Ensembling Sparse Autoencoders

TL;DR

This paper explores ensembling multiple sparse autoencoders to enhance feature diversity, stability, and downstream task performance in language models, demonstrating significant empirical improvements over single autoencoder approaches.

Contribution

It introduces ensemble methods for sparse autoencoders, including bagging and boosting, to improve feature extraction and downstream task effectiveness.

Findings

Ensembling SAEs improves activation reconstruction.

Ensembling increases feature diversity and stability.

Ensembling outperforms single SAEs in downstream tasks.

Abstract

Sparse autoencoders (SAEs) are used to decompose neural network activations into human-interpretable features. Typically, features learned by a single SAE are used for downstream applications. However, it has recently been shown that SAEs trained with different initial weights can learn different features, demonstrating that a single SAE captures only a limited subset of features that can be extracted from the activation space. Motivated by this limitation, we propose to ensemble multiple SAEs through naive bagging and boosting. Specifically, SAEs trained with different weight initializations are ensembled in naive bagging, whereas SAEs sequentially trained to minimize the residual error are ensembled in boosting. We evaluate our ensemble approaches with three settings of language models and SAE architectures. Our empirical results demonstrate that ensembling SAEs can improve the reconstruction of language model activations, diversity of features, and SAE stability. Furthermore, ensembling SAEs performs better than applying a single SAE on downstream tasks such as concept detection and spurious correlation removal, showing improved practical utility.