Add ReversePredictivity (muzellec2026) benchmark

[S4b1n3]

Benchmark: Muzellec2026

Description

This benchmark evaluates model alignment with neural data using the Reverse Predictivity metric. We are using the initial benchmark MajajHong2015, but measuring how models can be predicted by neural data.

Components

• Metric: Added Reverse Predictivity to regression correlation metrics.
• Ceiling: In the regression Y is not noisy (model features) so we don't use internal consistency. But X is noisy so we introduced predictor_consistency metric, computing the internal consistency of the predictor, and use this as the ceiling.

Testing

• All unit tests pass locally
• Benchmark loads correctly
• Ceiling is valid (0.88 ± 0.005)
• Tested with alexnet: score = 0.10071875

Citation

@Article{muzellec2025reverse,
title={Reverse Predictivity: Going Beyond One-Way Mapping to Compare Artificial Neural Network Models and Brains},
author={Muzellec, Sabine and Kar, Kohitij},
journal={bioRxiv},
pages={2025--08},
year={2025},
publisher={Cold Spring Harbor Laboratory}

https://www.biorxiv.org/content/10.1101/2025.08.08.669382v1.full

[KartikP]

PR looks good to me! Made two commits to clean up unnecessary pkl file and change to pyproject.toml.

Upon scoring infrastructure change, this PR will be merged. This will occur next week.

[KartikP]

LGTM!