architecturally sound, correctly implemented, and ready to take on the enterprise workloads over cloud. The Fauna team had been working on the FaunaDB tests extensively with Kyle Kingsbury, a computer safety researcher at Jepsen, for three months. “Our mandate for him was not merely to test the basic properties of the system, but rather to poke into the dark corners and exhaustively validate..FaunaDB”, states the Fauna team. Jepsen team states that Fauna had written their own Jepsen tests, which have been refined and expanded throughout the collaboration between Jepsen and Fauna. Jepsen evaluated FaunaDB 2.5.4 and 2.5.5, along with several development builds up to 2.6.0-rc10. Jepsen team used three replicas, along with 5-10 nodes striped evenly across replicas for the tests. Additionally, the log node topologies in 2.5.4 and 2.5.5 were explicitly partitioned, with a copy in each of the replicas. The Jepsen team waited for data movement to complete as well as for all indices to signal readiness before initiating the testing process. Fauna team states that FaunaDB’s core operations on single instances in 2.5.5 appeared quite “solid”. During the tests, the Fauna team reliably managed to create, read, update, and delete the records transactionally atsnapshot, serializable, and strict serializable isolation. Also, the acknowledged instance updates were never lost. FaundaDB also managed to pass additional tests, while covering features such as indexes and temporality. By the release of FaunaDB 2.6.0-rc10, Fauna managed to address all the issues identified by Jepsen. However, progress is still needed around some minor work and schema changes. Other than that, FaunaDB also provides the “highest possible level of correctness”. FaunaDB team is currently planning on promoting SI or serializable transactions to strict serializability which is the gold standard for concurrent systems. Another noticeable fact about FaunaDB is that it is self-operating. FaunaDB has been especially designed to offer support for online addition and removal of nodes with appropriate backpressure. Also, it is architecturally sound. FaunaDB combines Calvin’s cross-shard transactional protocol with the Raft’s consensus system for individual shards. Finally, the Jepsen team states that the bugs found in FaunaDB are implementation problems, and Fauna will be working on fixing the detected bugs as soon as possible. “FaunaDB’s approach is fundamentally sound...Calvin-based systems like FaunaDB could play an important future role in the distributed database landscape”, states the Jepsen team. For more information, check out the official Jepsen results post. MariaDB CEO says big proprietary cloud vendors “strip-mining open-source technologies and companies” Red Hat Satellite to drop MongoDB and will support only PostgreSQL backend Uber releases AresDB, a new GPU-powered real-time Analytics Engine

Google AI research team announced that it’s open sourcing GPipe, a distributed machine learning library for efficiently training Large-scale Deep Neural Network Models, under the Lingvo Framework, yesterday. GPipe makes use of synchronous stochastic gradient descent and pipeline parallelism for training. It divides the network layers across accelerators and pipelines execution to achieve high hardware utilization. GPipe also allows researchers to easily deploy accelerators to train larger models and to scale the performance without tuning hyperparameters. Google AI researchers had also published a paper titled “GPipe: Efficient Training of Giant Neural Networks using Pipeline Parallelism" last year in December. In the paper, researchers demonstrated the use of pipeline parallelism to scale up deep neural networks to overcome the memory limitation on current accelerators. Let’s have a look at major highlights of GPipe. GPipe helps with maximizing the memory and efficiency GPipe helps with maximizing the memory allocation for model parameters. Researchers conducted experiments on Cloud TPUv2s. Each of these Cloud TPUv2s consists of 8 accelerator cores and 64 GB memory (8 GB per accelerator). Generally, without GPipe, a single accelerator is able to train up to 82 million model parameters because of the memory limitations, however, GPipe was able to bring down the immediate activation memory from 6.26 GB to 3.46GB on a single accelerator. Researchers also measured the effects of GPipe on the model throughput of AmoebaNet-D to test its efficiency. Researchers found out that there was almost a linear speedup in training. GPipe also enabled 8 billion parameter Transformer language models on 1024-token sentences using speedup of 11x.                                        Speedup of AmoebaNet-D using GPipe Putting the accuracy of GPipe to test Researchers used GPipe to verify the hypothesis that scaling up existing neural networks can help achieve better model quality. For this experiment, an AmoebaNet-B with 557 million model parameters and input image size of 480 x 480  was trained on the ImageNet ILSVRC-2012 dataset. Researchers observed that the model was able to reach 84.3% top-1 / 97% top-5 single-crop validation accuracy without the use of any external data. Researchers also ran the transfer learning experiments on the CIFAR10 and CIFAR100 datasets, where they observed that the giant models improved the best published CIFAR-10 accuracy to 99% and CIFAR-100 accuracy to 91.3%. “We are happy to provide GPipe to the broader research community and hope it is a useful infrastructure for efficient training of large-scale DNNs”, say the researchers. For more information, check out the official GPipe Blog post. Google researchers propose building service robots with reinforcement learning to help people with mobility impairment Google AI researchers introduce PlaNet, an AI agent that can learn about the world using only images Researchers release unCaptcha2, a tool that uses Google’s speech-to-text API to bypass the reCAPTCHA audio challenge

