arXiv cs.AI· Quentin Fruytier, Akshay Malhotra, Shahab Hamidi-Rad, Aditya Sant, Aryan Mokhtari, Sujay Sanghavi·· 13 小时前AI 评分24
用 WAE-MMD 与可编程先验雕刻潜空间：解耦表示学习新方法
Sculpting the Latent Space: Disentanglement with WAE-MMD and Programmable Priors
AI 导读
研究提出用非参数 MMD 直接塑造聚合后验、实现可编程先验的 WAE-MMD 方法，相比主流基线将检测到的纠缠降低 70%，达到无监督统计独立性 SOTA。作者同时用无监督指标 Latent Predictability Score（LPS）量化了 VAE 与对抗训练在聚合后验上难以强制因子化表示的问题。
正文 · 原文
Abstract:Learning disentangled representations, where semantic features are independently captured in a latent space, is dominated by Variational Auto Encoders (VAE) and Adversarial Training. In this paper, we provide direct quantitative and visual evidence that these methods consistently fail to enforce the target factorized representation on the aggregate posterior, leading to entangled representations. We quantify this failure using the Latent Predictability Score (LPS), an unsupervised metric of mutual predictability that empirically lower bounds the level of entanglement in a latent space. To address these limitations, we leverage the non-parametric Maximum Mean Discrepancy (MMD) to directly sculpt the aggregate posterior enabling programmable priors. We demonstrate (1) state-of-the-art unsupervised statistical independence, with a 70% reduction in detected entanglement compared to leading baselines; (2) the flexibility to optionally inject semi-supervised information for explicit semantic alignment; and (3) precise shaping of the aggregate posterior distribution (validated through quantization-aware training) without sacrificing reconstruction quality.
| Comments: | 34 pages, 17 figures. Accepted at the SIAM International Conference on Data Mining (SDM 2026). v2: camera-ready revision; retitled; adds the LPS lower-bound lemma, NashAE baseline, CelebA, semi-supervised alignment and quantization-aware training experiments. Code: this https URL |
| Subjects: | Machine Learning (cs.LG); Artificial Intelligence (cs.AI) |
| Cite as: | arXiv:2510.11953 [cs.LG] |
| (or arXiv:2510.11953v2 [cs.LG] for this version) | |
| https://doi.org/10.48550/arXiv.2510.11953 arXiv-issued DOI via DataCite |
Submission history
From: Quentin Fruytier [view email]
[v1]
Mon, 13 Oct 2025 21:26:01 UTC (4,713 KB)
[v2]
Wed, 21 Oct 2026 17:17:00 UTC (5,565 KB)
来源：arXiv cs.AI · arxiv.org