Justin Deschenaux

Excited to share our latest work on EvoTune, a novel method integrating LLM-guided evolutionary search and reinforcement learning to accelerate the discovery of algorithms! 1/12🧵

A dream come true! I presented "No Representation, No Trust" on my favorite RL podcast, TalkRL!
Make sure to check it out to learn why training with PPO for too long makes your agent collapse!

I am in Vancouver for NeurIPS 2024 until December 16th if you want to meet, DM or email me.
We have two accepted papers from my lab:
1. Building on Efficient Foundations: Effective Training of LLMs with Structured Feedforward Layers, on Wednesday, East Exhibit Hall A-C #2010 (1/3)

Ah, I see, you got me here 😆

Where is this paper? I cannot find it

Is the paper on ArXiv?

@haeggee.bsky.social Is missing from the list 😢

Maybe we can set our bluesky handle, in addition to twitter, in our huggingface profile? 👀

They have released many of their models, maybe that’s why you thought that? I also thought that the weights were released

Amazing repo 🤩 Thank you for including our work ☺️

For people coming from an engineering background (ie without a background in SDEs, I feel like the variational diffusion model perspective (arxiv.org/pdf/2107.00630) is relatively intuitive, as then diffusion is kind of a generalized VAE

A paper a day keeps the FOMO away, episode 7.

Among "oldies but goldies", this tutorial by Rabiner on Hidden Markov Models (HMMs) is dear to my heart. HMMs are one of the simplest statistical models where some variables are not observed, and we love them for it. 🧵

www.cs.ubc.ca/~murphyk/Bay...

Thanks for sharing those high quality resources! 🙌

In a gratuitous attempt to acquire more followers myself 😁, I've made a start on a "starter pack". Hopefully as more people from 🐦 make it over to 🦋, we can extend this a bit. Suggestions welcome!

I've noticed not all accounts seem to be eligible to be added, anyone know what's up with that? 🤔