Week Ending 5.22.2022

 

RESEARCH WATCH: 5.22.2022

SPONSORED BY

 

This week was extremely active for "Computer Science - Artificial Intelligence", with 265 new papers.

  • The paper discussed most in the news over the past week was by a team at DeepMind: "A Generalist Agent" by Scott Reed et al (May 2022), which was referenced 27 times, including in the article Google's DeepMind says it is close to achieving 'human-level' artificial intelligence in Mail Online UK. The paper author, Nando de Freitas (DeepMind), was quoted saying "the game is over". The paper got social media traction with 187 shares. A Twitter user, @HochreiterSepp, posted "ArXiv Gato: a single generalist policy. Can play Atari, caption images, chat, stack blocks. Output determined by context (text, joint torques, buttons). 1.2B para. decoder-only transformer with 24 layers. Impressive results on control, robotics, language".

  • Leading researcher Yoshua Bengio (Université de Montréal) came out with "FedILC: Weighted Geometric Mean and Invariant Gradient Covariance for Federated Learning on Non-IID Data" @gastronomy tweeted "> Federated learning is a distributed machine learning approach which enables a shared server model to learn".

  • The paper shared the most on social media this week is "The Primacy Bias in Deep Reinforcement Learning" by Evgenii Nikishin et al (May 2022) with 101 shares. @jm_alexia (Alexia Jolicoeur-Martineau) tweeted "Super neat and simple idea. I wouldn't be surprised if it also helps non-RL models. It's a bit like going to sleep and retrying again fresh the next day. I wonder if sleep could be beneficial because of similar reasons".

  • The most influential Twitter user discussing papers is McKenna (¤, ¤) who shared "Why and How zk-SNARK Works" by Maksym Petkus (Jun 2019) and said: "Brilliant paper on ZKSnarks for anyone who wants an introduction to zero-knowledge cryptography". Note that this paper was published about two years ago.

This week was active for "Computer Science - Computer Vision and Pattern Recognition", with 242 new papers.

  • The paper discussed most in the news over the past week was "Detecting Deepfakes with Self-Blended Images" by Kaede Shiohara et al (Apr 2022), which was referenced 9 times, including in the article Seeing is deceiving in Mirage News. The paper got social media traction with 23 shares. The researchers called self - blended images (SBIs) to detect deepfakes. On Twitter, @srvmshr commented "CVPR oral paper from our lab. First one for & hopefully more to come", while @summarizedml commented "A novel synthetic training data method for detecting deepfakes. 📄".

  • Leading researcher Sergey Levine (University of California, Berkeley) published "Planning to Practice: Efficient Online Fine-Tuning by Composing Goals in Latent Space" The authors propose Planning to Practice (PTP), a method that makes it practical to train goal - conditioned policies for long - horizon tasks that require multiple distinct types of interactions to solve. @HochreiterSepp tweeted "ArXiv Training goal-conditioned policies that reach goals on command. 1 Decompose problem by a planner that suggests subgoals. 2 Hybrid approach pretrains subgoal generator and policy by offline reinforcement learning, then fine-tunes policy online".

  • The paper shared the most on social media this week is "AvatarCLIP: Zero-Shot Text-Driven Generation and Animation of 3D Avatars" by Fangzhou Hong et al (May 2022) with 125 shares. @summarizedml (SummarizedML) tweeted "A novel zero-shot text-driven framework for 3D avatar generation and animation. 📄".

This week was active for "Computer Science - Computers and Society", with 38 new papers.

This week was very active for "Computer Science - Human-Computer Interaction", with 41 new papers.

  • The paper discussed most in the news over the past week was "Productivity Assessment of Neural Code Completion" by Albert Ziegler et al (May 2022), which was referenced 1 time, including in the article GitHub Publishes Productivity Assessment of Neural Code Completion Systems in SyncedReview.com. The paper got social media traction with 25 shares. On Twitter, @arXiv_cs_CL_ja said "Productivity Assessment of Neural Code Completion Comment: To appear in: Proceedings of the 6th ACM SIGPLAN International Symposium on Machine Programming (MAPS '22), June 13, 2022 ニューラルコードの合成は、スニペットの生成が人間のソフトウェ", while @summarizedml observed "We find that the rate with which shown suggestions are accepted, rather fixmethan more specific metrics regarding the persistence of completions in the code fixmeover 📄".

This week was extremely active for "Computer Science - Learning", with 523 new papers.

Over the past week, 11 new papers were published in "Computer Science - Multiagent Systems".

  • The paper discussed most in the news over the past week was "Learning Eco-Driving Strategies at Signalized Intersections" by Vindula Jayawardana et al (Apr 2022), which was referenced 14 times, including in the article On the road to cleaner, greener, and faster driving in Science Daily. The paper author, Cathy Wu (University of Delaware), was quoted saying "This is a really interesting place to intervene. No one’s life is better because they were stuck at an intersection. With a lot of other climate change interventions, there is a quality-of-life difference that is expected, so there is a barrier to entry there. Here, the barrier is much lower". The paper was shared 1 time in social media. The authors propose a reinforcement learning (RL) approach to learn effective eco - driving control strategies. On Twitter, @summarizedml said "A reinforcementlearning approach to learn effective eco-driving control strategies at intersections. 📄".

Over the past week, 29 new papers were published in "Computer Science - Neural and Evolutionary Computing".

This week was very active for "Computer Science - Robotics", with 86 new papers.

  • The paper discussed most in the news over the past week was by a team at DeepMind: "A Generalist Agent" by Scott Reed et al (May 2022)

  • Leading researcher Sergey Levine (University of California, Berkeley) published "Planning to Practice: Efficient Online Fine-Tuning by Composing Goals in Latent Space" The investigators propose Planning to Practice (PTP), a method that makes it practical to train goal - conditioned policies for long - horizon tasks that require multiple distinct types of interactions to solve. @HochreiterSepp tweeted "ArXiv Training goal-conditioned policies that reach goals on command. 1 Decompose problem by a planner that suggests subgoals. 2 Hybrid approach pretrains subgoal generator and policy by offline reinforcement learning, then fine-tunes policy online". This paper was also shared the most on social media with 22 tweets. @HochreiterSepp (Sepp Hochreiter) tweeted "ArXiv Training goal-conditioned policies that reach goals on command. 1 Decompose problem by a planner that suggests subgoals. 2 Hybrid approach pretrains subgoal generator and policy by offline reinforcement learning, then fine-tunes policy online".


EYE ON A.I. GETS READERS UP TO DATE ON THE LATEST FUNDING NEWS AND RELATED ISSUES. SUBSCRIBE FOR THE WEEKLY NEWSLETTER.