Sign-In to the Lottery: Reparameterizing Sparse Training From Scratch

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Gadhikar, Advait, Jacobs, Tom, Zhou, Chao, Burkholz, Rebekka
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866908606491787264
author Gadhikar, Advait
Jacobs, Tom
Zhou, Chao
Burkholz, Rebekka
author_facet Gadhikar, Advait
Jacobs, Tom
Zhou, Chao
Burkholz, Rebekka
contents The performance gap between training sparse neural networks from scratch (PaI) and dense-to-sparse training presents a major roadblock for efficient deep learning. According to the Lottery Ticket Hypothesis, PaI hinges on finding a problem specific parameter initialization. As we show, to this end, determining correct parameter signs is sufficient. Yet, they remain elusive to PaI. To address this issue, we propose Sign-In, which employs a dynamic reparameterization that provably induces sign flips. Such sign flips are complementary to the ones that dense-to-sparse training can accomplish, rendering Sign-In as an orthogonal method. While our experiments and theory suggest performance improvements of PaI, they also carve out the main open challenge to close the gap between PaI and dense-to-sparse training.
format Preprint
id arxiv_https___arxiv_org_abs_2504_12801
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Sign-In to the Lottery: Reparameterizing Sparse Training From Scratch
Gadhikar, Advait
Jacobs, Tom
Zhou, Chao
Burkholz, Rebekka
Machine Learning
Computer Vision and Pattern Recognition
The performance gap between training sparse neural networks from scratch (PaI) and dense-to-sparse training presents a major roadblock for efficient deep learning. According to the Lottery Ticket Hypothesis, PaI hinges on finding a problem specific parameter initialization. As we show, to this end, determining correct parameter signs is sufficient. Yet, they remain elusive to PaI. To address this issue, we propose Sign-In, which employs a dynamic reparameterization that provably induces sign flips. Such sign flips are complementary to the ones that dense-to-sparse training can accomplish, rendering Sign-In as an orthogonal method. While our experiments and theory suggest performance improvements of PaI, they also carve out the main open challenge to close the gap between PaI and dense-to-sparse training.
title Sign-In to the Lottery: Reparameterizing Sparse Training From Scratch
topic Machine Learning
Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2504.12801