Saved in:
Bibliographic Details
Main Authors: Peng, Shuman, Khoeini, Arash, Vaswani, Sharan, Ester, Martin
Format: Preprint
Published: 2024
Subjects:
Online Access:https://arxiv.org/abs/2411.13073
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917842140528640
author Peng, Shuman
Khoeini, Arash
Vaswani, Sharan
Ester, Martin
author_facet Peng, Shuman
Khoeini, Arash
Vaswani, Sharan
Ester, Martin
contents The quality of self-supervised pre-trained embeddings on out-of-distribution (OOD) data is poor without fine-tuning. A straightforward and simple approach to improving the generalization of pre-trained representation to OOD data is the use of deep ensembles. However, obtaining an effective ensemble in the embedding space with only unlabeled data remains an unsolved problem. We first perform a theoretical analysis that reveals the relationship between individual hyperspherical embedding spaces in an ensemble. We then design a principled method to align these embedding spaces in an unsupervised manner. Experimental results on the MNIST dataset show that our embedding-space ensemble method improves pre-trained embedding quality on in-distribution and OOD data compared to single encoders.
format Preprint
id arxiv_https___arxiv_org_abs_2411_13073
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Improving OOD Generalization of Pre-trained Encoders via Aligned Embedding-Space Ensembles
Peng, Shuman
Khoeini, Arash
Vaswani, Sharan
Ester, Martin
Machine Learning
Computer Vision and Pattern Recognition
The quality of self-supervised pre-trained embeddings on out-of-distribution (OOD) data is poor without fine-tuning. A straightforward and simple approach to improving the generalization of pre-trained representation to OOD data is the use of deep ensembles. However, obtaining an effective ensemble in the embedding space with only unlabeled data remains an unsolved problem. We first perform a theoretical analysis that reveals the relationship between individual hyperspherical embedding spaces in an ensemble. We then design a principled method to align these embedding spaces in an unsupervised manner. Experimental results on the MNIST dataset show that our embedding-space ensemble method improves pre-trained embedding quality on in-distribution and OOD data compared to single encoders.
title Improving OOD Generalization of Pre-trained Encoders via Aligned Embedding-Space Ensembles
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2411.13073