Vedant Shah

N E R D

(1/n)🚨Train a model solving DFT for any geometry with almost no training data
Introducing Self-Refining Training for Amortized DFT: a variational method that predicts ground-state solutions across geometries and generates its own training data!
📜 arxiv.org/abs/2506.01225
💻 github.com/majhas/self-...

hank you to our funders for this project: CIFAR, NSERC, and Abundant Intelligences. Thank you also for meeting me with the rich discussions @tyrellturing.bsky.social, @veds12.bsky.social, @mnoukhov.bsky.social and @arnaghosh.bsky.social that gave clarity to the problem.

New preprint! 🧠🤖

How do we build neural decoders that are:
⚡️ fast enough for real-time use
🎯 accurate across diverse tasks
🌍 generalizable to new sessions, subjects, and even species?

We present POSSM, a hybrid SSM architecture that optimizes for all three of these axes!

🧵1/7

I will be presenting our work at the MATH-AI workshop at #NeurIPS2024 today.

Location: West Meeting Room 118-120
Time: 11:00 AM - 12:30 PM; 4:00 PM - 5:00 PM

Come by if you want to chat about designing difficult evaluation benchmarks, follow-up work, and mathematical reasoning in LLMs!

I will be at #NeurIPS2024 this week and will be presenting our work

"AI-Assisted Generation of Difficult Math Questions"

at the MATH-AI Workshop on Saturday 🚀!

Would love to chat if you are interested in topics related to LLM reasoning and systematic generalization!

arxiv.org/abs/2407.21009

Re: the scale is dead debate. Isn't it pretty obvious that just scaling is never going to work if your method breaks down on OOD inputs? The world is non-stationary, so it's constantly presenting new OOD inputs.