Exploring Expert Specialization through Unsupervised Training in Sparse Mixture of Experts

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Nikolic, Strahinja, Oguz, Ilker, Psaltis, Demetri
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866916947374899200
author Nikolic, Strahinja
Oguz, Ilker
Psaltis, Demetri
author_facet Nikolic, Strahinja
Oguz, Ilker
Psaltis, Demetri
contents Understanding the internal organization of neural networks remains a fundamental challenge in deep learning interpretability. We address this challenge by exploring a novel Sparse Mixture of Experts Variational Autoencoder (SMoE-VAE) architecture. We test our model on the QuickDraw dataset, comparing unsupervised expert routing against a supervised baseline guided by ground-truth labels. Surprisingly, we find that unsupervised routing consistently achieves superior reconstruction performance. The experts learn to identify meaningful sub-categorical structures that often transcend human-defined class boundaries. Through t-SNE visualizations and reconstruction analysis, we investigate how MoE models uncover fundamental data structures that are more aligned with the model's objective than predefined labels. Furthermore, our study on the impact of dataset size provides insights into the trade-offs between data quantity and expert specialization, offering guidance for designing efficient MoE architectures.
format Preprint
id arxiv_https___arxiv_org_abs_2509_10025
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Exploring Expert Specialization through Unsupervised Training in Sparse Mixture of Experts
Nikolic, Strahinja
Oguz, Ilker
Psaltis, Demetri
Machine Learning
Artificial Intelligence
Understanding the internal organization of neural networks remains a fundamental challenge in deep learning interpretability. We address this challenge by exploring a novel Sparse Mixture of Experts Variational Autoencoder (SMoE-VAE) architecture. We test our model on the QuickDraw dataset, comparing unsupervised expert routing against a supervised baseline guided by ground-truth labels. Surprisingly, we find that unsupervised routing consistently achieves superior reconstruction performance. The experts learn to identify meaningful sub-categorical structures that often transcend human-defined class boundaries. Through t-SNE visualizations and reconstruction analysis, we investigate how MoE models uncover fundamental data structures that are more aligned with the model's objective than predefined labels. Furthermore, our study on the impact of dataset size provides insights into the trade-offs between data quantity and expert specialization, offering guidance for designing efficient MoE architectures.
title Exploring Expert Specialization through Unsupervised Training in Sparse Mixture of Experts
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2509.10025