Tech News - Data

Just after RedHat announced its plans to drop MongoDB from its Satellite system management solution because of it being licensed under SSPL, GNU has followed suit. Earlier this week, GNU announced its plans move its GNU Health Federation message and authentication server -Thalamus- from MongoDB to PostgreSQL. As listed on the post, the main reason for this switch is because MongoDB decided to change the license of the server to their Server Side Public License (SSPL). Because of this decision, many GNU/Linux distributions are no longer including the Mongodb server. In addition to these reasons, GNU expresses their concerns that even the organizations like the OSI and Free Software Foundation are showing their reluctance to accept this idea. Adding to this hesitation of accepting the license; rejection from a large part of the Libre software community and the immediate end of support from GPL versions of MongoDB has lead to the adoption of PostgreSQL for Thalamus. Dr. Luis Falcon, President of GNU Solidario says that one of the many reasons for choosing PostgreSQL was its JSON(B) support that  provides the flexibility and scalability found in document oriented engines. The upcoming thalamus server will be designed to support PostgreSQL. To stay updated with further progress on this announcement, head over to the GNU blog. GNU Bison 3.3  released with major bug fixes, yyrhs and yyphrs tables, token constructors and more GNU ed 1.15 released! GitHub now supports the GNU General Public License (GPL) Cooperation Commitment as a way of promoting effective software regulation  

It was just two weeks back when Chinese developers protested over the “996 work schedule”, that requires employees to work from 9 am to 9 pm, 6 days a week in China. And now the news of Chinese tech firms firing employees over the age of 30 is trending. Shelly Banjo, Roving Asia Tech Reporter, at Bloomberg published a post in May last year where she shed light on the grotesque reality of the ageism work culture in China. One such example is of Helen He, a tech recruiter in Shanghai, who has been instructed by her bosses to not hire people over 35 years of age. “Most people in their 30s are married and have to take care of their family—they’re not able to focus on the high-intensity work. If a 35-year-old candidate isn’t seeking to be a manager, a hiring company wouldn’t even give that CV a glance”, said Helen. Ageism within tech firms is at its peak in China and is taking a toll over its employees, as they are forced to move out of the industry. Banjo states that the “30+ middle-age crisis”,  is rife in China. For instance, close to three-quarters of tech workers in China are below thirty in age and the employers further promote this concept. The core reason behind Chinese employers fortifying this practice is because employees over thirty years of age are not considered as efficient as the young. Apart from that, anybody over 30 years old is likely to be experienced and demand a higher wage. Whereas, young employees can be hired at lower wages (as most don’t have a family to look after), contributing to higher profits on the lower scale for the company. Banjo states that China’s 996 work schedule and its hiring young policies reflect China’s obsession with achieving global tech domination. However, the consequences of this obsession are hard-hitting. Banjo gives an example of Ou Jianxin, a 42-year-old, research engineer at ZTE Corp. who committed suicide after he was fired from his role without informing him of the reason. However, after the news went public, people had their suspicions and many blamed it on his age, saying that, he would have already been considered “too old” to be an engineer in China. Another example presented by Banjo is of a job search results on Zhaopin.com that over 10,000 job postings calling for applicants younger than 35. One such job posting is from e-commerce retailer JD.com Inc, which seeks an applicant with a master’s degree for a senior manager position between the age of 20-28. Moreover, although China has national laws that prohibit discrimination based on gender, religion, and disability, there are no laws based on declining someone an offer based on his/her age. “Age-dismissal victims rarely ask for help from lawyers,” says Lu Jun, a social activist and visiting scholar at Fordham University School of Law. But there are some who have fought against the ageism policy. For instance, in 2011 the Shenzhen Stock Exchange had put up a recruitment notice on its website asking for applicants younger than 28. But, the director of a local nonprofit wrote an open letter about this listing to the municipal bureau of human resources, after which the media made the story viral, leading to stock exchange investigating into the listing and taking it down. Others have a different way to fight ageism. One such example presented is of Liu Huai Yi, 33, who was fired from his IT role at Nokia Corp. in Chengdu. Liu says the incident pushed him “to change and improve.. skills to get a better job. I don’t buy the idea that after 35 you can’t get a job. Someone in IT has to just keep learning to keep up.” After a long job search, Liu got another IT job in a multinational healthcare company. However, the ageist policies are not just a part of China’s work culture but have also spread to other parts of the world. For instance, in March 2018, ProPublica conducted an investigation that showed that IBM cut 20,000 older employees in the U.S. over a course of last five years to “sharply increase hiring of people born after 1980.” A user named “duxup” on Hacker News commented, “A little bit like the US maybe? I was already in a technical field, was laid off after a company buyout, I changed careers and took a coding Bootcamp. The other two older guys (at Bootcamp) and I got a lot of "culture" related questions. One dude was actually told by the recruiter that they were worried he was too old, he was surprised they'd actually say it to him so he asked ... and the recruiter repeated herself happily”. Another user “jdietrich” commented on Hacker News, “Older workers expect higher wages and are less willing to tolerate unpaid overtime. They're harder to bullshit with cheap "perks" like foosball tables and beer on Friday. An experienced developer might be better, but who cares when you can get two junior devs for the same price and they'll work 14 hour days during your quarterly "crunch"? They don't believe in the 10x developer, they don't even believe in the 1.1x developer; their employees are just meat in a seat”. What the US-China tech and AI arms race means for the world – Frederick Kempe at Davos 2019 China’s Huawei technologies accused of stealing Apple’s trade secrets, reports The Information Is China’s facial recognition powered airport kiosks an attempt to invade privacy via an easy flight experience

