How can we distribute data points more uniformly for multi-dimensional problems in robotics, finance, & computational science? 🧵

MIT CSAIL’s AI-powered method uses graph neural networks to do this, self-optimizing to avoid clumps & voids. This "Message-Passing Monte Carlo (MPMC)" approach significantly improves accuracy in simulations & computations:

PNAS Paper: bit.ly/3ZLvGi0
bit.ly/4ehj1If MPMC uses graph neural networks to allow points to "communicate" for better uniformity.

Often, the more evenly you can spread out those data points, the more accurately you can simulate complex systems.

Training LLMs can be much cheaper than previously thought.

While companies like @OpenAI and @Meta use billions of dollars to train theirs, CSAIL & @myshell_ai research shows that just 0.1 million USD is sufficient for training LLaMA2-level LLMs.

Introducing the open-source project JetMoE:

A thread 📊research.myshell.ai/jetmoe JetMoE democratizes the training of high-performance LLMs, and makes it achievable by a wide range of research institutes and companies.

We at @MIT_CSAIL are so proud of the role our alum Dr. Katie Bouman played in the development of the first-ever picture of a black hole. She's been psyched about all the #blackhole interest & just wanted to clarify a few things. (1/7) In our first tweet about this, we linked to a 2016 story about an algorithm she led the development of while at CSAIL. That algorithm was intended to take a picture of a black hole, but didn’t create the final image. (cont.)(2/7)