

















Abstract:Vector quantization is common in deep models, yet its hard assignments block gradients and hinder end-to-end training. We propose DiVeQ, which treats quantization as adding an error vector that mimics the quantization distortion, keeping the forward pass hard while letting gradients flow. We also present a space-filling variant (SF-DiVeQ) that assigns input to a curve constructed by the lines connecting codewords, resulting in less quantization error and full codebook usage. Both methods train end-to-end without requiring auxiliary losses or temperature schedules. In VQ-VAE image compression, VQGAN image generation, and DAC speech coding tasks across various data sets, our proposed methods improve reconstruction and sample quality over alternative quantization approaches.
| Subjects: | Machine Learning (cs.LG) |
| Cite as: | arXiv:2509.26469 [cs.LG] |
| (or arXiv:2509.26469v4 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2509.26469 arXiv-issued DOI via DataCite |
From: Mohammad Hassan Vali [view email]
[v1]
Tue, 30 Sep 2025 16:17:21 UTC (9,671 KB)
[v2]
Mon, 23 Mar 2026 10:28:35 UTC (9,520 KB)
[v3]
Tue, 12 May 2026 11:42:38 UTC (10,900 KB)
[v4]
Tue, 26 May 2026 12:24:30 UTC (9,520 KB)
此内容由惯性聚合(RSS阅读器)自动聚合整理,仅供阅读参考。 原文来自 — 版权归原作者所有。