Alibaba, the ecommerce leader and multinational conglomerate surprises the advertising market with a smart copywriting tool, AI-CopyWriter. The digital marketing and big data arm Alimama developed artificial intelligence-powered copywriting tool. The tool is backed with powerful deep learning mechanism and natural language processing technology to deliver thousands of marketing content in just couple of seconds. This AI copywriting tool runs through the tens of millions of sampled data in the back end and generates copies for products in just few seconds. The tool delivers high efficiency with more than 20,000 copy lines in single second, reducing the repetitive copywriting jobs of advertising and marketing teams. This mind-blowing product is simple to use. One simply needs to insert the url link of a product page and the smart copy engine returns with results of numerous innovative copy ideas with just a button click away. According to Alimama, the AI Copywriting tool has been validated through the Turing test and can generate tens of thousands of copy lines in one second. This tool is capable of providing tone-specific copy lines such as funny, loving, poetical or a promotional one along with a adjustments of their word characters. Prior to this tool, the team at Alimama innovated a smart banner designer tool for small and mid-sized businesses which they can use to redesign and resize the advertising banners on e-commerce platforms with just a slide of a mouse. The team also recently released a smart video editing tool powered with AI technology, through which advertising and promotion teams can generate a quick 20 seconds video in almost less than minute. The tool has already been proven very successful by renowned apparel chain Espirit, US fashion brand Dickies and by website aggregators such as Taobao and Tmall. “The AI copywriter is a really amazing tool. Based on a massive database of existing copy and advanced AI technologies, the tool can reduce the repetitive and tedious copywriting workload for our teams.” says E-Commerce Head - Asia Pacific market at Esprit. Google’s translation tool is now offline Adobe to spot fake images using Artificial Intelligence Microsoft start AI School to teach Machine Learning and Artificial Intelligence

The American Civil Liberties Union (ACLU) has revealed that Amazon has been selling its facial recognition software, called Rekognition, to a number of law enforcement agencies in the U.S. Using a freedom of information requests, the ACLU obtained correspondence between the respective departments and Amazon. According to the ACLU, Rekognition is a dangerous step towards a surveillance state. It could, the organization argues, lead to serious infringement on civil liberties. Here's what ACLU had to say in a post published on Tuesday 22 May: People should be free to walk down the street without being watched by the government. By automating mass surveillance, facial recognition systems like Rekognition threaten this freedom, posing a particular threat to communities already unjustly targeted in the current political climate. Once powerful surveillance systems like these are built and deployed, the harm will be extremely difficult to undo. How is Rekognition currently being used? Two U.S. police departments are using Rekognition. In Oregon, the Washington County Sheriff's Office is using the facial recognition tool to identify persons of interest from a database of 300,000 mugshots. This is a project that has been underway for some time. Chris Adzima, Senior Information Systems Analyst for the Washington County Sheriff’s Office, wrote a guest post on the AWS website outlining how they were using Rekognition in June 2017. Once the architecture was in place, the team built a mobile app to make the technology usable for officers. In Orlando, meanwhile, police have been using AWS for 'consulting and advisory services.' They are seeking to implement Rekognition in a project referred to in the documentation as 'Orlando Safety Video POC'. Orlando City police are paying $39,000 for AWS' time on the project. Civil liberties organizations pen an open letter to Jeff Bezos The ACLU, along with a number of other organizations, including the Electronic Frontier Foundation and Data for Black Lives, penned an open letter to Jeff Bezos to express their concern. In an appeal to Amazon's past commitment to civil liberties, the letter stated: In the past, Amazon has opposed secret government surveillance. And you have personally supported First Amendment freedoms and spoken out against the discriminatory Muslim Ban. But Amazon’s Rekognition product runs counter to these values. As advertised, Rekognition is a powerful surveillance system readily available to violate rights and target communities of color. The letter presents an impassioned plea for Amazon to consider the way in which it is its complicit with government agencies. It also offers a serious warning about the potential consequences of facial recognition technology in the hands of law enforcement. Amazon defends collaborating with police Amazon has been quick to defend itself. In a statement emailed to various news organizations, the company said "Our quality of life would be much worse today if we outlawed new technology because some people could choose to abuse the technology. Imagine if customers couldn’t buy a computer because it was possible to use that computer for illegal purposes? Like any of our AWS services, we require our customers to comply with the law and be responsible when using Amazon Rekognition.” However, the key issue with Amazon's statement is that the analogy with personal computers doesn't quite hold. Individuals aren't responsible for maintaining the law, and neither do they hold the same power that law enforcement agencies do. Technology might change how individuals behave, but that behavior must still comply with the law. The current scenario is a little different; the concern is around how technology might actually change the way the law functions. There isn't, strictly speaking at least, any way of governing how that happens. Whatever you make of Amazon's work with law enforcement, it's clear that we are about to enter a new era of disruption and innovation in public institutions. For some people, collaboration between public and private realms opens up plenty of opportunities. But there are many dangers that must be monitored and challenged. Read next: Top 10 Tools for Computer Vision [Link] Admiring the many faces of Facial Recognition with Deep Learning [Link]

