brain-encoding

Official repository for "A Dual-Stream Neural Network Explains the Functional Segregation of Dorsal and Ventral Visual Pathways in Human Brains", NeurIPS-23.

Load and model the brain data of the Algonauts Project 2023 Challenge.

Code for my Master's Thesis "Deep Neural Encoding Models of the Human Visual Cortex to Predict fMRI Responses to Natural Visual Scenes" and my submission for the "Algonauts Project 2023 Challenge".

MIND: Mixture-of-Experts Integrated Decoder for Multimodal Brain Encoding (ICASSP 2026)

A lightweight multimodal brain encoding model distilled from TRIBE v2. Tiny-TRIBE predicts fMRI cortical responses to naturalistic video stimuli using compact encoders (~14M parameters) and knowledge distillation from the 4.7B parameter teacher model.

Validates whether META's TRIBE v2 predicted brain activations help LLMs understand emotion. 3-condition A/B/C experiment, 100 samples, ElevenLabs audio + local LLM inference.

Neural attention analytics — predict brain responses to any video using TRIBE v2 (Meta FAIR)

Fork of Meta FAIR's TRIBE v2 multimodal brain-response model, carrying a fix for WhisperX on non-CUDA devices (int8 rather than float16) that was sent upstream as PR #20. Used in production by CineNeuro.

Neuroimaging preprocessing, brain decoding, and visual brain encoding using fMRI, EEG/MEG, CNNs, VLMs, and transformer representations

Exploring AI brain encoding models and neural activation mapping

Don't follow the herd. Predict it. Brain-encoded swarm social simulation engine. First project built on Meta TRIBE v2.

Subcortical response analysis for video stimuli using TRIBE v2.

Modular brain encoding pipeline aligning transformer-based NLP embeddings with fMRI data for brain-language analysis.

Drop a clip. See which cortical regions it would drive in an average brain. In-silico fMRI around Meta TRIBE v2 — encoding, not decoding.

Predicts a human audience's second-by-second neural engagement with a movie trailer before anyone watches it. Meta FAIR TRIBE v2 (V-JEPA2 + Wav2Vec-BERT + Llama 3.2) turns video into 20,484 fMRI vertices, mapped across 7 brain regions into 5 emotions, with scene peaks, personas and a PDF report.