Progress on Our BRAINIAC Initiative 🧠
Just submitted our NSF ACCESS allocation progress report, and I wanted to share some highlights! The BRAINIAC project (Broad‐scope Reasoning Artificial Intelligence for Nano-micro Identification, Assessment, and Categorization) is really taking off! 🌟
What we’ve accomplished 📊
- Our proof-of-concept is now published as a cover article in Molecular Systems Design & Engineering! 🎉
- Our SGNN model achieved 89% classification accuracy on sensor data - that’s 4x better than random baselines!
- We’ve successfully extracted and structured data from over 1,400 research papers using LLM-assisted text mining
BRAINIAC combines three powerful approaches:
- LLM-Assisted Text Mining to convert research papers into structured datasets 📚
- Spiking Graph Neural Networks that fuse global properties with molecular details 🔄
- High-Fidelity Simulations (DFT/AIMD) for mechanistic validation 🔬
The coolest part? We’re bridging the gap between isolated ML approaches and how human experts actually think and reason about materials!
Challenges & next steps 🔮
Scaling up isn’t easy! We’re working on:
- Processing hundreds of thousands more publications, More! and More! We want to scale it up to 1 Million+ Publications!
- Including time-series sensor response data
- Running more simulations across diverse material families
- Building an API so others can use BRAINIAC for their research
our MSDE paper!