Google has been one of the leading forces in the area of text-to-speech (TTS) conversions. The company has further leaped ahead in this domain with the launch of Tacotron 2. The new technique is a combination of Google’s Wavenet and the original Tacotron—Google’s previous speech generation projects. WaveNet is a generative model of time domain waveforms. It produces natural sounding audio fidelity and is already used in some complete TTS systems. However, the inputs to WaveNet need significant domain expertise to produce as they require elaborate text-analysis systems and a detailed pronunciation guide. Tacotron is a sequence-to-sequence architecture for producing magnitude spectrograms from a sequence of characters i.e. it synthesizes speech directly from words. It uses a single neural network trained from data alone for production of the linguistic and acoustic features .Tacotron uses the Griffin-Lim algorithm for phase estimation. Griffin-Lim produces characteristic artifacts and lower audio fidelity than approaches like WaveNet. Although Tacotron was efficient with respect to patterns of rhythm and sound, it wasn’t actually suited for producing a final speech product. Tacotron 2 is a conjunction of the above described approaches. It features a tacotron style, recurrent sequence-to-sequence feature prediction network that generates mel spectrograms. Followed by a modified version of WaveNet which generates time-domain waveform samples conditioned on the generated mel spectrogram frames. Source: https://arxiv.org/pdf/1712.05884.pdf In contrast to Tacotron, Tacotron 2 uses simpler building blocks, using vanilla LSTM and convolutional layers in the encoder and decoder. Also, each decoder step corresponds to a single spectrogram frame. The original WaveNet used linguistic features, phoneme durations, and log F0 at a frame rate of 5 ms. However, these lead to significant pronunciation issues when predicting spectrogram frames spaced this closely. Hence, the WaveNet architecture used in Tacotron 2  work with 12.5 ms feature spacing by using only 2 upsampling layers in the transposed convolutional network. Here’s how it works: Tacotron 2 uses a sequence-to-sequence model optimized for TTS in order to map a sequence of letters to a sequence of features that encode the audio. These sequence of features include an 80-dimensional audio spectrogram with frames computed every 12.5 milliseconds. They are used for capturing word pronunciations, and various other qualities of human speech such as volume, speed and pitch. Finally, these features are converted to a waveform of 24 kHz using a WaveNet-like architecture. Tacotron 2 system can be trained directly from data without relying on complex feature engineering. It achieves state-of-the-art sound quality close to that of natural human speech. Their model achieves a mean opinion score (MOS) of 4.53 comparable to a MOS of 4.58 for professionally recorded speech. Google has also provided some Tacotron 2 audio samples that demonstrate the results of their TTS system. In the future, Google would work on improving their system to pronounce complex words, generate audio in realtime, and directing a generated speech to sound happy or sad. The entire paper is available for reading at Arxiv archives here.

Yesterday, Danilo Poccia, an Evangelist at Amazon Web Services announced the PostgreSQL-compatible edition of Aurora Serverless will be generally available. Aurora PostgreSQL Serverless lets customers create database instances that only run when needed and automatically scale up or down based on demand. If a database isn’t needed, it will shut down until it is needed. With Aurora Serverless, users have to pay on a per-second basis for the database capacity one uses when the database is active, plus the usual Aurora storage costs. Last year, Amazon had made the Aurora Serverless MySQL generally available. How the Aurora PostgreSQL Serverless storage works When a database is created with Aurora Serverless, users set the minimum and maximum capacity. The client applications transparently connect to a proxy fleet that routes the workload to a pool of resources that are automatically scaled. Scaling is done quickly, as the resources are ‘warm’ and ready to be added to serve user requests. Image Source: Amazon blog The storage layer is independent from the computer resources, used by the database, as the storage is not provisioned in advance. The minimum storage is 10GB, however based on the database usage, the Amazon Aurora storage will automatically grow, up to 64 TB, in 10GB increments with no impact to database performance. How to create an Aurora Serverless PostgreSQL Database Create a database from the Amazon RDS console, using Amazon Aurora as engine. Select the PostgreSQL version compatible with Aurora serverless. After selecting the version, the serverless option becomes available. Currently, its is version 10.5. Enter an identifier to the new DB cluster, choose the master username, and let Amazon RDS generate a password. This will let users retrieve their credentials during database creation. Select the minimum and maximum capacity for the database, in terms of Aurora Capacity Units (ACUs), and in the additional scaling configuration, choose to pause compute capacity after 5 minutes of inactivity. Based on the setting, Aurora Serverless will automatically create scaling rules for thresholds for CPU utilization, connections, and the available memory. Aurora Serverless PostgreSQL will now be available in US East (N. Virginia and Ohio), US West (Oregon), EU (Ireland), and Asia Pacific (Tokyo). Many developers are happy with the announcement. https://twitter.com/oxbits/status/1148840886224265218 https://twitter.com/sam_jeffress/status/1148845547110854656 https://twitter.com/maciejwalkowiak/status/1148829295948771331 Visit the Amazon blog for more details. How do AWS developers manage Web apps? Amazon launches VPC Traffic Mirroring for capturing and inspecting network traffic Amazon adds UDP load balancing support for Network Load Balancer

