Saved in:
Bibliographic Details
Main Authors: Dooms, Thomas, Gauderis, Ward, Wiggins, Geraint A., Oramas, Jose
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2504.02667
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909563467333632
author Dooms, Thomas
Gauderis, Ward
Wiggins, Geraint A.
Oramas, Jose
author_facet Dooms, Thomas
Gauderis, Ward
Wiggins, Geraint A.
Oramas, Jose
contents We propose $χ$-net, an intrinsically interpretable architecture combining the compositional multilinear structure of tensor networks with the expressivity and efficiency of deep neural networks. $χ$-nets retain equal accuracy compared to their baseline counterparts. Our novel, efficient diagonalisation algorithm, ODT, reveals linear low-rank structure in a multilayer SVHN model. We leverage this toward formal weight-based interpretability and model compression.
format Preprint
id arxiv_https___arxiv_org_abs_2504_02667
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Compositionality Unlocks Deep Interpretable Models
Dooms, Thomas
Gauderis, Ward
Wiggins, Geraint A.
Oramas, Jose
Machine Learning
We propose $χ$-net, an intrinsically interpretable architecture combining the compositional multilinear structure of tensor networks with the expressivity and efficiency of deep neural networks. $χ$-nets retain equal accuracy compared to their baseline counterparts. Our novel, efficient diagonalisation algorithm, ODT, reveals linear low-rank structure in a multilayer SVHN model. We leverage this toward formal weight-based interpretability and model compression.
title Compositionality Unlocks Deep Interpretable Models
topic Machine Learning
url https://arxiv.org/abs/2504.02667