Saved in:
Bibliographic Details
Main Authors: Salishev, Sergey, Akhremchik, Ian
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.14004
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909897372729344
author Salishev, Sergey
Akhremchik, Ian
author_facet Salishev, Sergey
Akhremchik, Ian
contents Quantized neural networks can be viewed as a chain of noisy channels, where rounding in each layer reduces capacity as bit-width shrinks; the floating-point (FP) checkpoint sets the maximum input rate. We track capacity dynamics as the average bit-width decreases and identify resulting quantization bottlenecks by casting fine-tuning as a smooth, constrained optimization problem. Our approach employs a fully differentiable Straight-Through Estimator (STE) with learnable bit-width, noise scale and clamp bounds, and enforces a target bit-width via an exterior-point penalty; mild metric smoothing (via distillation) stabilizes training. Despite its simplicity, the method attains competitive accuracy down to the extreme W1A1 setting while retaining the efficiency of STE.
format Preprint
id arxiv_https___arxiv_org_abs_2508_14004
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle GDNSQ: Gradual Differentiable Noise Scale Quantization for Low-bit Neural Networks
Salishev, Sergey
Akhremchik, Ian
Machine Learning
Information Theory
Numerical Analysis
68T07, 90C26
I.2.6; E.4; G.1.6
Quantized neural networks can be viewed as a chain of noisy channels, where rounding in each layer reduces capacity as bit-width shrinks; the floating-point (FP) checkpoint sets the maximum input rate. We track capacity dynamics as the average bit-width decreases and identify resulting quantization bottlenecks by casting fine-tuning as a smooth, constrained optimization problem. Our approach employs a fully differentiable Straight-Through Estimator (STE) with learnable bit-width, noise scale and clamp bounds, and enforces a target bit-width via an exterior-point penalty; mild metric smoothing (via distillation) stabilizes training. Despite its simplicity, the method attains competitive accuracy down to the extreme W1A1 setting while retaining the efficiency of STE.
title GDNSQ: Gradual Differentiable Noise Scale Quantization for Low-bit Neural Networks
topic Machine Learning
Information Theory
Numerical Analysis
68T07, 90C26
I.2.6; E.4; G.1.6
url https://arxiv.org/abs/2508.14004