The ‘Neural Information Processing Systems’, or ‘NIPS’, a well-known machine learning and computational neuroscience conference adopted ‘NeurIPS’ as an alternative acronym for the conference, last week. The acronym ‘NIPS’ had been under the spotlight worldwide over the past few years as some members of the community thought of the acronym as “sexist” and pointed out that it is offensive towards women. “Something remarkable has happened in our community. The name NeurIPS has sprung up organically as an alternative acronym, and we’re delighted to see it being adopted”, mentioned the NeurIPS team. NIPS team also added that they have taken a couple of measures to support the new acronym. This is why all signage and the program booklet for the 2018 meeting will have either the full conference name or NeurIPS mentioned to refer to the conference. Sponsors have also been asked to make sure that they make the required changes within their document materials. A branding company has also been hired to get a new logo designed for the conference. Moreover, the conference site has been moved to neurips.cc. “one forward-thinking member of the community purchased neurips.com and described the site’s purpose as ‘host[ing] the conference content under a different acronym... until the board catches up,” as mentioned on NeurIPS news page. NIPS organizers had conducted a  poll, back in August, on the NIPS website asking people if they agree or disagree with the name change. Around 30% of the respondents had answered that they support the name change (28% males and about 44% females) while 31% ‘strongly disagreed’ with the name change proposal (31% male and 25% female). This had led to NIPS keeping the name as it is. However, many people were upset by the board’s decision, and when the emphasis on a name change within the community became evident, the name got revised. One such person who was greatly dissatisfied with the decision was Anima Anandkumar, director of Machine Learning at Nvidia, who had started a petition on change.org last month. The petition managed to gather 1500 supporters as of today. “The acronym of the conference is prone to unwelcome puns, such as the perhaps subversively named pre-conference “TITS” event and juvenile t-shirts such as “my NIPS are NP-hard”, that add to the hostile environment that many ML researchers have unfortunately been experiencing” reads the petition. Anima pointed out that some of these incidents trigger uncomfortable memories for many researchers who have faced harassing behavior in the past. Moreover, Anandkumar tweeted out with #ProtestNIPS in support of the conference changing its name, which received over 300 retweets. https://twitter.com/AnimaAnandkumar/status/1055262867501412352 After the board’s decision to rebrand the name, Anandkumar tweeted out thanking everyone for their support for #protestNIPS. “ I wish we could have started with a clean slate and done away with problematic legacy, but this is a compromise. I hope we can all continue to work towards better inclusion in #ml”. Other than Anandkumar, many other people had been equally active in amplifying the support for #protestNIPS. People in support of #protestNIPS Jeff Dean, head of Google AI Dean had tweeted in support of Anandkumar, saying that NIPS should take the issue of name change seriously: https://twitter.com/JeffDean/status/1055289282930176000 https://twitter.com/JeffDean/status/1063679694283857920 Dr. Elana J Fertig, Associate Professor of Applied Mathematics, Johns Hopkins Elana had also tweeted in support of #protestNIPS. “These type of attitudes cannot be allowed to prevail in ML. Women need to be welcome to these communities. #WomenInSTEM” https://twitter.com/FertigLab/status/1063908809574354944 Daniela Witten, professor of (bio)statistics, University of Washington Witten tweeted saying: “I am so disappointed in @NipsConference for missing the opportunity to join the 21st century and change the name of this conference. But maybe the worst part is that their purported justification is based on a shoddy analysis of their survey results”. https://twitter.com/daniela_witten/status/1054800517421924352 https://twitter.com/daniela_witten/status/1054800519607181312 https://twitter.com/daniela_witten/status/1054800521582731264 “Thanks to everyone who has taken the time to share thoughts and concerns regarding this important issue. We were considering alternative acronyms when the community support for NeurIPS became apparent. We ask all attendees this year to respect this solution from the community and to use the new acronym in order that the conference focus can be on science and ideas”, mentioned the NeurIPS team. NIPS 2017 Special: Decoding the Human Brain for Artificial Intelligence to make smarter decisions NIPS 2017 Special: A deep dive into Deep Bayesian and Bayesian Deep Learning with Yee Whye Teh NIPS 2017 Special: How machine learning for genomics is bridging the gap between research and clinical trial success by Brendan Frey

