Compression-enabled interpretability of voxelwise encoding models.

Voxelwise encoding models based on convolutional neural networks (CNNs) are widely used as predictive models of brain activity evoked by natural movies. Despite their superior predictive performance, the huge number of parameters in CNN-based models have made them difficult to interpret. Here, we in...

Full description

Saved in:
Bibliographic Details
Main Authors: Fatemeh Kamali, Amir Abolfazl Suratgar, Mohammadbagher Menhaj, Reza Abbasi-Asl
Format: Article
Language:English
Published: Public Library of Science (PLoS) 2025-02-01
Series:PLoS Computational Biology
Online Access:https://doi.org/10.1371/journal.pcbi.1012822
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850039479935434752
author Fatemeh Kamali
Amir Abolfazl Suratgar
Mohammadbagher Menhaj
Reza Abbasi-Asl
author_facet Fatemeh Kamali
Amir Abolfazl Suratgar
Mohammadbagher Menhaj
Reza Abbasi-Asl
author_sort Fatemeh Kamali
collection DOAJ
description Voxelwise encoding models based on convolutional neural networks (CNNs) are widely used as predictive models of brain activity evoked by natural movies. Despite their superior predictive performance, the huge number of parameters in CNN-based models have made them difficult to interpret. Here, we investigate whether model compression can build more interpretable and more stable CNN-based voxelwise models while maintaining accuracy. We used multiple compression techniques to prune less important CNN filters and connections, a receptive field compression method to select receptive fields with optimal center and size, and principal component analysis to reduce dimensionality. We demonstrate that the model compression improves the accuracy of identifying visual stimuli in a hold-out test set. Additionally, compressed models offer a more stable interpretation of voxelwise pattern selectivity than uncompressed models. Finally, the receptive field-compressed models reveal that the optimal model-based population receptive fields become larger and more centralized along the ventral visual pathway. Overall, our findings support using model compression to build more interpretable voxelwise models.
format Article
id doaj-art-41c1d635d8f34dc8a6ad2c0609ab782a
institution DOAJ
issn 1553-734X
1553-7358
language English
publishDate 2025-02-01
publisher Public Library of Science (PLoS)
record_format Article
series PLoS Computational Biology
spelling doaj-art-41c1d635d8f34dc8a6ad2c0609ab782a2025-08-20T02:56:20ZengPublic Library of Science (PLoS)PLoS Computational Biology1553-734X1553-73582025-02-01212e101282210.1371/journal.pcbi.1012822Compression-enabled interpretability of voxelwise encoding models.Fatemeh KamaliAmir Abolfazl SuratgarMohammadbagher MenhajReza Abbasi-AslVoxelwise encoding models based on convolutional neural networks (CNNs) are widely used as predictive models of brain activity evoked by natural movies. Despite their superior predictive performance, the huge number of parameters in CNN-based models have made them difficult to interpret. Here, we investigate whether model compression can build more interpretable and more stable CNN-based voxelwise models while maintaining accuracy. We used multiple compression techniques to prune less important CNN filters and connections, a receptive field compression method to select receptive fields with optimal center and size, and principal component analysis to reduce dimensionality. We demonstrate that the model compression improves the accuracy of identifying visual stimuli in a hold-out test set. Additionally, compressed models offer a more stable interpretation of voxelwise pattern selectivity than uncompressed models. Finally, the receptive field-compressed models reveal that the optimal model-based population receptive fields become larger and more centralized along the ventral visual pathway. Overall, our findings support using model compression to build more interpretable voxelwise models.https://doi.org/10.1371/journal.pcbi.1012822
spellingShingle Fatemeh Kamali
Amir Abolfazl Suratgar
Mohammadbagher Menhaj
Reza Abbasi-Asl
Compression-enabled interpretability of voxelwise encoding models.
PLoS Computational Biology
title Compression-enabled interpretability of voxelwise encoding models.
title_full Compression-enabled interpretability of voxelwise encoding models.
title_fullStr Compression-enabled interpretability of voxelwise encoding models.
title_full_unstemmed Compression-enabled interpretability of voxelwise encoding models.
title_short Compression-enabled interpretability of voxelwise encoding models.
title_sort compression enabled interpretability of voxelwise encoding models
url https://doi.org/10.1371/journal.pcbi.1012822
work_keys_str_mv AT fatemehkamali compressionenabledinterpretabilityofvoxelwiseencodingmodels
AT amirabolfazlsuratgar compressionenabledinterpretabilityofvoxelwiseencodingmodels
AT mohammadbaghermenhaj compressionenabledinterpretabilityofvoxelwiseencodingmodels
AT rezaabbasiasl compressionenabledinterpretabilityofvoxelwiseencodingmodels