Week Ending 2.6.2022

 

RESEARCH WATCH: 2.6.2022

 

This week was active for "Computer Science", with 1,151 new papers.

  • The paper discussed most in the news over the past week was "DRAWNAPART: A Device Identification Technique based on Remote GPU Fingerprinting" by Tomer Laor (Ben-Gurion University of the Negev) et al (Jan 2022), which was referenced 37 times, including in the article The Next Graphics Card Crisis Could Be The Most Worrying Yet in Forbes.com. The paper got social media traction with 100 shares. The authors report on a new technique that can significantly extend the tracking time of fingerprint - based tracking methods. On Twitter, @FobbeSean observed "New and hard to evade: Tracking browsers via unique fingerprinting of GPUs with small WebGL applications (Laor et al 2022): Includes a discussion of countermeasures, Tor Browser does not appear to be affected".

  • Leading researcher Yoshua Bengio (Université de Montréal) came out with "Trajectory Balance: Improved Credit Assignment in GFlowNets".

  • The paper shared the most on social media this week is by a team at University of Wisconsin-Madison: "VOS: Learning What You Dont Know by Virtual Outlier Synthesis" by Xuefeng Du et al (Feb 2022) with 432 shares. The researchers present VOS, a novel framework for OOD detection by adaptively synthesizing virtual outliers that can meaningfully regularize the models decision boundary during training. @martin_gorner (Martin Görner) tweeted "This is sweet 🥧 ! Finally a solid way of of teaching a neural network to know what it does not know. (OOD = Out Of Domain, i.e. not one of the classes in the training data.) Congrats".

This week was very active for "Computer Science - Artificial Intelligence", with 172 new papers.

  • The paper discussed most in the news over the past week was by a team at Massachusetts Institute of Technology: "Natural Language Descriptions of Deep Visual Features" by Evan Hernandez et al (Jan 2022), which was referenced 9 times, including in the article Demystifying Machine-Learning Systems: Automatically Describing Neural Network Components in Natural Language in SciTechDaily. The paper author, Evan Hernandez, was quoted saying "In a neural network that is trained to classify images, there are going to be tons of different neurons that detect dogs. But there are lots of different types of dogs and lots of different parts of dogs. So even though ‘dog’ might be an accurate description of a lot of these neurons, it is not very informative. We want descriptions that are very specific to what that neuron is doing. This isn’t just dogs; this is the left side of ears on German shepherds". The paper got social media traction with 35 shares. A user, @cogconfluence, tweeted "Our #ICLR2022 work is out! 🎉 MILAN (mutual-information-guided linguistic annotation of neurons) describes in natural language what individual units in neural networks do. Paper: w/ Teona Bagashvili, Antonio Torralba, &".

  • Leading researcher Yoshua Bengio (Université de Montréal) came out with "Adaptive Discrete Communication Bottlenecks with Dynamic Vector Quantization" @summarizedml tweeted "Vector Quantization (VQ) is a method for discretizing latent representations and has become a major part of the deep learning toolkit 📄".

  • The paper shared the most on social media this week is by a team at Carnegie Mellon University: "Tutorial on amortized optimization for learning to optimize over continuous domains" by Brandon Amos (Feb 2022) with 162 shares. @EugeneVinitsky (Eugene Vinitsky) tweeted "If you hear me raving about how RL is amortized optimization, know that this excellent tutorial is the source".

Over the past week, 189 new papers were published in "Computer Science - Computer Vision and Pattern Recognition".

  • The paper discussed most in the news over the past week was by a team at North Carolina State University: "Learning Auxiliary Monocular Contexts Helps Monocular 3D Object Detection" by Xianpeng Liu et al (Dec 2021), which was referenced 11 times, including in the article NC State University develops new AI method to boost computer vision in Autonomous Vehicle International. The paper author, Tianfu Wu (North Carolina State University), was quoted saying "We live in a 3D world, but when you take a picture, it records that world in a 2D image". The paper got social media traction with 5 shares. The authors propose a simple yet effective formulation for monocular 3D object detection without exploiting any extra information. A user, @summarizedml, tweeted "Monocular 3D object detection without exploiting any extra information . 📄".

  • Leading researcher Sergey Levine (University of California, Berkeley) came out with "Fully Online Meta-Learning Without Task Boundaries" @FinSentim tweeted "Jathushan Rajasegaran et al. including study how meta-learning can be applied to tackle online problems of this nature, simultaneously adapting to changing tasks and input distributions and meta-training the model in order to adapt more quickly in the future. #ML".

  • The paper shared the most on social media this week is by a team at University of Wisconsin-Madison: "VOS: Learning What You Dont Know by Virtual Outlier Synthesis" by Xuefeng Du et al (Feb 2022)

This week was active for "Computer Science - Computers and Society", with 34 new papers.

  • The paper discussed most in the news over the past week was "Health Advertising on Facebook: Privacy & Policy Considerations" by Andrea Downing et al (Jan 2022), which was referenced 4 times, including in the article Health Sites Let Ads Track Visitors Without Telling Them in Wired News. The paper author, Andrea Matwyshyn, was quoted saying "It’s entirely expected from my perspective that findings like this keep coming up for the category that I call 'health-ish' data that does not cleanly fall under the limited privacy protections that currently exist in US laws". The paper got social media traction with 80 shares. The researchers analyzed content and marketing tactics of digital medicine companies to evaluate various types of cross site tracking middleware used to extract health information from users without permission. A Twitter user, @mattsmear, said "“Privacy Zuckering” happens when a user is tricked into publicly sharing more information than a user really intended to share. When... employed to elicit public data from patient populations online, one might consider the sensitivity of health data involved." 👇💣👇💣👇💣👇💣".

  • The paper shared the most on social media this week is "Measuring Disparate Outcomes of Content Recommendation Algorithms with Distributional Inequality Metrics" by Tomo Lazovich et al (Feb 2022) with 133 shares. @rasmus_kleis (Rasmus Kleis Nielsen) tweeted ""The top 1% of authors get almost 80% of all Tweet impressions in our dataset" et al finds in really interesting paper May suggest big accounts would have leverage w/platform (pay us!), but do they? If 1% was of 200m+ daily active users that'd be 2 million accounts!".

This week was very active for "Computer Science - Human-Computer Interaction", with 37 new papers.

This week was extremely active for "Computer Science - Learning", with 531 new papers.

Over the past week, 12 new papers were published in "Computer Science - Multiagent Systems".

Over the past week, 20 new papers were published in "Computer Science - Neural and Evolutionary Computing".

This week was very active for "Computer Science - Robotics", with 71 new papers.

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

 
Artificial Intelligence, Research WatchCraig SmithComment