Yesterday, the W3C and FIDO alliance approved using WebAuthn as an official web standard, eliminating password-based logins. WebAuthn or Web Authentication was first introduced in November 2015 as a way of replacing passwords for securing online accounts. It is now already supported by most browsers, including Chrome, Firefox, Edge, and Safari as well as in Android and Windows 10. WebAuthn allows users to log into their internet accounts using biometrics, mobile devices, and/or FIDO security keys which offer higher security over passwords alone. WebAuthn is an important component of the FIDO Alliance’s FIDO2 set of specifications. FIDO2 is a standard that supports public key cryptography and multifactor authentication. Per the official press release, FIDO2 attempts to address traditional authentication issues in four ways: Security: FIDO2 cryptographic login credentials are unique across every website; biometrics or other secrets like passwords never leave the user’s device and are never stored on a server. This security model eliminates the risks of phishing, all forms of password theft, and replay attacks. Convenience: Users log in with simple methods such as fingerprint readers, cameras, FIDO security keys, or their personal mobile device. Privacy: Because FIDO keys are unique for each internet site, they cannot be used to track users across sites. Scalability: Websites can enable FIDO2 via an API call across all supported browsers and platforms on billions of devices consumers use every day. “Web Authentication as an official web standard is the pinnacle of many years of industry collaboration to develop a practical solution for stronger authentication on the web,” said Brett McDowell, executive director of the FIDO Alliance in a statement. “With this milestone, we're moving into a new era of ubiquitous, hardware-backed FIDO Authentication protection for everyone using the internet.” WebAuthn is already implemented on sites such as Dropbox, Facebook, GitHub, Salesforce, Stripe, and Twitter. With it becoming the official standard, it is expected to have other sites use it leading to more password-free logins across the web. Announcing W3C Publishing Working Group’s updated scope and goals Microsoft Edge introduces Web Authentication for passwordless web security It’s a win for Web accessibility as courts can now order companies to make their sites WCAG 2.0 compliant.

It was just earlier last week when US Reps. Jackie Speier, Ilhan Omar, Rashida Tlaib, and 11 others wrote a letter demanding Google and Apple to ban Absher, a Saudi Government app, that allows Saudi men to control where the women can travel. “Keeping this application in your stores allows your companies and your American employees to be accomplices in the oppression of Saudi Arabian women and migrant workers”, reads the letter written by the US reps. to Apple and Google. However, Google has decided to keep Absher on the Google play store. Google communicated this decision to the office of representative Jackie Speier, stating that the app does not violate any agreements, reported INSIDER last week. Apple is yet to make a decision. Absher app is based on Saudi “guardian” law and comes with features aimed to restrict women’s travel to specific airports and routes. Also, in case the woman decides to flee from the country without permission, she can get instantly caught with Absher’s automatic SMS feature. This SMS feature sends instant messages to the guardian for times when she crosses borders or makes airport check-ins without permission. Google and Apple had decided to investigate and review the app under the rising pressure in mid-Feb when US Senator Ron Wyden had written to Apple and Google demanding them to remove Absher app from Google Play store. Apart from Wyden, Activist groups including Human Rights Watch and Amnesty International had also slammed Apple and Google, earlier last month, for hosting Absher. Speier told INSIDER that the responses from Google and Apple so far are ‘deeply unsatisfactory’. “As of today, the Absher app remains available in both the Apple App store and the Google Play Store. Facilitating the detention of women seeking asylum and fleeing abuse and control unequivocally causes harm. I will be following up on this issue with my colleagues," said Rep. Speier. An AI startup now wants to monitor your kids’ activities to help them grow ‘securly’ Babysitters now must pass Perdictim’s AI assessment to be “perfect” to get the job Twitter blocks Predictim, an online babysitter-rating service, for violating its user privacy policies; Facebook may soon follow suit

Last week the team at Facebook open-sourced homomorphic hashing for a secure update propagation. It is difficult to ensure consistency while propagating updates across a large network of peers. Even traditional methods aren’t useful as they can introduce compromises with respect to efficiency and scalability. To address this problem, the research team of Facebook worked on this project and released a paper based on the same. The paper focuses on formalizing the problem of secure update propagation and propose a system that allows a centralized distributor to propagate signed updates across a network. The researchers show that their system is secure against an attacker who can maliciously modify any update and its signature. The researchers have opted for a cryptographic primitive known as homomorphic hashing, introduced by Bellare, Goldreich, and Goldwasser. The researchers have also studied about the instantiation of the lattice-based homomorphic hash, LtHash of Bellare and Miccancio. which is a specific homomorphic hashing algorithm. The paper provides a detailed security analysis of the collision resistance of LtHash. It gives an idea about the implementation of LtHash using a selection of parameters that ensure security. This implementation has been deployed to secure update propagation in production at Facebook and is also included in the Folly open-source library. The challenges of securing update propagation A central distributor who is responsible for managing the master database and propagating updates to a set of subscribers in an efficient manner. To make this possible, the distributor has to be in charge of directly sending the updates to each subscribed client. As the number of subscribed clients and the rate of updates increases, this approach fails. This might saturate the network interface controller and further leaving it unable to finish distributing one update before the next one is ready to be pushed. A better approach is to delegate the propagation through the clients. Some