Congratulations, Ko! So well deserved

In closing, I want to thank my amazing advisors for supervising this work. I could not have imagined a better place to earn my PhD than the Department of Cognitive Science at @jhuartssciences.bsky.social

Taken together, our results do support a spatiotemporal hierarchy for computing social interactions in the lateral visual stream.

In our joint regression procedure we find that communicative interactions predict STS activity within 200 ms of stimulus onset. This reveals a rapid representation of high-level social interaction features in the brain.

In the joint regression, we find that social primitives are broadly represented in mid- and high-level regions of the STS with a similar latency. Combined with the finding that LOC and aSTS are predicted with similar latency, this suggests there may be skip connections between EVC and the STS.

In our final analysis, we use our novel regression procedure to jointly predict fMRI responses using EEG and stimulus features. We find that low-level visual features predict EVC activity with a short temporal latency and that communication is represented in the STS with the longest latency.

The EEG signal predicts early visual cortex (EVC) with a very short latency. However, mid-level and high-level regions in the lateral visual pathway are represented with a similar latency, suggesting that the lateral visual stream may not be organized in a strict feedforward hierarchy.

We find that visual social primitives, such as how far apart two people are in a video, are decodable from EEG earlier than social interactions features like whether two people are communicating.

Monkey See, Model Knew: LLMs accurately Predict Human AND Macaque Visual Brain Activity
Colin Conwell, @emaliemcmahon.bsky.social, Akshay Vivek Jagadeesh, Kasper Vinken, @amrahs-inolas.bsky.social, @jacob-prince.bsky.social, George Alvarez, @taliakonkle.bsky.social and Marge Livingstone

Woo! So excited to read it!