The Timescale team announced a release candidate, TimescaleDB 1.0, designed to support full SQL, yesterday. TimescaleDB is the first open-source time-series database that scales for fast ingest and complex queries while providing full SQL support. It also natively supports full SQL. TimescaleDB 1.0 comes with features such as native Grafana integration, first-class Prometheus support, and dozens of other new features. The timescale is the company behind the first open-source time-series database and is powered by PostgreSQL. TimescaleDB has helped businesses across the world with mission-critical applications such as industrial data analysis, complex monitoring systems, operational data warehousing, financial risk management, geospatial asset tracking, and more. TimescaleDB 1.0 key Features TimescaleDB 1.0 offers first-class Prometheus support for long-term storage along with native Grafana integration. First-class Prometheus Support There is now an added native support in TimescaleDB to act as a remote storage backend for Prometheus (a monitoring system and time-series database). This adds many benefits to Prometheus such as a full SQL interface, long-term replicated storage, support for late data, data updates, and the ability to JOIN monitoring data against other business data. Native Grafana Integration TimscaleDB 1.0 now comes with a graphical SQL query builder for Grafana and additional support. In addition to these two major features, there are other TimescaleDB features: TimescaleDB 1.0  is fast, flexible, and built to scale. It supports full SQL i.e. it is similar to PostgreSQL on the outside but is architected for time-series internally. TimescaleDB 1.0 provides the largest ecosystem of any time-series database such as Tableau, Grafana, Apache Kafka, Apache Spark, Prometheus, Zabbix support. It is now enterprise ready and offers reliability and tooling of PostgreSQL, enterprise-grade security, and production-ready SLAs. TimescaleDB 1.0 manages time-series data. It offers automatic space-time partitioning, a hypertable abstraction layer, adaptive chunk sizing, and other new functions for easier time-series analytics in SQL. It also comprises features such as geospatial analysis, JSON support, along with easy schema management. TimescaleDB has managed to achieve some significant milestones since its launch in April 2017. It managed to surpass 1 million downloads and 5,000 GitHub stars. It has Bloomberg, Comcast, Cray, Cree, and LAIKA as production users. “Based on all the adoption we’re seeing, it’s becoming clear to us that all data is essentially a time-series data. We’re building TimescaleDB to accommodate this growing need for a performant, easy-to-use, SQL-centric, and enterprise-ready time-series database,” says Ajay Kulkarni, Timescale founder on the TimeScale announcement page. To get started, download TimescaleDB (installation instructions). You can explore the first release candidate for TimescaleDB 1.0 at Github or on Docker. For more information, check out the official release notes. Introducing Watermelon DB: A new relational database to make your React and React Native apps highly scalable Say hello to FASTER: a new key-value store for large state management by Microsoft IBM Files Patent for “Managing a Database Management System using a Blockchain Database”

On 10th July, Stripe’s API services went down twice, from 16:36–17:02 UTC and again from 21:14–22:47 UTC. Though the services recovered immediately, it had caused significantly elevated error rates and response times. Two days after the incident, i.e., on 12th July, Stripe has shared a root cause analysis on the repeated degradation, as requested by the users. David Singleton, Stripe CTO describes the summary of API failures as  “two different database bugs and a configuration change interacted in an unforeseen way, causing a cascading failure across several critical services.” What was the cause of Stripe’s first API degradation? Three months ago, Stripe had upgraded to a new minor version and had performed the necessary testing to maintain a quality assured environment. This included executing a phased production rollout with the less critical as well as the increasingly critical clusters. Though it operated properly for the first three months, on the day of the event, it failed due to the presence of multiple stalled nodes. This occurred due to a shard, which was unable to elect a new primary state. [box type="shadow" align="" class="" width=""]“Stripe splits data by kind into different database clusters and by quantity into different shards. Each cluster has many shards, and each shard has multiple redundant nodes.”[/box] As the shard was used widely, its unavailability caused the compute resources for the API to starve and thus resulted in a severe degradation of the API services. The Stripe team detected  the failed election within a minute and started incident response within two minutes. The team forced the election of a new primary state, which led to restarting the database cluster. Thus, 27 minutes after the degradation, the Stripe API fully recovered. What caused Stripe’s API to degrade again? Once the Stripe’s API recovered, the team started investigating the root cause of the first degradation. They identified a code path in the new version of the database’s election protocol and decided to revert back to the previous known stable version for all the shards of the impacted cluster. This was deployed within four minutes. Until 21.14 UTC, the cluster was working fine. Later, the automated alerts fired indicating that some shards in the cluster were again unavailable, including the shard implicated in the first degradation. Though the symptoms appeared to be the same, the second degradation was caused due to a different reason. The prior reverted stable version interacted poorly with a configuration change to the production shards. Once the CPU starvation was observed, the Stripe team updated the production configuration and restored the affected shards. Once the shard was verified as healthy, the team began increasing the traffic back up, including prioritizing services as required by user-initiated API requests. Finally, Stripe’s API services were recovered at 22:47 UTC. Remedial actions taken The Stripe’s team has undertaken certain measures to ensure such degradation does not occur in the future An additional monitoring system has been implemented to alert whenever nodes stop reporting replication lag. Several changes have been introduced to prevent failures of individual shards from cascading across large fractions of API traffic. Further, Stripe will introduce more procedures and tooling to increase safety using which operators can make rapid configuration changes during incident response. Reactions to Stripe’s analysis of the API degradation has been mixed. Some users believe that the Stripe team should have focussed more on mitigating the error completely, rather than analysing the situation, at that moment. A Hacker News comment read, “In my experience customers deeply detest the idea of waiting around for a failure case to re-occur so that you can understand it better. When your customers are losing millions of dollars in the minutes you're down, mitigation would be the thing, and analysis can wait. All that is needed is enough forensic data so that testing in earnest to reproduce the condition in the lab can begin. Then get the customers back to working order pronto. 20 minutes seems like a lifetime if in fact they were concerned that the degradation could happen again at any time. 20 minutes seems like just enough time to follow a checklist of actions on capturing environmental conditions, gather a huddle to make a decision, document the change, and execute on it. Commendable actually, if that's what happened.” Few users appreciated Stripe’s analysis report. https://twitter.com/thinkdigitalco/status/1149767229392769024 Visit the Stripe website for a detailed timeline report. Twitter experienced major outage yesterday due to an internal configuration issue Facebook, Instagram and WhatsApp suffered a major outage yesterday; people had trouble uploading and sending media files Cloudflare suffers 2nd major internet outage in a week. This time due to globally deploying a rogue regex rule.

