Week Ending 5.3.2020
RESEARCH WATCH: 5.3.2020
Over the past week, 97 new papers were published in "Computer Science - Artificial Intelligence".
The paper discussed most in the news over the past week was by a team at Google: "Chip Placement with Deep Reinforcement Learning" by Azalia Mirhoseini et al (Apr 2020), which was referenced 9 times, including in the article AI Designs Computer Chips for More Powerful AI in Discover Magazine. The paper author, Jeff Dean (Google), was quoted saying "It’s a multi-week process to actually go from the design you want to actually having it physically laid out on a chip with the right constraints in area and power and wire length and meeting all the design roles or whatever fabrication process you’re doing". The paper got social media traction with 611 shares. The researchers present a learning - based approach to chip placement, one of the most complex and time - consuming stages of the chip design process. On Twitter, @theshawwn posted ""Whereas existing baselines require human experts in the loop and take several weeks to generate, our method can generate placements in under six hours that outperform or match their manually designed counterparts." It's how TPUs are designed, apparently. Neat!".
Leading researcher Yoshua Bengio (Université de Montréal) came out with "Learning To Navigate The Synthetically Accessible Chemical Space Using Reinforcement Learning" @saikrishna_gvs tweeted "Indeed! It would be interesting to see if this approach can be used to develop #antivirals for #COVID19. Apart from achieving state of the art results, this approach has an added advantage that every #drug generated is actually synthesizable".
The paper shared the most on social media this week is by a team at Wright State University: "Explainable Deep Learning: A Field Guide for the Uninitiated" by Ning Xie et al (Apr 2020) with 576 shares. @gastronomy (Tanat Tonguthaisri) tweeted "> Deep neural network (DNN) is an indispensable machine learning tool for achieving human-level performance on many learning tasks. Yet, due to its blac".
This week was active for "Computer Science - Computer Vision and Pattern Recognition", with 212 new papers.
The paper discussed most in the news over the past week was by a team at University of Waterloo: "COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images" by Linda Wang et al (Mar 2020), which was referenced 22 times, including in the article How deep learning can improve how we conduct scientific research in The Next Web. The paper author, Alexander Wong (University of Waterloo), was quoted saying "The hope is that the AI can help radiologists to more rapidly and accurately differentiate between COVID-19 infections and other forms of infections (especially important since flus are prevalent still this time of year), and more importantly, reduce the burden for radiologists but enabling other front-line health workers with less expertise to better make diagnosis". The paper got social media traction with 149 shares. A user, @hoangdta, tweeted "This AI acquired 100% sensitivity on Covid-19 cases detection ! The dataset is quite small, I would love to see scientists join force to tackle this problem. It would be HUGE if we can exploit the power of deep learning in the battle against Covid-19".
Leading researcher Yoshua Bengio (Université de Montréal) published "DiVA: Diverse Visual Feature Aggregation for Deep Metric Learning".
The paper shared the most on social media this week is "Consistent Video Depth Estimation" by Xuan Luo et al (Apr 2020) with 873 shares. @theowatson (Theodore Watson) tweeted "Wow. This is a huge leap forward for depth from RGB. NN estimated but then feeding back projection errors into the training. Via".
This week was active for "Computer Science - Computers and Society", with 39 new papers.
The paper discussed most in the news over the past week was by a team at Massachusetts Institute of Technology: "Apps Gone Rogue: Maintaining Personal Privacy in an Epidemic" by Ramesh Raskar et al (Mar 2020), which was referenced 17 times, including in the article The Importance of Equity in Contact Tracing in ITSecurityNews.info. The paper author, Ramesh Raskar (Massachusetts Institute of Technology), was quoted saying "We are dedicated to privacy-first solutions — user location and contact history should never leave a user’s phone without direct consent". The paper got social media traction with 96 shares. On Twitter, @dhmackenzie observed "Hi James, it's probably worth you taking a look first at this paper from the Safe Paths team, which explains some of the thinking. Safe Paths is a Privacy-First contact tracing app. Privacy has absolutely been a key consideration from day 1".
This week was active for "Computer Science - Human-Computer Interaction", with 34 new papers.
The paper shared the most on social media this week is by a team at Georgia Institute of Technology: "CNN Explainer: Learning Convolutional Neural Networks with Interactive Visualization" by Zijie J. Wang et al (Apr 2020) with 426 shares. @nicolangnl (Nico Lang) tweeted "First there was the playground for neural networks: Now there is this great tool for learning and teaching the concept of CNNs. Well done!".
This week was very active for "Computer Science - Learning", with 387 new papers.
The paper discussed most in the news over the past week was by a team at University of Waterloo: "COVID-Net: A Tailored Deep Convolutional Neural Network Design for Detection of COVID-19 Cases from Chest X-Ray Images" by Linda Wang et al (Mar 2020)
Leading researcher Yoshua Bengio (Université de Montréal) published "Learning To Navigate The Synthetically Accessible Chemical Space Using Reinforcement Learning" @saikrishna_gvs tweeted "Indeed! It would be interesting to see if this approach can be used to develop #antivirals for #COVID19. Apart from achieving state of the art results, this approach has an added advantage that every #drug generated is actually synthesizable".
The paper shared the most on social media this week is by a team at Salesforce: "The AI Economist: Improving Equality and Productivity with AI-Driven Tax Policies" by Stephan Zheng et al (Apr 2020) with 1062 shares. The investigators train social planners that discover tax policies in dynamic economies that can effectively trade - off economic equality and productivity. @lrvarshney (Lav Varshney) tweeted "you have to have a good model for howthink and how they're going to respond to the changes you're making" but using AI is a step in this direction".
The most influential Twitter user discussing papers is Judea Pearl who shared "Dark, Beyond Deep: A Paradigm Shift to Cognitive AI with Humanlike Common Sense" by Yixin Zhu et al (Apr 2020) and said: "Good catch. This link should work I hope".
Over the past week, 16 new papers were published in "Computer Science - Multiagent Systems".
Over the past week, 28 new papers were published in "Computer Science - Neural and Evolutionary Computing".
The paper discussed most in the news over the past week was "The Cost of Training NLP Models: A Concise Overview" by Or Sharir et al (Apr 2020), which was referenced 3 times, including in the article AI21 Labs Asks: How Much Does It Cost to Train NLP Models? in SyncedReview.com. The paper got social media traction with 135 shares. A Twitter user, @billiout, said "According to the following study from training a single BIG NLP model can cost about $10k. That's unacceptable! Both for the environmental burden as well as for the independent researchers who don't have access to these resources.#NLProc".
This week was active for "Computer Science - Robotics", with 47 new papers.
The paper discussed most in the news over the past week was by a team at Google: "Learning Agile Robotic Locomotion Skills by Imitating Animals" by Xue Bin Peng et al (Apr 2020), which was referenced 2 times, including in the article A system to reproduce different animal locomotion skills in robots in Tech Xplore. The paper author, Jason Peng, was quoted saying "The most exciting result for us was that the same underlying method can learn a pretty large variety of skills ranging from walking to dynamic hopping and turning and all of the skills learned in simulation can also be transferred to a real robot". The paper got social media traction with 153 shares. A user, @svlevine, tweeted "Robots that learn by imitating animals! Check out how RL, imitation, and real-world adaptation can enable legged robots to perform a wide range of agile skills from mocap of dogs and artist-created animations. w/ et al".
Leading researcher Sergey Levine (University of California, Berkeley) published "Emergent Real-World Robotic Skills via Unsupervised Off-Policy Reinforcement Learning". This paper was also shared the most on social media with 43 tweets.