Making Foundation Models Probabilistic via Singular Value Ensembles

TL;DR

This paper introduces Singular Value Ensemble (SVE), a parameter-efficient implicit ensemble method that enhances uncertainty quantification in foundation models by modulating singular values, achieving ensemble-like performance with minimal additional parameters.

Contribution

SVE leverages the singular vectors of pretrained models' weight matrices to create diverse ensembles by only training singular values, reducing computational costs and parameter overhead.

Findings

SVE achieves uncertainty calibration comparable to deep ensembles.

SVE improves model calibration while maintaining accuracy.

SVE requires less than 1% additional parameters over the base model.

Abstract

Foundation models have become a dominant paradigm in machine learning, achieving remarkable performance across diverse tasks through large-scale pretraining. However, these models often yield overconfident, uncalibrated predictions. The standard approach to quantifying epistemic uncertainty, training an ensemble of independent models, incurs prohibitive computational costs that scale linearly with ensemble size, making it impractical for large foundation models. We propose Singular Value Ensemble (SVE), a parameter-efficient implicit ensemble method that builds on a simple, but powerful core assumption: namely, that the singular vectors of the weight matrices constitute meaningful subspaces of the model's knowledge. Pretrained foundation models encode rich, transferable information in their weight matrices. If the singular vectors are indeed meaningful (orthogonal) "knowledge directions". To obtain a model ensemble, we modulate only how strongly each direction contributes to the output. Rather than learning entirely new parameters, we freeze the singular vectors and only train per-member singular values that rescale the contribution of each direction in that shared knowledge basis. Ensemble diversity emerges naturally as stochastic initialization and random sampling of mini-batches during joint training cause different members to converge to different combinations of the same underlying knowledge. SVE achieves uncertainty quantification comparable to explicit deep ensembles while increasing the parameter count of the base model by less than 1%, making principled uncertainty estimation accessible in resource-constrained settings. We validate SVE on NLP and vision tasks with various different backbones and show that it improves calibration while maintaining predictive accuracy.