Beyond Discreteness: Sample Complexity Analysis of Straight-Through Estimator for 1-bit Quantization

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Jeong, Halyun, Xin, Jack, Yin, Penghang
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866913178361790464
author Jeong, Halyun
Xin, Jack
Yin, Penghang
author_facet Jeong, Halyun
Xin, Jack
Yin, Penghang
contents Training quantized neural networks requires addressing the non-differentiable and discrete nature of the underlying optimization problem. To tackle this challenge, the straight-through estimator (STE) has become the most widely adopted heuristic, allowing backpropagation through discrete operations by introducing biased yet valid surrogate gradients. However, its theoretical properties remain largely unexplored, with few existing analyses focus on the generalization error by assuming an infinite amount of training data. In contrast, this work presents the first sample complexity analysis of STE in the context of neural network quantization. Our theoretical results highlight the critical role of sample size in the success of STE, a key insight absent from existing studies. Specifically, by analyzing the quantization-aware training of a two-layer neural network with binary weights and activations, we derive the sample complexity bounds in terms of the data dimensionality that guarantee the convergence of STE-based optimization to the global minimum for both ergodic and non-ergodic analyses. Moreover, in the presence of label noises, we prove an intriguing recurrence property of STE-gradient method, where the iterate repeatedly escape from and return to the optimal binary weights. Finally, we empirically demonstrate that STE fails for general non-Gaussian data but its effectiveness can be restored through normalization, underscoring its practical importance in effective quantization.
format Preprint
id arxiv_https___arxiv_org_abs_2505_18113
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond Discreteness: Sample Complexity Analysis of Straight-Through Estimator for 1-bit Quantization
Jeong, Halyun
Xin, Jack
Yin, Penghang
Machine Learning
Optimization and Control
68T07, 68Q32, 90C26, 94A12
Training quantized neural networks requires addressing the non-differentiable and discrete nature of the underlying optimization problem. To tackle this challenge, the straight-through estimator (STE) has become the most widely adopted heuristic, allowing backpropagation through discrete operations by introducing biased yet valid surrogate gradients. However, its theoretical properties remain largely unexplored, with few existing analyses focus on the generalization error by assuming an infinite amount of training data. In contrast, this work presents the first sample complexity analysis of STE in the context of neural network quantization. Our theoretical results highlight the critical role of sample size in the success of STE, a key insight absent from existing studies. Specifically, by analyzing the quantization-aware training of a two-layer neural network with binary weights and activations, we derive the sample complexity bounds in terms of the data dimensionality that guarantee the convergence of STE-based optimization to the global minimum for both ergodic and non-ergodic analyses. Moreover, in the presence of label noises, we prove an intriguing recurrence property of STE-gradient method, where the iterate repeatedly escape from and return to the optimal binary weights. Finally, we empirically demonstrate that STE fails for general non-Gaussian data but its effectiveness can be restored through normalization, underscoring its practical importance in effective quantization.
title Beyond Discreteness: Sample Complexity Analysis of Straight-Through Estimator for 1-bit Quantization
topic Machine Learning
Optimization and Control
68T07, 68Q32, 90C26, 94A12
url https://arxiv.org/abs/2505.18113