of the subscribers can participate in forwarding the distributor’s original updates to other subscribers. According to researchers, this approach would reduce the number of connections the distributor manages and the bandwidth will remain unaffected. But the major issue is to ensure consistency. Each subscriber needs to trust a set of intermediate subscribers to have correctly propagated the original updates. The challenge is to maintain the integrity of the distributor’s updates across a network of subscribers that could alter those updates. And this is what is referred to as the secure update propagation problem. Experimental approaches by the researchers The possible solution could be that the distributor can use digital signatures to assert the authenticity and integrity of the messages it distributes. The distributor can generate a public and private key pair, publish the public key to every subscriber upon joining the network while keeping the private key secret. The signatures can then be constructed over the contents of the update or the contents of the updated database. In the case of signing each update, handling update propagation securely is for the distributor to directly sign the contents of each update that sent to its subscribers. The signature can be used to verify the contents before applying it to the database. While this approach prevents an attacker from modifying updates maliciously, it also adds complications to the handling of batch updates and offline database validation. So another approach suggested by researchers is to rely on a signature algorithm computed over the database contents after each update. But in this case, the distributor must iterate over the entire database to produce the signature. Hashing approach - LtHash An alternative approach is hashing where the distributor can use a hash function to hash the entire database into a small digest. The resulting digest can be directly signed, as opposed to having the distributor sign the database itself. The collision-resistant property of the hash function and the unforgeability of the signature algorithm ensures integrity through the sequence of updates. However, an ideal solution would allow the distributor and its subscribers to update the database hash entirely irrespective of the size of the database. This is possible with the use of homomorphic hashing. The team used LtHash, a specific homomorphic hashing algorithm based on lattice cryptography, for creating an efficiently updatable checksum of a database. The checksum, along with a signature from the distributor of the database, allows a subscriber to validate the integrity of database updates. LtHash was chosen in the favor of other homomorphic hashing algorithms for its performance and efficient implementation. LtHash can take a set of arbitrarily long elements as input, and produce a 2KB hash value as output. Two LtHash outputs can be “added” together by breaking each output into 16-bit chunks and performing component-wise vector addition modulo 216. To know more about the implementation of secure update propagation, check out the paper. Facebook announces ‘Habitat’, a platform for embodied ArtificiaI Intelligence research Facebook open sources Magma,a software platform for deploying mobile networks The Verge spotlights the hidden cost of being a Facebook content moderator, a role Facebook outsources to 3rd parties to make the platform safe for users  

Facebook has threatened to pull investment projects from Canada and Europe if the lobbying demands stated by Sheryl Sandberg, COO at Facebook were not met, The Guardian reports. Facebook was planning to build a data center in Canada to create jobs. The leaked memo, as seen by CW and the Guardian reveals that the deal was to be made only if Christian Paradis, Canada's then minister of industry, sends a letter of reassurance to Sandberg. According to her, the letter should reassure Facebook that the existence of the data center would not be used by the country to extend its legal jurisdiction over non-Canadian data held by Facebook. Sandberg told the officials from the European Union and Canada that if she did not receive any reassurances, then Facebook will consider other options for investment and growth. On the same day, Facebook received the letter from Canada guaranteeing the independence of non-Canadian data. The EU is yet to give such an assurance. Because of the company’s relationship with the Irish government, Facebook was hoping to influence the EU as well. These confidential documents apparently got leaked online. They were filed under seal as part of a lawsuit in California between Facebook and an app developer, Six4Three. These confidential documents show a global lobbying operation by Facebook that targets legislators around the world, including countries like the U.K., United States, Canada, India, and Brazil. In a statement to Business Insider, Facebook said, "Like the other documents that were cherry-picked and released in violation of a court order last year, these by design tell one side of a story and omit important context. As we've said, these selective leaks came from a lawsuit where Six4Three, the creators of an app known as Pikinis, hoped to force Facebook to share information on friends of the app's users. These documents have been sealed by a Californian court so we're not able to discuss them in detail." According to Computer Weekly, one of the original reporters of the news,  Marne Levine, then Facebook's vice-president of global public policy, wrote in one memo, "Sheryl took a firm approach and outlined that a decision on the datacentre was imminent. She emphasized that if we could not get comfort from the Canadian government on the jurisdiction issue we had other options.” Levine also described in the leaked messages as to how the Facebook staff distracted aides to Paradis so that other lobbyists could initiate a discussion with the ministers directly. This made Levine get the mobile numbers of the three government ministers. According to Levine, Sheryl Sandberg got comfortable around former UK chancellor George Osborne. The motive was to make him speak out against EU data laws, according to the leaked internal memo. This news is a real eye-opener in terms of how Facebook operates, which