Google open-sourced Bidirectional Encoder Representations from Transformers (BERT) last Friday for NLP pre-training. Natural language processing (NLP) consists of topics like sentiment analysis, language translation, question answering, and other language-related tasks. Large datasets for NLP containing millions, or billions, of annotated training examples is scarce. Google says that with BERT, you can train your own state-of-the-art question answering system in 30 minutes on a single Cloud TPU, or a few hours using a single GPU. The source code built on top of TensorFlow. A number of pre-trained language representation models are also included. BERT features BERT improves on recent work in pre-training contextual representations. This includes semi-supervised sequence learning, generative pre-training, ELMo, and ULMFit. BERT is different from these models, it is the first deeply bidirectional, unsupervised language representation, pre-trained using only a plain text corpus - Wikipedia. Context-free models like word2vec generate a single word embedding representation for every word. Contextual models, on the other hand, generate a representation\ of each word based on the other words in the sentence. BERT is deeply bidirectional as it considers the previous and next words. Bidirectionality It is not possible to train bidirectional models by simply conditioning each word on words before and after it. Doing this would allow the word that’s being predicted to indirectly see itself in a multi-layer model. To solve this, Google researchers used a straightforward technique of masking out some words in the input and condition each word bidirectionally in order to predict the masked words. This idea is not new, but BERT is the first technique where it was successfully used to pre-train a deep neural network. Results On The Stanford Question Answering Dataset (SQuAD) v1.1, BERT achieved 93.2% F1 score surpassing the previous state-of-the-art score of 91.6% and human-level score of 91.2%. BERT also improves the state-of-the-art by 7.6% absolute on the very challenging GLUE benchmark, a set of 9 diverse Natural Language Understanding (NLU) tasks. For more details, visit the Google Blog. Intel AI Lab introduces NLP Architect Library FAT Conference 2018 Session 3: Fairness in Computer Vision and NLP Implement Named Entity Recognition (NER) using OpenNLP and Java

