Unifying Diffusion Models' Latent Space, with Applications to CycleDiffusion and Guidance
abs: https://t.co/d3NRBZTL4a pic.twitter.com/sP0PLwfHHk

— AK (@_akhaliq) October 12, 2022

This looks like the Vision Transformers architecture we have been waiting for: MaxViT https://t.co/WbzgJ50PjB
1/ State of the Art accuracy on ImageNet (no pre-training on huge datasets)
2/ Linear complexity wrt. image size (thanks to a clever attention design) pic.twitter.com/5bW0N7n3s5

— Martin Görner (@martin_gorner) October 11, 2022

All the related work section should be like the one of the recent "Attention Beats Concatenation for Conditioning Neural Fields" of Rebain et al. https://t.co/MBQfH7sE6S

Look at this cool git-like branching graph putting all the cited work in perspective! pic.twitter.com/HXLSyXe4xE

— Thomas Wolf (@Thom_Wolf) October 11, 2022

Understanding HTML with Large Language Models
abs: https://t.co/nKCCunfLxr
project page: https://t.co/cktfbuul4R pic.twitter.com/RzOaKOvGqP

— AK (@_akhaliq) October 11, 2022

MaxViT : combines ConvNet modules and 2 types of self attention (local n'y block, and on a subsampled grid).
Since DETR (hi @alcinos26 !), I've become convinced that combining Conv and attention/dynamic routing was the Right Thing. https://t.co/DNOBsqL54Z

— Yann LeCun (@ylecun) October 10, 2022

GLM-130B reaches INT4 quantization w/ no perf degradation, allowing effective inference on 4*3090 or 8*2080 Ti GPUs, the most ever affordable GPUs required for using 100B-scale models?

Paper: https://t.co/f2bj1N8JTN
Model weights & code & demo & lessons: https://t.co/aKZNGEDmks pic.twitter.com/kVRV0b8Y56

— Tsinghua KEG (@thukeg) October 10, 2022

Content-Based Search for Deep Generative Models
abs: https://t.co/6yAYV5XNqO
project page: https://t.co/fTF1qDsYyh pic.twitter.com/jxEDXIagrJ

— AK (@_akhaliq) October 7, 2022

XDoc: Unified Pre-training for Cross-Format Document Understanding
abs: https://t.co/bHZuRhbzDP pic.twitter.com/CQudsSPM4e

— AK (@_akhaliq) October 7, 2022

Decomposed Prompting: A Modular Approach for Solving Complex Tasks
abs: https://t.co/yhdw7lTUmr pic.twitter.com/icyxZt9k21

— AK (@_akhaliq) October 6, 2022

Evaluate & Evaluation on the Hub: Better Best Practices for Data and Model Measurement
abs: https://t.co/zPhPvQm045 pic.twitter.com/MJRKsXSFzV

— AK (@_akhaliq) October 6, 2022

Ask Me Anything: A simple strategy for prompting language models
abs: https://t.co/FxJBE7cLG1
github: https://t.co/GH31qYBcY9 pic.twitter.com/unUnTITbDE

— AK (@_akhaliq) October 6, 2022

SHAP, LIME, PFI, ... you can interpret ML models with many different methods.

It's all fun and games until two methods disagree.

What if LIME says X1 has a positive contribution, SHAP says negative?

A thread about the disagreement problem, and how to approach it:

— Christoph Molnar (@ChristophMolnar) October 4, 2022