might also be used as an inspiration by other tech companies in countries where their data demands are not being met. This also seems to be a winning situation for Facebook as it is not only getting its demands fulfilled but also receiving enough support from the government's end in doing it. “In a lot of ways Facebook is more like a government than a traditional company,” Facebook CEO Mark Zuckerberg has said in an interview. Well, it seems Mark Zuckerberg is on his point this time. The involvement of government is a matter of concern for most of the users. One of the users commented on HackerNews, “Just for a little context, I think it's worth mentioning that this news comes to light when Canadians are thinking quite a bit about companies lobbying the gov't, as a bit of a scandal is brewing with the current liberal gov't[0].” Another user commented, “The Canadians agreed to not regulate other countries data. This seems pretty reasonable. Why should the Canadian government regulate how an American tech company handles German data? It makes a lot more sense for each country to have jurisdiction over data from (1) its own citizens, (2) residents on its soil or (3) data physically stored on its soil.” Facebook announces ‘Habitat’, a platform for embodied ArtificiaI Intelligence research Facebook open sources Magma, a software platform for deploying mobile networks The Verge spotlights the hidden cost of being a Facebook content moderator, a role Facebook outsources to 3rd parties to make the platform safe for users

Yesterday, Google researchers released three different research papers which describe their investigations in easy-to-adapt robotic autonomy by combining deep Reinforcement Learning with long-range planning. This research is made for people with a mobility impairment that makes them home-bound. The researchers propose to build service robots, trained using reinforcement learning to improve the independence of people with limited mobility. The researchers have trained the local planner agents in order to perform basic navigation behaviors and traverse short distances safely without collisions with moving obstacles. These local planners take noisy sensor observations, such as a 1D lidar that helps in providing distances to obstacles, and output linear and angular velocities for robot control. The researchers trained the local planner in simulation with AutoRL (AutomatedReinforcement Learning) which is a method that automates the search for RL rewards and neural network architecture. These local planners transfer to both real robots and to new, previously unseen environments. This works as building blocks for navigation in large spaces. The researchers then worked on a roadmap, a graph where nodes are locations and edges connect the nodes only if local planners can traverse between them reliably. Automating Reinforcement Learning (AutoRL) In the first paper, Learning Navigation Behaviors End-to-End with AutoRL, the researchers trained the local planners in small, static environments. It is difficult to work with standard deep RL algorithms, such as Deep Deterministic Policy Gradient (DDPG). To make it easier, the researchers automated the deep Reinforcement Learning training. AutoRL is an evolutionary automation layer around deep RL that searches for a reward and neural network architecture with the help of a large-scale hyperparameter optimization. It works in two phases, reward search, and neural network architecture search. During the reward search, AutoRL concurrently trains a population of DDPG agents, with each having a slightly different reward function. At the end of the reward search phase, the reward that leads the agents to its destination most often gets selected. In the neural network architecture search phase, the process gets repeated. The researchers use the selected reward and tune the network layers. This turns into an iterative process and which means AutoRL is not sample efficient. Training one agent takes 5 million samples while AutoRL training around 10 generations of 100 agents requires 5 billion samples which is equivalent to 32 years of training. The advantage is that after AutoRL, the manual training process gets automated, and DDPG does not experience catastrophic forgetfulness. Another advantage is that AutoRL policies are robust to the sensor, actuator and localization noise, which generalize to new environments. PRM-RL In the second paper, PRM-RL: Long-Range Robotic Navigation Tasks by Combining Reinforcement Learning and Sampling-based Planning, the researchers explain Sampling-based planners that tackle long-range navigation by approximating robot motions. In this paper, the researchers have combined PRMs with hand-tuned RL-based local planners (without AutoRL) for training robots locally and then adapting them to different environments. The researchers trained a local planner policy in a generic simulated training environment, for each robot. Then they build a PRM with respect to that policy, called a PRM-RL, over a floor plan for the deployment environment. For building a PRM-RL, the researchers connected the sampled nodes with the help of Monte Carlo simulation. The resulting roadmap can be tuned to both the abilities and geometry of the particular robot. Though the roadmaps for robots with the same geometry having different sensors and actuators will have different connectivity. At execution time, the RL agent easily navigates from roadmap waypoint to waypoint. Long-Range Indoor Navigation with PRM-RL In the third paper, the researchers have made several improvements to the original PRM-RL. They replaced the hand-tuned DDPG with AutoRL-trained local planners, which improves long-range navigation. They have also added Simultaneous Localization and Mapping (SLAM) maps, which robots use at execution time, as a source for building the roadmaps. As the SLAM maps are noisy, this change closes the “sim2real gap”, a phenomenon where simulation-trained agents significantly underperform when they are transferred to real-robots. Lastly, they have added distributed roadmap building to generate very large scale roadmaps containing up to 700,000 nodes. The team compared PRM-RL to a variety of different methods over distances of up to 100m, well beyond the local planner range. The team realized that PRM-RL had 2 to 3 times the rate of success over baseline because the nodes were connected appropriately for the robot’s capabilities. To conclude, Autonomous robot navigation can improve the independence of people with limited mobility. This is possible by automating the learning of basic, short-range navigation behaviors with AutoRL and using the learned policies with SLAM maps for building roadmaps. To know more about this news, check out the Google AI blog post. Google launches Flutter 1.2, its first feature update, at Mobile World Congress 2019 Google released a paper showing how it’s fighting disinformation on its platforms Google introduces and open-sources Lingvo, a scalable TensorFlow framework for Sequence-to-Sequence Modeling  

