Data Science News Week in Review: November 25th
Every week we’re bringing data science news to you… some highlights include reviewing Facebook’s DeepFovea AI, IBM’s Consortium for Sequencing the Food Supply Chain, and more. Read on:
[Related Article: The Best Machine Learning Research of September 2019]
Facebook’s DeepFovea AI promises power-efficient VR foveated rendering
In a paper released on November 18th, Facebook introduced a human-like “peripheral vision” GAN. The system works to reproduce images while only knowing 10% of the source imagery.
Jigsaw releases data set to help develop AI that detects toxic comments
In April, Jigsaw started a Kaggle competition by releasing part of a large labeled data set of toxic comments. Recently, they’ve announced they’re putting out the full set, complete with labels from over 9000 human “toxicity raters,” all in hopes to create AI that can accurately detect toxic comments and predict how toxic people will see it as.
IBM and the Unitary Fund Unite for Open Source Projects for Quantum Computing
On November 20th, IBM announced they were partnering with the Unitary Fund to provide special funding for grants and priority access to some IBM Q systems, so IBM can increase support for their community of Quantum enthusiasts. They anticipate being able to build even more open-source software and publicly accessible systems.
OpenAI Safety Gym enhances reinforcement learning
In order to train reinforcement learning systems that are better and safer before being deployed in risky human environments, OpenAI has released an open-source “Safety Gym.” The gym is a set of environments that attempts to quantify risk and allows for agents to learn from their mistakes before it puts humans at risk; at the gym, agents have three possible scenarios, different difficulty ratings, and tasks to complete.
Designing conversational experiences with sentiment analysis in Amazon Lex
Amazon just released a tutorial on building a bot, adding logic to make responses based on user sentiment (understanding how users feel and crafting better answers), and configuring hand-over to an agent to continue the conversation. You can now do all this natively within Amazon Lex.
What’s in your food? A new technology platform shows early signs of promise
IBM Research released a paper, published November 19, describing their proof of concept program, designed to detect expected and unexpected ingredients within your food. Their approach involves evaluating the DNA and RNA of food and comparing that to a database of thousands of plant and animal genomes.
NVIDIA and Microsoft Team Up to Aid AI Startups
On November 20, NVIDIA and Microsoft announced they have joined forces and opened their start-up resources to some of the most promising young companies. Now, members of Microsoft for Startups will also have access to all the tools and resources within NVIDIA Inception, and vice versa.
Google Cloud tackles adoption roadblocks with AI explainability toolkit
In a recent Whitepaper, Google Cloud dove deeper into their AI Explanations product. They discuss some of the key issues in creating explainable AI, and offer some answers with an in-depth toolkit based on their years of research.
Apple’s AI can predict cognitive impairment from iOS app usage
In a recent paper, a new AI system from Apple has been able to predict and detect early signs of cognitive impairment — namely dementia — from users’ app usage. The system was trained on 113 older adults, 31 diagnosed with impairment.
[Related Article: Apple Pay Card’s Credit Determining AI: Gender Biased?]
Google details DeepMind AI’s role in Play Store app recommendations
DeeMind has recently improved on Google’s current Play Store recommendation system, in hopes to create a more personalized experience for each user. They’ve done this using an LSTM model that can recommend based on multiple objectives, with specific attempts at de-biasing their candidate generator model.
