Research map / Learning theory and algorithms / Kernels and regression theory
Autoencoders that don't overfit towards the Identity
Harald Steck
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| replace weight decay on U and V with L2 norm on the product UV^T (ablation) | linear autoencoder · MSD | nDCG@100: 0.249 -> 0.266 | Improved |
| replace weight decay with dropout-style L2 on UV^T (ablation) | linear autoencoder · ML-20M | nDCG@100: 0.376 -> 0.407 | Improved |
| use item-specific Lambda instead of constant lambda in L2 regularization (ablation) | linear autoencoder · MSD | nDCG@100: 0.266 -> 0.314 | Improved |
| use item-specific Lambda instead of constant lambda (ablation) | linear autoencoder · ML-20M | nDCG@100: 0.407 -> 0.417 | Improved |
| add full emphasis (emphasized denoising) to DLAE (ablation) | DLAE · ML-20M | nDCG@100: 0.413 -> 0.420 | Improved |
| add emphasized denoising to dropout training (ablation) | DLAE · Netflix | nDCG@100: 0.383 -> 0.394 | Improved |
| add emphasized denoising to dropout training (ablation) | DLAE · MSD | nDCG@100: 0.309 -> 0.320 | Improved |
| train with sampled dropout vs analytic dropout-induced L2 (ablation) | LAE · Netflix | nDCG@100: 0.384 vs 0.383 | No clear effect |
| enforce (near-)orthogonality constraint diag(UV^T)=0 instead of unconstrained diagonal (ablation) | DLAE low-rank · Netflix | nDCG@100 | Improved |
Closest papers by meaning
- Denoising Autoencoders for Overgeneralization in Neural Networks (TPAMI 2019 · similarity 0.87)
- High-dimensional Asymptotics of Denoising Autoencoders (NeurIPS 2023 · similarity 0.85)
- Denoising Criterion for Variational Auto-Encoding Framework (AAAI 2017 · similarity 0.85)
- Learning Dynamics of Linear Denoising Autoencoders (ICML 2018 · similarity 0.85)
- Supervised autoencoders: Improving generalization performance with unsupervised regularizers (NeurIPS 2018 · similarity 0.84)
- Deconstructing Denoising Diffusion Models for Self-Supervised Learning (ICLR 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