YouTube has disabled all comments from its videos featuring young children in order to curb the spread of pedophiles who are using YouTube to trade clips of young girls in states of undress. This issue was first discovered, when Matt Watson, a video blogger, posted a 20-minute clip detailing how comments on YouTube were used to identify certain videos in which young girls were in activities that could be construed as sexually suggestive, such as posing in front of a mirror and doing gymnastics. Youtube’s content regulation practices have been in the spotlight in recent years. Last week, YouTube received major criticism for recommending videos of minors and allowing pedophiles to comment on these posts, with a specific time stamp of the video of when an exposed private part of the young child was visible. YouTube was also condemned for monetizing these videos allowing advertisements for major brands like Nestle, Fortnite, Disney, Fiat, Fortnite, L’Oreal, Maybelline, etc to be displayed on these videos. Following this news, a large number of companies have suspended their advertising spending from YouTube and refused to do so until YouTube took strong actions. In the same week, YouTube told Buzzfeed News that it is demonetizing channels that promote anti-vaccination content. YouTube said that this type of content does not align with its policy and called it “dangerous and harmful” content. Actions taken by YouTube YouTube said that it will now disable comments worldwide on almost all videos of minors by default. It said the change would take effect over several months. This will include videos featuring young and older minors that could be at risk of attracting predatory behavior. They are further introducing new comments classifier powered by machine learning that will identify and remove twice as many predatory comments as the old one. YouTube has also banned videos that encourage harmful and dangerous challenges. “We will continue to take actions on creators who cause egregious harm to the community”, they wrote in a blog post. "Nothing is more important to us than ensuring the safety of young people on the platform," said YouTube chief executive Susan Wojcicki on Twitter. https://twitter.com/SusanWojcicki/status/1101182716593135621 Despite her apologetic comments, she was on the receiving end of a brutal backlash with people asking her to resign from the organization. https://twitter.com/g8terbyte/status/1101221757233573899 https://twitter.com/KamenGamerRetro/status/1101186868052398080 https://twitter.com/SpencerKarter/status/1101305878014242822 The internet is slowly becoming a harmful place for young tweens. Not just Youtube, recently, TikTok, the popular video-sharing app which is a rage among tweens, was accused of illegally collecting personal information from children under 13. It was fined $5.7m by the US Federal Trade Commission. TikTok has now implemented features to accommodate younger US users in a limited, separate app experience that has additional safety and privacy protections. Similar steps have, however, not been implemented across their global operations. Nestle, Disney, Fortnite pull out their YouTube ads from paedophilic videos as YouTube’s content regulation woes continue. Youtube promises to reduce recommendations of ‘conspiracy theory’. Ex-googler explains why this is a ‘historic victory’. Is the YouTube algorithm’s promoting of #AlternativeFacts like Flat Earth having a real-world impact?

Mozilla, a popular free and open-source web browser, released the largest public dataset of human voices available for use, called Common Voice, yesterday. The dataset consists of 18 different languages (including English, French, German, Mandarin Chinese, Welsh, Kabyle, etc) and adds about 1,400 hours of recorded voice clips from more than 42,000 contributors. “With this release, the continuously growing Common Voice dataset is now the largest ever of its kind, with tens of thousands of people contributing their voices and originally written sentences to the public domain (CC0)”, states the Mozilla team. The  Common Voice dataset is unique and rich in diversity as it represents a global community of voice contributors. These contributors can also opt-in to offer other information such as age, sex, and accent so that their voice clips get attached to data that is useful in training speech engines. Mozilla had enabled multi-language support back in June 2018, making Common Voice more global and inclusive. Mozilla also involves different communities contributing towards the project who have helped with launching the data collection efforts in 22 different languages and 70 more in progress on the Common Voice site. With the help of these communities, Mozilla has made the latest additions to the Common Voice dataset including languages such as Dutch, Hakha-Chin, Esperanto, Farsi, Basque, and Spanish. It also plans to continue working with these communities to retain the diversity in the voices represented. As per the Mozilla team, these public contributors are not only able to track the progress per language in recording and validation but have also improved the prompts that vary from clip to clip. Mozilla has also added a new option to create a saved profile, that helps the contributors keep track of their progress and metrics across different languages. It also offers optional demographic profile information that further helps improve the audio data used in training speech recognition accuracy. Apart from the dataset, Mozilla also has goals towards contributing to a more diverse and innovative voice technology ecosystem in the future. It aims to release voice-enabled products while also making sure to support researchers and smaller players. “For Common Voice, our focus in 2018 was to build out the concept, make it a tool for any language community to use, optimize the website, and build a robust backend. Our overall aim remains: Providing more and better data to everyone in the world who seeks to build and use voice technology”, states the Mozilla team. For more information on this announcement, check out the official Mozilla blog post. Mozilla partners with Scroll to understand consumer attitudes for an ad-free experience on the web Mozilla shares key takeaways from the Design Tools survey Mozilla partners with Ubisoft to Clever-Commit its code, an artificial intelligence assisted assistant

