Most ML folks I know have @AnthropicAI's Toy Models of Superposition paper on their reading list, but too few have read it.
— Emmanuel Ameisen (@mlpowered) October 19, 2022
It is one of the most interesting interpretability paper I've read in a while and it can benefit anyone using deep learning.
Here are my takeaways! pic.twitter.com/XrQ3Pp6b6b
The @runwayml Stable Diffusion Inpainting model is now available in the latest version of the @huggingface 🧨diffusers library (v0.6.0) 🎉
— AK (@_akhaliq) October 19, 2022
Github: https://t.co/cVVfWMwZgB
model: https://t.co/RWuqRbt2qj pic.twitter.com/SfHmUYFDAd
Token Merging: Your ViT But Faster
— AK (@_akhaliq) October 19, 2022
abs: https://t.co/14DPSqst90
github: https://t.co/oJMk2nSuvw pic.twitter.com/4hJKEZovEp
Differentially Private Diffusion Models
— AK (@_akhaliq) October 19, 2022
abs: https://t.co/IW2WU2ega5
project page: https://t.co/3gxE40jRu6 pic.twitter.com/HAVldjJDqG
UniTune: Text-Driven Image Editing by Fine Tuning an Image Generation Model on a Single Image
— AK (@_akhaliq) October 19, 2022
abs: https://t.co/AU8m80CjQD pic.twitter.com/A7UD1fLM9B
Awesome Computer Vision
— Jean de Nyandwi (@Jeande_d) October 18, 2022
A curated list of computer vision resources: books, courses, papers, software, datasets, pre-trained models, tutorials, talks, blogs, and songs :-)https://t.co/twm6btvY24 pic.twitter.com/1MrFrHscoF
Model Criticism for Long-Form Text Generation (https://t.co/V59Z94D9Oy w/ Yuntian Deng & @volokuleshov)
— Sasha Rush (@srush_nlp) October 18, 2022
Researchers have observed that LM likelihood doesn't directly correlate with the emergence of long-form coherence. We quantify this through a model criticism framework. 1/
Bayesian recipe of the day 🧑🍳: Take a simple latent variable model where the latents are Gaussian and the observed variables are linear Gaussian. Fit the linear projection parameters by maximum likelihood estimation. Then, posterior inference gives you (Probabilistic) PCA! 🤯 pic.twitter.com/lTtKLhvVO0
— Gilles Louppe (@glouppe) October 18, 2022
You Only Live Once: Single-Life Reinforcement Learning
— AK (@_akhaliq) October 18, 2022
abs: https://t.co/PG3tqv89DA pic.twitter.com/WrwzfA2Bg8
gghighlight 0.4.0 is now on CRAN with this experimental feature! You can use `line_label_type` arg. The details can be found athttps://t.co/qonDNQ5gni #rstats https://t.co/OrtSjgUl5M
— Hiroaki Yutani (@yutannihilat_en) October 18, 2022
Table-To-Text generation and pre-training with TabT5
— AK (@_akhaliq) October 18, 2022
abs: https://t.co/YyZ9hUexVx pic.twitter.com/lkjdnMo1He
This is compelling. Didn't see anything in the paper about chaining, but if it can iteratively allow the user to describe small changes, then one could build some tools that overcome many of the challenges of one-shot generation. https://t.co/eii2CqKq0z
— Mark Parody Riedl (@riedl@sigmoid.social) (@mark_riedl) October 18, 2022
