It is so sad to see something like this happening to the best research system in the world, and probably the best ever in history. In China, for years scientists had wished to have something nearly as good -- in fact, Chinese NSF has been emulating NSF in recent years, which, in my personal view, has contributed to much of the recent scientific and technological advances in China.
We release Diamond Maps💎 unlocking accurate and efficient guidance for diffusion models. Our experiments show that our methods scale incredibly well. Excited to see what people will build with this!
Accurate guidance has been a notoriously hard problem, but in this work, we’re bringing TWO (!) solutions to the table. The recipe for success:
1️⃣ Speed: Use distilled models (flow maps, mean flows, consistency models).
2️⃣ Exploration: Inject stochasticity to properly explore your search space.
Because this fundamentally improves anything using flow matching and diffusion, we see a lot of potential for applications across audio, robotics, molecules, and beyond.
Paper: https://t.co/wxtWWRrnw7
Code: https://t.co/WocPtT6orn
Huge thanks to an amazing team: Douglas Chen, @LucaEyring, @ishin_shah, Giri Anantharaman, @electronickale, @zeynepakata, Tommi Jaakkola, @nmboffi, and @max_simchowitz. It was awesome bringing this to life together!
the crucial thing to ANORA is that to Ani believing in the relationship doesn't just nail what it's like to be naive about the future in your 20s, it's that being with Vanya will so radically change her life and social circumstances she has no choice but to believe it's real
New paper out!
We introduce “LEAPS”, a neural sampling algorithm for discrete distributions via continuous-time Markov chains (“discrete diffusion”). We introduce a novel importance sampling scheme and novel symmetries built into neural networks.
https://t.co/jpLZI85QLK
(1/4)
Excited to share that 6 papers were accepted at ICLR 2025! ✨ #ICLR2025
We proposed long-context perplexity, invariant in-context learning, and constrained tool decoding for better training and usage of LLMs. We also looked into some fundamental questions, such as OOD generalization of in-context learning, the interplay between monosemanticity and robustness, and the nature of projection heads. Check the pic for a brief intro (and save time scrolling over the thread).
I'm on the market and would love to discuss potential opportunities!
Excited to share our new 7B LLM @liquidai .
Strong evals on diverse tasks (including several evals from the synthetic arena that I lead), long context strength at low memory cost, and edge-device / on-prem deployment options for customers. Great work from the team :).
We've long suspected that Spotify was recruiting fake artists and juicing them in its algorithm and placing them on the most popular playlists. Liz Pelly collected the internal docs + other evidence to finally prove it.
I have exciting news!🌟 In Jan 2025, I will begin my position as an Assistant Professor at @UTHealthSPH in the Center for Human Genetics & Department of Epidemiology :)
I'm hiring at all levels, so please send your trainees my way!! More info at https://t.co/GlQdIHNRDa 🧬
New version + code for our NeurIPS paper is now out:
“The Empirical Impact of Neural Parameter Symmetries, or Lack Thereof.”
We study how symmetries in weight-space impact optimization and loss landscape geometry of neural nets, via "counterfactual" NNs w/o symmetries.
1/n
Excited to share some of the models we have been working on at Liquid! It's been quite fun to work on alternative architectures at small parameter scales, lots more to explore.
🧵 1/ What’s at stake with race-based equations for lung function? Clinical, occupational, and financial reclassifications for millions of patients.
I’m delighted to share this labor of love, published today in @NEJM and unveiled at #ATS2024.
https://t.co/9V0Bck071g
New #NVIDIA ICLR spotlight paper:
What if neural nets could process, analyze, interpret, and edit other neural nets weights? These models, termed metanetworks, are useful for various tasks.
We develop new metanetworks for processing diverse input neural net architectures.
1/n
I'm so excited to share what @AlanLiu96, @AminNathanamin and I have been working on!
Please reach out if you or someone you know works at a private practice, so we can save them from the agony of talking to insurance!
Excited to introduce CARE: Contrastive Augmentation-induced Rotational Equivariance, an equivariant contrastive learning approach that trains augmentations of input data to correspond to orthogonal transformations of the embedding space.
Paper: https://t.co/zA4CHkrp6F
excited to be presenting spotlight work at ICLR soon: “Relational Attention: Generalizing Transformers for Graph-Structured Tasks” 😇
the name says it all: we generalized the standard transformer to learn node AND EDGE representations, resulting in a powerful graph processor!
1/ New preprint!🔔
Excited to share my internship work at @CuraiHQ.
Using model distillation techniques, we’ve developed an approach to create light-weight guardrail models that monitor the output of generative language models like GPT-4!
w/@nairvarun18@elliotschu@anithakan
My brother is coming to @MIT_CSAIL for his PhD!!
Could not be prouder of @CD1AO and all the hard work that brought him here.
Excited to finally live in the same city again after 9 years!