Earlier this month, OpenAI released a limited version of GPT-2, its unsupervised language model, with a warning that it could be used for automating the production of fake content. While many machine learning researchers supported their decision for putting AI safety first, some felt that OpenAI is spreading fear and hindering reproducibility while others felt it was a PR stunt. AI Village, a community of hackers and data scientists working together to spread awareness about the use and misuse of AI, also shared its views on GPT-2 and its threat models. AI Village in the blog post said, “...people need to know what these algorithms that control their lives are capable of. This model seems to have capabilities that could be dangerous and it should be held back for a proper review.” These are the potential threat models in which GPT-2 can be used, according to AI Village: The bot-based misinformation threat model Back in 2017, when FCC launched a public comments website, it faced a massive coordinated botnet attack. This botnet posted millions of comments alongside humans about anti-net neutrality. Researchers were able to detect this disinformation by using regex to filter for all these comments with near certainty. AI Village said that if these comments were generated by GPT-2, we wouldn’t have been able to find that these comments were written by a botnet. Amplifying human generated disinformation We have seen a significant amount of bot-activity on different online platforms. Very often, these activities just amplify fake content created by humans by giving upvotes and likes. How these bots work is that they log in on any online platform, like the target post, and then log off for the next bot. This behavior is quite different from a human, who actually scroll through posts and stay on these social media platforms for some time. This is what gives away whether the user is a bot or an actual human. This metadata of login times, locations, site activity can prove to be really helpful in detecting bots. Automated spear phishing In a paper published in 2016, two data scientists, John Seymour, and Philip Tully introduced SNAP_R. It is a recurrent neural network that can learn to tweet phishing posts targeting specific end users. The GPT-2 language model could also be used for automated spear phishing campaigns. How can we prevent the misuse of AI? OpenAI with this decision wanted to start a discussion about the responsible release of machine learning models, and AI Village hopes that having more such discussions could prevent AI threats on our society. “We need to have an honest discussion about misinformation & disinformation online. This discussion needs to include detecting and fighting botnets, and users who are clearly not who they say they are from generating disinformation.” In recent years, we have seen many breakthroughs in AI, but comparatively very less effort has been put into finding ways to prevent the malicious use of AI. Generative Adversarial Networks are now capable of producing headshots that are indistinguishable from photos. Deepfakes for video and audio have advanced so much that they almost seem real. Currently, we do not have any mechanism in place for researchers to responsibly release their work that could potentially be used for evil. “With truly dangerous AI, it should be locked up. But after we verify that it's a threat and scope out that threat”, states the blog post. AI Village believes that we need to have more thorough discussions about such AI systems and the damages they can do. Last year, in a paper, AI Village listed down some of the ways through which AI researchers, companies, legislators, security researchers, and educators can come together to prevent and mitigate the AI threats: Policymakers and technical researchers should come together to investigate, prevent, and mitigate potential malicious uses of AI. Researchers before opening their work to the general public should think about the dual use of their work. They should also proactively reach out to relevant actors when harmful applications are foreseeable. We should have the best practices and more sophisticated methods in place to address dual use concerns. We should try to expand the range of stakeholders and domain experts involved in discussions of these challenges. You can read the AI Village post on its official website. OpenAI’s new versatile AI model, GPT-2 can efficiently write convincing fake news from just a few words Artificial General Intelligence, did it gain traction in research in 2018? OpenAI team publishes a paper arguing that long term AI safety research needs social scientists