Facebook’s artificial intelligence research group - FAIR - just turned five. In a blog post, published yesterday, Facebook executives discussed the accomplishments FAIR has made over the last five years, and where it might be heading in the future. The team was formed with an aim to advance state-of-the-art AI via open research. FAIR has grown since its inception and now has labs in the USA and Europe. Their team has worked broadly with the open-source community and some of their papers have received awards. A significant part of FAIR research is around the keys to reasoning, prediction, planning, and unsupervised learning. These areas of investigation, in turn, require a better theoretical understanding of various fields related to artificial intelligence. They believe that long-term research explorations are necessary to unlock the full potential of artificial intelligence. Important milestones achieved by the FAIR team Memory networks FAIR developed a new class of machine learning models that can overcome the limitations in neural networks, i.e. long-term memory. These new models can remember previous interactions to answer general knowledge questions while keeping previous statements of a conversation in context. Self-supervised learning and generative models FAIR was fascinated by a new unsupervised learning method, Generative Adversarial Networks (GANs) in 2014 proposed by researchers from MILA at Université de Montréal. From 2015, FAIR published a series of papers that showcased the practicality of GANs. FAIR researchers and Facebook engineers have used adversarial training methods for a variety of applications, including long-term video prediction and creating graphic designs in fashion pieces. A scalable Text classification In 2016 FAIR built fastText, a framework for rapid text classification and learning word representations. In a 2017 paper, FAIR proposed a model that assigns vectors to “subword units” (sequences of 3 or 4 characters) rather than to whole words. This allowed the system to create representations for words that were not present in training data. This resulted in a model which could classify billions of words by learning from untrained words. Also, FastText is now available in 157 languages. Translation research FAIR developed a CNN-based neural machine translation architecture and published a paper on it in 2017. ‘Multi-hop’ CNNs are easier to train on limited data sets and can also better understand misspelled or abbreviated words; they’re designed to mimic the way humans translate sentences, by taking multiple glimpses at the sentence they are trying to translate. The results were a 9x increase in speed over RNNs while maintaining great accuracy. AI tools In 2015, the FAIR community open-sourced Torch deep learning modules to speed up training of larger neural nets. Torchnet was released in 2016 to build effective and reusable learning systems. Further,, they released Caffe2, a modular deep learning framework for mobile computing. After that, they collaborated with Microsoft and Amazon to launch ONNX, a common representation for neural networks. ONNX makes it simple to move between frameworks. A new benchmark for computer vision In 2017, FAIR researchers won the ‘International Conference on Computer Vision Best Paper’ for Mask R-CNN, which combines object detection with semantic segmentation. The paper stated: “Without bells and whistles, Mask R-CNN outperforms all existing, single-model entries on every task, including the COCO 2016 challenge winners.” Faster training and bigger data sets for Image Recognition Facebook’s Applied Machine Learning (AML) team discussed how they trained image recognition networks on large sets of public images with hashtags. The biggest dataset included 3.5 billion images and 17,000 hashtags. This was a breakthrough made possible by FAIR’s research on training speed. FAIR was able to train ImageNet, an order of a magnitude faster than the previous best. According to FAIR, “Our ultimate goal was to understand intelligence, to discover its fundamental principles, and to make machines significantly more intelligent” They continue to expand their research efforts into various areas such as developing machines that can acquire models of the real world with self-supervised learning, training machines to reason, and to plan, conceive complex sequences of actions. This is the reason why the community is also working on robotics, visual reasoning, and dialogue systems. Facebook AI research and NYU school of medicine announces new open-source AI models and MRI dataset as part of their FastMRI project The US Air Force lays groundwork towards artificial general intelligence based on hierarchical model of intelligence Is Anti-trust regulation coming to Facebook following fake news inquiry made by a global panel in the House of Commons, UK?

Facebook has recently been granted a patent titled “Selection and Presentation of News Stories Identifying External Content to Social Networking System Users” on July 31st, 2018. It aims to analyze the user data to curate a personalized news feed for the users. This will also include providing users with control over the kind of news they want to see. Facebook wants to add a Filter option in its news feed. This will make it easier for the users to find relevant news items. As per the patent application, “the news stories may be filtered based on filter criteria allowing a viewing user to more easily identify new stories of interest”. For instance, the filter can be added to view stories associated with either some other user or some news source. You can also add a keyword filter to get all the stories related to that specific keyword.  Facebook news feed filter tool   There are a lot of groups and pages on Facebook which helps reflect the user’s interests. The kind of content that the user posts also says a lot about his/her preferences. As there is a lot of user data present, Facebook automatically analyzes the user’s profile to optimize the news feed as per the choice of the user. There is also a ranking criterion involved when it comes to filtering news feed. The patent reads “news stories are scored and ranked based on their scores. News stories may be ranked based on the popularity of the news story among users of the social networking system. Popularity may be based on the number of views, likes, comments, shares or individual posts of the news story in the social networking system.” News stories can also be ranked based on the chronological order.   Facebook news feed filter tool patent Once Facebook is done analyzing the user profile, filtering the feed based on filter criteria, and ranking the stories based on the ranking criteria, a newly customized news feed will be generated and presented to the user. Facebook has been taking measures to curb fake news from its feed. The news filter tool is expected to help further. It will prevent irrelevant and fake news from occurring on users’ news feed as the users can choose to see news only from trusted resources. In fact, Facebook recently acquired Bloomsbury AI to fight fake news. Additionally, the latest news sources, accounts, groups, and pages will also be recommended to users based on data analyzed. With so much data floating around on Facebook feeds, this patent idea seems like a much-needed one. There are no details currently on when or if this feature will hit the Facebook feed. What do you think about Facebook’s news feed filter tool patent? Let us know in the comments below. Facebook launched new multiplayer AR games in Messenger Facebook launches a 6-part Machine Learning video series Facebook open sources Fizz, the new generation TLS 1.3 Library  

