Accepted at #NeurIPS2022! Thank you to all reviewers and ACs! I hope to see many of you in New Orleans!!

How does one remove or reduce bias in AI when even domain experts find it challenging to identify bias in an arbitrary dataset? Automatic bias detection has emerged as a promising approach to improve the performance of predictive algorithms... (1/3)

New paper! DiffDock: Diffusion Steps, Twists, and Turns for Molecular Docking arxiv.org/abs/2210.01776 1. Diffusion over molecule position, rotation, and torsion angles 2. From 23% accuracy to 38% on a time-split 🤗 3. Confidence estimates with high selective accuracy 1/3

In drug design, molecular docking is typically treated as a regression problem. While deep learning models are typically faster than traditional search-based methods, there have been no significant improvements in accuracy. Now, a diffusion generative model... (1/4)

2014: MIT professor Regina Barzilay is diagnosed w/#breastcancer. 2017: she develops a ML model that could have diagnosed her cancer years earlier. 2020: she wins AAAI's $1m award for her work. More: bit.ly/Barzilay1m #BreastCancerAwarenessMonth #BreastCancerSurvivor

Tal Schuster, Robin Jia, and I are giving a (fully virtual) tutorial on "Uncertainty Estimation for NLP" at COLING 2022 on Monday morning 8am Korea time (7pm EST Sunday). Join us! Website: sites.google.com/view/uncertain… Underline event: underline.io/events/360/ses… #uncertaintyNLP

Recently White House Office of Science & Technology Policy released a report outlining a bold vision for biotech + biomanufacturing, and R&D needs to achieve these goals. What do #JameelClinic researchers think about these goals? We asked 3rd year PhD student Rachel (Menghua) Wu to share her thoughts on the matter: 🧵 1/3

PSA from the doggos of #JameelClinic: Just 8 days left! Don't snooze on the chance to share your research at #AICures2023 & get FREE admission to the conference!

In our ContextViT paper with the wonderful YujiaBao we looked hard at the problem of learning vision transformers (SSL or SL) when you have data arriving in batches that can exhibit distribution shift/context shift (such as experimental batches) arxiv.org/abs/2305.19402 1/N

Another preprint that has been a lot of fun is our work on Channel Vision Transformers with YujiaBao and Srinivasan Sivanandan , a vision transformer variant which explicitly models different channels separately and creates one token per channel. Paper: arxiv.org/abs/2309.16108# 1/N

Our latest preprint on ML for influenza vaccine selection is out! We utilized historical antigenic analysis data from the World Health Organization (WHO) to demonstrate that our model selects vaccine strains that reliably improve over the annual recommendations. (1/4)

voyage-code-2 is the sota model for code retrieval

In RAG applications, LLMs often reprocess the same database chunks for different queries—leading to high latency and cost from handling massive input tokens. We introduce KVLink, an efficient approach to reuse pre-computed KV caches of retrieved documents, drastically cutting

Excited to share my #ICLR2025 paper, with JC Hütter and friends! Genetic perturbation screens allow biologists to manipulate and measure the genes in cells = discover causal relationships! BUT they are expensive to run, expensive to interpret. ... We use LLMs to help!