On Tuesday, Stanford University researchers introduced two recent datasets collected by the Stanford NLP Group to further advance the field of machine reading. These two new datasets CoQA (Conversational Question Answering), and HotpotQA work towards incorporating more “reading” and “reasoning” in the task of question answering and move beyond questions that can be answered by simple pattern matching. CoQA aims to solve the problem by introducing a context-rich interface of a natural dialog about a paragraph of text. The second one, HotpotQA goes beyond the scope of one paragraph and presents the challenge of reasoning over multiple documents to arrive at the answer. Lately, solving the task of machine reading or question answering is becoming an important section  towards a powerful and knowledgeable AI system. Recently, large-scale question answering datasets like the Stanford Question Answering Dataset (SQuAD) and TriviaQA have progressed a lot in this direction. These datasets have enabled good results in allowing researchers to train deep learning models What is CoQA? Most of the question answering systems are limited to answering questions independently. But usually while having a conversation there happens to be a few interconnected questions. Also, it is more common to seek information by engaging in conversations involving a series of interconnected questions and answers. CoQA is a Conversational Question Answering dataset developed by the researchers at Stanford University to address this limitation and working in the direction of conversational AI systems. Features of CoQA dataset The researchers didn’t restrict the answers to be a contiguous span in the passage. As a lot of questions can’t be answered by a single span in the passage, which will limit the naturalness of the conversations. For example, for a question like How many times a word has been repeated?, the answer can be simply three despite text in the passage not spelling this out directly. Most of the QA datasets mainly focus on a single domain, which makes it difficult to test the generalization ability of existing models. The CoQA dataset is collected from seven different domains including, children’s stories, literature, middle and high school English exams, news, Wikipedia, Reddit, and science. The CoQA challenge launched in August 2018, has received a great deal of attention and has become one of the most competitive benchmarks. Post the release of Google’s BERT models, last November, a lot of progress has been made, which has lifted the performance of all the current systems. Microsoft Research Asia’ state-of-the-art ensemble system “BERT+MMFT+ADA” achieved 87.5% in-domain F1 accuracy and 85.3% out-of-domain F1 accuracy. These numbers are now approaching human performance. HotpotQA: Machine Reading over Multiple Documents We often find ourselves in need of reading multiple documents to find out about the facts about the world. For instance, one might wonder, in which state was Yahoo! founded? Or, does Stanford have more computer science researchers or Carnegie Mellon University? Or simply, How long do I need to run to burn the calories of a Big Mac? The web does contain the answers to many of these questions, but the content is not always in a readily available form, or even available at one place. To successfully answer these questions, there is a need for a QA system that finds the relevant supporting facts and to compare them in a meaningful way to yield the final answer. HotpotQA is a large-scale question answering (QA) dataset that contains about 113,000 question-answer pairs. These questions require QA systems to sift through large quantities of text documents for generating an answer. While collecting the data for HotpotQA, the researchers have annotators to specify the supporting sentences they used for arriving at the final answer. To conclude, CoQA considers those questions that would arise in a natural dialog given a shared context, with challenging questions that require reasoning beyond one dialog turn. While, HotpotQA focuses on multi-document reasoning, and challenges the research community for developing new methods to acquire supporting information. To know more about this news, check out the post by Stanford. Stanford experiment results on how deactivating Facebook affects social welfare measures Thank Stanford researchers for Puffer, a free and open source live TV streaming service that uses AI to improve video-streaming algorithms Stanford researchers introduce DeepSolar, a deep learning framework that mapped every solar panel in the US

Redis Labs, the provider of Redis Enterprise, announced details about its 14th consecutive quarter of double-digit growth and its annual growth of more than 60% in the company’s 2019 fiscal year. It was just last week, when Redis Labs, a California-based computer software startup, announced that it has raised $60 million in Series E financing round led by a new and leading private equity firm, Francisco Partners. Moreover, Redis Labs finished the 2019 fiscal year with more than 250 full-time employees, with its global headcount increasing 50 percent in the past year. Also, since the company scales its global go-to-market team, new offices have been opened in Austin, Texas, and Bangalore (India) to drive adoption of Redis Enterprise. Based on the record growth results and recently secured funding, Redis Labs aims to accelerate its plans in the new fiscal year across different departments including sales, marketing, and product development. This would help them meet the demands for a multi-model database that is capable of delivering the performance, deployment flexibility, and seamless scaling to advance instant experiences. Redis Labs is already continuing to expand the business with other Global 1000 enterprises such as Alliance Data, ANZ Bank, Applied Materials, Carrefour, Dick's Sporting Goods, Thomas Cook, Mercedes Benz, Nordea, UIPath, and WestPac. Other than that, Alvin Richards has been promoted to Chief Product Officer from Chief Education Officer, that he had been appointed as back in 2017. This will help him continue the company’s market leadership and deliver innovation for the multi-model database market. Redis was named as 2019 technology of the year for the second time by IDG's InfoWorld. Also, Redis secured the place of the seventh most popular database in DB-Engines ranking among more than 300 databases. Apart from having the highest rating among the top seven database providers, it is also the first database that achieved 1 billion launches on Docker Hub in 2018. “Redis Enterprise delivers the requirements for a multi-model cloud-native database that operates at record-breaking performance with unmatched cost efficiency”, mentioned Ofer Bengal, co-founder, and CEO at Redis Labs in an email sent to us. RedisGraph v1.0 released, benchmarking proves its 6-600 times faster than existing graph databases Redis Cluster Features Overview Redis Labs moves from Apache2 modified with Commons Clause to Redis Source Available License (RSAL)

