Rajat Gupta

New paper: Towards a Human-like Open-Domain Chatbot. Key takeaways: 1. "Perplexity is all a chatbot needs" ;) 2. We're getting closer to a high-quality chatbot that can chat about anything Paper: https://arxiv.org/abs/2001.09977  Blog: https://ai.googleblog.com/2020/01/towards-conversational-agent-that-can.html …pic.twitter.com/5SOBa58qx3

BigGAN samples are famously photo-realistic but limited in diversity for some classes. Slightly modifying only the class embeddings (network unchanged) can reduce the diversity gap by ~50%! Work with Long Mai and led by fantastic @MkQili!! Paper & video: http://anhnguyen.me/project/biggan-am/ …pic.twitter.com/MhIOxCepV5

Very happy to share our latest work accepted at #ICRL2020: we prove that a Self-Attention layer can express any CNN layer. 1/5 Paper: https://openreview.net/pdf?id=HJlnC1rKPB … Interactive website : https://epfml.github.io/attention-cnn/  Code: https://github.com/epfml/attention-cnn … Blog: http://jbcordonnier.com/posts/attention-cnn/ …pic.twitter.com/X1rNS1JvPt

Drifting in an autonomous vehicle. Uses rotation rate of the vehicle’s velocity vector to track the path, while yaw acceleration is used to stabilize sideslip. Could help autonomous vehicles in emergencies. Fun results. https://ddl.stanford.edu/sites/g/files/sbiybj9456/f/marty_avec2018_fullpaper.pdf …pic.twitter.com/tSF3R1jTHu

Still slowly making my way through this year's NeurIPS talks. Esp like to stumble by good talks from slightly different areas, e.g. tonight liked "ML Meets Single-Cell Biology" https://slideslive.com/38921722/machine-learning-meets-singlecell-biology-insights-and-challenges … Incredible that we're mapping out cell state markov chains for tissues.pic.twitter.com/eJcCC2AGDB

i can't make this uppic.twitter.com/ECxLVtQgFz

BackPACK: Packing more into backprop "we introduce BackPACK, an efficient framework built on top of PyTorch, that extends the backpropagation algorithm to extract additional information from first- and second-order derivatives" https://arxiv.org/abs/1912.10985  https://f-dangel.github.io/backpack/ pic.twitter.com/sX9Rp6We1w

Can we simulate both a user and system and learn to autocomplete in an unsupervised way? Yes! We frame the autocomplete task as a cooperative communication game. #StanfordNLP Talk: Dec 14 9:45-10AM (West 118) #NeurIPS2019 https://arxiv.org/abs/1911.06964 pic.twitter.com/w73UKQabhe

Our new paper, Deep Learning for Symbolic Mathematics, is now on arXiv https://arxiv.org/abs/1912.01412  We added *a lot* of new results compared to the original submission. With @f_charton (1/7)pic.twitter.com/GrhQRT5WRW

Excited to share our work on Contrastive Learning of Structured World Models! C-SWMs learn object-factorized models & discover objects without supervision, using a simple loss inspired by work on graph embeddings Paper: http://arxiv.org/abs/1911.12247  Code: https://github.com/tkipf/c-swm  1/5pic.twitter.com/2n38bhOaKc

My most recent @GoogleAI residency project is out: "Hamiltonian Neural Networks" Blog: https://greydanus.github.io/2019/05/15/hamiltonian-nns/ … Paper: https://arxiv.org/abs/1906.01563  Starting from noisy (pixel) data, we can learn _exact_ conservation of energy-like quantities.pic.twitter.com/SY2jZBr5an

I've been working with @GuggerSylvain for a long time to create something that combines the best of @ProjectJupyter Notebooks with the best of traditional software development approaches. It's called nbdev. We're releasing it today as open source. https://www.fast.ai/2019/12/02/nbdev/ … 1/pic.twitter.com/u9x26L4uRf

Large language models are starting to capture larger swaths of English Grammar, and several of us at NYU have gotten interested in trying to get a broad overview of where models are succeeding and failing. [new dataset alert; thread]pic.twitter.com/9DcqCEAtUR

In our recent collaboration with @berkeley_ai we show how to generate realistic complex scenes from scratch! While the problem is extremely challenging, we show how to achieve SOTA in unconditional generation and improve conditional generation using SPADE http://arxiv.org/pdf/1911.11357.pdf …pic.twitter.com/fVXkEvMdkt

Introducing the SHA-RNN :) - Read alternative history as a research genre - Learn of the terrifying tokenization attack that leaves language models perplexed - Get near SotA results on enwik8 in hours on a lone GPU No Sesame Street or Transformers allowed. https://arxiv.org/abs/1911.11423 pic.twitter.com/RN5TPZ3xWH