Isotropy Matters: Soft-ZCA Whitening of Embeddings for Semantic Code Search

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Diera, Andor, Galke, Lukas, Scherp, Ansgar
Format: Preprint
Published: 2024
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913588222885888
author Diera, Andor
Galke, Lukas
Scherp, Ansgar
author_facet Diera, Andor
Galke, Lukas
Scherp, Ansgar
contents Low isotropy in an embedding space impairs performance on tasks involving semantic inference. Our study investigates the impact of isotropy on semantic code search performance and explores post-processing techniques to mitigate this issue. We analyze various code language models, examine isotropy in their embedding spaces, and its influence on search effectiveness. We propose a modified ZCA whitening technique to control isotropy levels in embeddings. Our results demonstrate that Soft-ZCA whitening improves the performance of pre-trained code language models and can complement contrastive fine-tuning.
format Preprint
id arxiv_https___arxiv_org_abs_2411_17538
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Isotropy Matters: Soft-ZCA Whitening of Embeddings for Semantic Code Search
Diera, Andor
Galke, Lukas
Scherp, Ansgar
Computation and Language
Low isotropy in an embedding space impairs performance on tasks involving semantic inference. Our study investigates the impact of isotropy on semantic code search performance and explores post-processing techniques to mitigate this issue. We analyze various code language models, examine isotropy in their embedding spaces, and its influence on search effectiveness. We propose a modified ZCA whitening technique to control isotropy levels in embeddings. Our results demonstrate that Soft-ZCA whitening improves the performance of pre-trained code language models and can complement contrastive fine-tuning.
title Isotropy Matters: Soft-ZCA Whitening of Embeddings for Semantic Code Search
topic Computation and Language
url https://arxiv.org/abs/2411.17538