At the MariaDB OpenWorks held earlier this week, MariaDB CEO Michael Howard took a stab at big proprietary cloud vendors and accused them of "strip-mining open-source technologies and companies," and “abusing the license and privilege, not giving back to the community." His keynote at the event described his plans for MariaDB, the future of MariaDB, and how he plans for MariaDB on becoming an ‘heir to Oracle and much more’. Furthermore, the entire keynote saw instances of Howard targeting his rivals- namely Amazon and Oracle- and comparing MariaDB mottos to its rivals. "We believe proprietary and closed licenses are dead. We believe you have to be a general-purpose database and not a relegated niche one, like--and nothing against it--time series. That's not going to be a general purpose database that will drive applications worldwide." MariaDB is an example of such a database. Accusations on Oracle and Amazon AWS Targeting Oracle, Howard said, "Now, you can migrate complex operational Oracle systems to MariaDB. Last year, we had one of the largest banks--Development Bank of Singapore--in the world forklift from Oracle to MariaDB. Since then, MariaDB has seen five times the number of Oracle migrations happening over the last year." Howard has also accused Amazon’s AWS of promoting its brand and making MariaDB instances on AWS look incompetent in the process. When Austin Rutherford, MariaDB's VP of Customer Success, showed the audience the result of a HammerDB benchmark on AWS EC2, AWS's default MariaDB instances did poorly. The AWS homebrew Aurora, which is built on top of MySQL, consistently beat the former database. The top-performing DBMS was MariaDB Managed Services on AWS. While these results initially were not a major cause of concern, Howard observed that one of the biggest retail drug companies in the world-and a MariaDB customer-had told MariaDB that "Amazon offers the most vanilla MariaDB around. There's nothing enterprise about it. We could just install MariaDB from source on EC2 and do as well." It was then that he "began to wonder, Is there something that they're deliberately crippling?" Further adding “There is something not kosher happening." Comparing MariaDB to Aurora, Howard said, "The best Aurora can do in a failover is 12 seconds. MariaDB can do it less than a second." ‘Heir to Oracle’ In his keynote, he speaks about making MariaDB the 'heir apparent' to Oracle, even including a checklist of what needs to be achieved to be that 'drop-in' replacement for the market-leading database. Source: Computerworld UK According to The Register, just last year, MariaDB released an Oracle compatibility layer, which allows customers to migrate their applications from Oracle to MariaDB, and also use their internal skills. “All these Oracle application developers and people familiar with Oracle – you can’t just say ‘jump off a cliff onto new ground’; you have to give them a bridge. Sometimes that’s emotional, sometimes it’s technical.” “It was so jarring to the proprietary vendors who pride themselves on secrecy, on taking advantage – at least monetarily, in the margins sense – from customers,” he said. “Open-source destroys these artificial definitions and boundaries that have been so, so much a part of the software industry.” Speaking to Computerworld UK , Howard further explained his views on the big cloud vendors. "Oracle as the example of on-premise lock-in and Amazon being the example of cloud lock-in. You could interchange the names, you can honestly say now that Amazon should just be called Oracle Prime, they have gone so aggressive. Fortunately or unfortunately, depending on whose position you want to take, it's all good for MariaDB because we can act as consumer protection. We are going to protect the brand quality and technical quality of our product no matter where it sits." Red Hat Satellite to drop MongoDB and will support only PostgreSQL backend Red Hat drops MongoDB over concerns related to its Server Side Public License (SSPL) GNU Health Federation message and authentication server drops MongoDB and adopts PostgreSQL

Today, the Facebook research team announced ‘Habitat’, a new platform for embodied AI research. According to the team, this is a “modular high-level library to train embodied AI agents across a variety of tasks, environments, and simulators”. This will empower a shift from ‘internet AI’ with static datasets to an embodied AI model with agents acting in realistic environments. The project was launched by the Facebook Reality Lab, Georgia Tech, SFU, Intel, and Berkeley to restore the disconnect between ‘internet AI’ and ‘embodied AI’.  It will standardize the entire ‘software stack’ for training embodied agents, and release modular high-level libraries to train and deploy embodied agents. An important objective of Habitat-API is to make it easy for users to use a 3D environment and set up a variety of embodied agent tasks in it. Habitat consists of Habitat-Sim, Habitat-API, and Habitat Challenge. #1 Habitat-Sim This is “a flexible, high-performance 3D simulator with configurable agents, multiple sensors, and generic 3D dataset handling”. It also has built-in support for SUNCG, MatterPort3D, Gibson and other datasets. Habitat-Sim achieves several thousand frames per second (FPS) running single-threaded, and reaches over 10,000 FPS multi-process on a single GPU on rendering a scene from the Matterport3D dataset. #2 Habitat-API Habitat-API defines embodied AI tasks, configuring embodied agents, training these agents, and benchmarking their performance on the defined tasks using standard metrics. #3 Habitat Challenge Habitat Challenge is an autonomous navigation challenge that benchmarks and accelerates progress in embodied AI. Participants can upload code and not predictions- unlike classical 'internet AI' image dataset-based challenges. The uploaded agents are evaluated to test for generalization You can head over to Facebook’s official announcement for more information on this news. Facebook and Google pressurized to work against ‘Anti-Vaccine’ trends after Pinterest blocks anti-vaccination content from its pinboards Facebook’s AI Chief at ISSCC talks about the future of deep learning hardware Regulate Google, Facebook, and other online platforms to protect journalism, says a UK report