The next version of the Python library, pandas 0.24.0 will not have support for Python 2. pandas is a popular Python library widely used for data manipulation and data analysis. It is used in areas like numerical tables and time series data. Jeff Reback, pandas maintainer Tweeted on Wednesday: https://twitter.com/jreback/status/1080603676882935811 Many major Python libraries removing Python 2 support One of the first tools to drop support for Python 2 was ipython in 2017. This was followed by matplotlib and more recently NumPy. Other popular libraries like scikit-learn and SciPy will also be removing support for Python 2 this year. IDEs like Spyder and Pythran are also included in the list. Python 2 support ending in 2020 Core Python developers will stop supporting Python 2 no later than the year 2020. This move is to control fragmentation and save on workforce for maintaining Python 2. Python 2 will no longer receive any new features and all support for it will cease next year. As stated on the official website: “2.7 will receive bugfix support until January 1, 2020. After the last release, 2.7 will receive no support.” Python 2 support was about to end in 2015 itself but was extended by five years considering the user base. Users seem to welcome the change to move forward as a comment on Hacker new says: “Time to move forward. Python 2 is so 2010.” NumPy drops Python 2 support. Now you need Python 3.5 or later. Python governance vote results are here: The steering council model is the winner NYU and AWS introduce Deep Graph Library (DGL), a python package to build neural network graphs

Yesterday, IBM launched its Blockchain World Wire, a global blockchain network for cross-border payments that will make use of Stablecoin by U.S. dollars and cryptocurrency to make near real-time cross border financial transactions. It is based on distributed ledger technology (DLT) for regulated financial firms. IBM Blockchain World Wire is a real-time global payments network that works towards clearing and settling foreign exchange, cross border payments and remittances. Currently, this network can transfer funds to more than 50 countries using 47 digital coins backed by fiat currencies. According to IBM, World Wire is the first blockchain network of its kind to integrate payment messaging and clearing and settlement on a single unified network while allowing participants to dynamically choose from a variety of digital assets for settlement. According to a report by Cheddar, six international banks have signed letters of intent to issue their own Stablecoins backed by their national fiat currencies including Brazil’s Banco Bradesco, South Korea’s Bank Busan and the Philippines’ Rizal Commercial Banking Corporation on IBM’s Blockchain World Wire. Advantages of Blockchain World Wire Faster payment processing Blockchain World Wire provides simultaneous clearing and settlement and eliminates multiple parties processing transactions. Lower costs The World Wire comes with reduced capital requirements for cross-border transactions. Even the clearing costs have been lowered. Transparency The World Wire provides end-to-end transparency and one exchange fee between all currencies which makes it easier. If two financial institutions that are transacting agree upon using either a Stablecoin, central bank digital currency or another digital asset as the bridge asset between any two currencies then they will be provided with trade and important settlement instructions. The institutions can use their existing payment systems by connecting it to World Wire’s APIs in order to convert the first fiat currency into the digital asset. Further, the World Wire converts the digital asset into the second fiat currency, that completes the transaction. The transaction details are recorded onto an immutable blockchain for clearing purpose. Marie Wieck, General Manager, IBM Blockchain, said, “We’ve created a new type of payment network designed to accelerate remittances and transform cross-border payments to facilitate the movement of money in countries that need it most. By creating a network where financial institutions support multiple digital assets, we expect to spur innovation and improve financial inclusion worldwide.” To know more about this news, check out IBM’s official website. Google expands its Blockchain search tools, adds six new cryptocurrencies in BigQuery Public Datasets Blockchain governance and uses beyond finance – Carnegie Mellon university podcast Stable version of OpenZeppelin 2.0, a framework for smart blockchain contracts, released!

Microsoft has come out with a production system that performs text-to-speech (TTS) synthesis using deep neural networks. This new production system makes it hard for you to distinguish the voice of computers from human voice recordings. The Neural text-to-speech synthesis has significantly reduced the ‘listening fatigue’ when talking about interaction with AI systems. It enables the system with human-like, natural sounding voice, that makes the interaction with chatbots and virtual assistants more engaging. This neural-network powered text-to-speech system was demonstrated by the Microsoft team at the Microsoft Ignite conference in Orlando, Florida, this week. Additionally, Neural text-to-speech converts digital texts such as e-books into audiobooks. It also enhances in-car navigation systems. Deep Neural networks are great at overcoming the limits of traditional text-to-speech systems. Neural networks are very accurate in matching the patterns of stress and intonation in spoken language, called prosody. They’re also quite effective in synthesizing the units of speech into a computer voice. Neural TTS Traditional text-to-speech systems generally break down the prosody into separate linguistic analysis and acoustic prediction steps that get governed by independent models. This usually results in muffled, buzzy voice synthesis. Whereas, neural networks perform prosody prediction and voice synthesis simultaneously. This results in a more fluid and natural-sounding voice. Microsoft makes use of the computational power of Azure to offer real-time streaming. This makes it useful for situations such as interacting with a chatbot or virtual assistant. This TTS capability is served in the Azure Kubernetes Service to ensure high scalability and availability. Only the preview of the text-to-speech service is available currently. The preview comes with two pre-built neural text-to-speech voices in English – Jessa, and Guy.  Microsoft will be making more languages available soon. It will also be offering customization services in 49 languages for customers wanting to build branded voices optimized for their specific needs. For more information, check out the official Microsoft Blog post. Microsoft acquires AI startup Lobe, a no code visual interface tool to build deep learning models easily DoWhy: Microsoft’s new python library for causal inference Say hello to FASTER: a new key-value store for large state management by Microsoft