Nghia Ticy

Sudeep Dasari wrote an excellent blog post on our work on RoboNet, cross-posted on the SAIL blog: http://sail.stanford.edu/blog/robonet/  We imagine a future where robots share data *across* research labs, just like the rest of machine learning. https://twitter.com/svlevine/status/1199220226807590912 …pic.twitter.com/c77Sxl0Vqt

PyTorch-Struct (v0.4 https://github.com/harvardnlp/pytorch-struct …). Library `genbmm` for speed/memory improvements. New CUDA kernels for log-space matmul and banded sparse matrices (https://github.com/harvardnlp/genbmm/blob/master/notebooks/Sparse.ipynb …)pic.twitter.com/5ty4kcds5f

Can adversarial examples improve image recognition? Check out our recent work: AdvProp, achieving ImageNet top-1 accuracy 85.5% (no extra data) with adversarial examples! Arxiv: https://arxiv.org/abs/1911.09665  Checkpoints: https://git.io/JeopW pic.twitter.com/bAu054LGt2

Fourier Transform Image Filter Tool http://www.ejectamenta.com/Imaging-Experiments/fourierimagefiltering.html …pic.twitter.com/AjjLaelcR7

EfficientDet: a new family of efficient object detectors. It is based on EfficientNet, and many times more efficient than state of art models. Link: https://arxiv.org/abs/1911.09070  Code: coming soonpic.twitter.com/2KYabAnpLL

We're releasing Safety Gym, environments and tools to evaluate reinforcement learning with safety constraints: https://openai.com/blog/safety-gym/ … Aims to ultimately help agents satisfy real-world safety requirements while training (eg not driving off a cliff, not writing abusive content).pic.twitter.com/VTwS4KoFS1

How to combine planning and RL? With LEAP, we use RL + LVMs to learn representations and skills, then plan over high-level subgoals. This allows us to plan with image observations, and solve (very) long-horizon tasks! w/ S. Nasiriany, @Vitchyr, S. Lin https://sites.google.com/view/goal-planning …pic.twitter.com/QTg09vthHp

Topology-disturbing objects: a new class of 3D optical illusion https://www.tandfonline.com/doi/full/10.1080/17513472.2017.1368133 …pic.twitter.com/Zu8zYtbcx6

We are releasing DeepFovea, a new state of the art in foveated rendering for augmented and virtual reality using an AI-powered system. It is the first practical video generative adversarial network with sparse input. https://ai.facebook.com/blog/deepfovea-using-deep-learning-for-foveated-reconstruction-in-ar-vr …pic.twitter.com/DcMbHbDEvD

*New paper* RandAugment: a new data augmentation. Better & simpler than AutoAugment. Main idea is to select transformations at random, and tune their magnitude. It achieves 85.0% top-1 on ImageNet. Paper: https://arxiv.org/abs/1909.13719  Code: https://git.io/Jeopl pic.twitter.com/equmk59K2i

Primes and Knots https://www.youtube.com/watch?v=jvoYgNYKyk0 …pic.twitter.com/Huqj2kpHLy

Learn about ReAgent, a full suite of open source tools designed to streamline the process of building ML models that make and rely on decisions. https://ai.facebook.com/blog/open-sourcing-reagent-a-platform-for-reasoning-systems/ …pic.twitter.com/hobFrg70Lo

NaSC: an app where you do maths like a normal person. It lets you type whatever you want and smartly figures out what is math and spits out an answer on the right pane. Then you can plug those answers in to future equations https://parnold-x.github.io/nasc/  & https://github.com/parnold-x/nasc pic.twitter.com/OHyxYPJ2qR

As we issue our fourth Community Standards Enforcement Report, we are sharing how recent advances in AI have helped us keep people safe from harmful content across our platforms. https://ai.facebook.com/blog/community-standards-report …pic.twitter.com/BTV08YwcBT

Going Critical - the way things move and spread, somewhat chaotically, across a network https://www.meltingasphalt.com/interactive/going-critical/ …pic.twitter.com/b9YDlq9oA4

Simulated worlds are like onions - Breaking down Biogrid’s GPU-accelerated world simulation https://medium.com/@haversine02/simulated-worlds-are-like-onions-dfa18cab3d85 … & @BioGridGamepic.twitter.com/zr5MxCDDuK