Making Models with AI

The CMU study introduces BrainDiVE, a new diffusion model that combines brain imaging data with advanced image generation technology to understand how our visual brain works. Unlike traditional methods that rely on pre-selected images, BrainDiVE can create new images that are predicted to activate specific brain regions. The system successfully generated appropriate images for areas that process faces, places, bodies, words and food while revealing new patterns in how these regions function.

This CMU study introduces mDLAG, a tool to understand how different brain regions communicate. Using recordings from three visual areas in monkey brains (V1, V2, and V3d), they discovered that brain regions processing the same part of visual space showed stronger communication patterns. The method successfully identified which areas were talking to each other and the direction of information flow, improving our understanding of brain connectivity. The researchers hope to expand it to a larger section of the brain.

This Columbia study developed a new method for interpreting brain activity signals that bypasses traditional "spike sorting", where invasive neural recordings are broken down to assign activation patterns to individual neurons. The method directly analyzes key features of neural signals to understand behavior. Testing across multiple brain regions and animal species showed this approach outperformed conventional methods. The technique works in near real-time and is especially valuable in brain areas where traditional analysis is challenging.