Research map / Label-efficient and robust learning / Recognition and long-tail
Denoising Autoencoders for Overgeneralization in Neural Networks
Giacomo Spigler
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 |
|---|---|---|---|
| add dAE confidence score | CNN · MNIST fooling 99% | fooling rate: 99% -> 0% | Improved |
| add COOL output layer | CNN · MNIST fooling 99% | fooling rate: 99% -> 4.5% | Improved |
| add dAE confidence score | CNN · Fashion-MNIST fooling 99% | fooling rate: 30.5% -> 0% | Improved |
| add dAE confidence score | CNN · Fashion-MNIST fooling 90% | thresholded accuracy: 90.91% -> 77.8% | Hurt |
| use symmetric instead of asymmetric decoder (ablation) | dAE · MNIST fooling | accuracy: 98.98% -> 99.14% | No clear effect |
| use sigmoid instead of ReLU activations (ablation) | COOL · MNIST fooling 99% | fooling rate: 4.5% -> 95.5% | Hurt |
| increase fooling learning rate from 0.00001 to 0.001 (ablation) | COOL · MNIST fooling 90% | fooling rate: 34.5% -> 56.5% | Improved |
| increase fooling learning rate from 0.00001 to 0.001 (ablation) | dAE · MNIST fooling 90% | fooling rate: 0% -> 0% | No clear effect |
| attack confidence score directly instead of classifier output (ablation) | dAE · MNIST fooling 99% | fooling rate: 0% -> 1% | Hurt |
| add L2 regularization lambda=10 to dAE (ablation) | dAE · MNIST fooling confidence 99% | thresholded accuracy: 87.84% | Hurt |
| dAE vs COOL vs 1-Class SVM | dAE · 1-class recognition | AUC: 0.753 -> 0.964 | Improved |
Closest papers by meaning
- Autoencoders that don't overfit towards the Identity (NeurIPS 2020 · similarity 0.87)
- Deconstructing Denoising Diffusion Models for Self-Supervised Learning (ICLR 2025 · similarity 0.84)
- Benign Overfitting in Adversarial Training of Neural Networks (ICML 2024 · similarity 0.84)
- Benign Overfitting and Grokking in ReLU Networks for XOR Cluster Data (ICLR 2024 · similarity 0.84)
- Rethinking Benign Overfitting in Two-Layer Neural Networks (ICML 2025 · similarity 0.83)
- Benign overfitting in leaky ReLU networks with moderate input dimension (NeurIPS 2024 · similarity 0.83)
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.
