encoding-models

Voxelwise Encoding Model tutorials from the Gallant lab.

Data and code for the Algonauts Project 2025 challenge.

Trained encoding models to generate in silico neural responses for arbitrary stimuli.

Interpretable text embeddings by asking LLMs yes/no questions (NeurIPS 2024)

Code to reproduce the results from the NSD-synthetic data release paper.

Scripts and QA for the hyperface fMRI dataset

Algorithms to investigate (through neural control) the representational relationships between univariate or multivariate in silico fMRI responses of multiple visual areas of the brain.

Curated papers, datasets, and tools for fMRI-based encoding models, decoding, and representational analyses (language, vision, video, audio, and mental state).

Code and analysis for Diaz & Henderson (2026), including VEDB visual-field preprocessing, SimCLR ResNet-18 pretraining, downstream evaluation, reference models, NSD voxelwise encoding models, and variance-partitioning analyses.

Brain alignment differences between LLMs are not measurable at any scale public fMRI provides. Each model's map is reliable (0.91-0.95); their difference is not (0.005-0.031). Includes a one-correlation diagnostic.

Layer-wise CNN encoding of mouse V1 responses (Allen Institute Neuropixels), with an adversarial audit that downgraded the headline claim.

Extract time-varying visual, audio, and semantic features from video/audio clips and relate them to human ratings or a recorded signal (e.g. a brain channel) with a cross-validated, lag-aware encoding model.