Research map / Architectures and efficiency / RNNs and normalization
Supervised autoencoders: Improving generalization performance with unsupervised regularizers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 reconstruction error (supervised autoencoder) (ablation) | NN · Deterding test | accuracy: 52.50 -> 54.98 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-Sigmoid · Deterding test | accuracy: 87.00 -> 90.67 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-Sigmoid · SUSY test | accuracy: 76.90 -> 77.79 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-ReLu · Deterding test | accuracy: 72.29 -> 85.47 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-ReLu · SUSY test | accuracy: 65.27 -> 71.99 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-ReLu · MNIST test | accuracy: 97.30 -> 98.00 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-Kernel · Deterding test | accuracy: 74.85 -> 92.52 | Helped |
| add reconstruction error (supervised autoencoder) (ablation) | NN-Kernel · SUSY test | accuracy: 77.38 -> 77.27 | No clear effect |
| add reconstruction error, larger hidden dimension (ablation) | NN-ReLu · CIFAR-10 grayscale | accuracy | Helped |
| set predictive loss weight to 0 (pure autoencoder, two-stage) (ablation) | SAE deep CNN · CIFAR-10 | test accuracy | Hurt |
| prediction loss weight 0.01 vs 1.0 (ablation) | SAE deep CNN · CIFAR-10 | test accuracy | Helped |
| initialize NN with pretrained autoencoder | NN · Deterding/SUSY/MNIST | test accuracy | No clear effect |
Closest papers by meaning
- Regularized linear autoencoders recover the principal components, eventually (NeurIPS 2020 · similarity 0.90)
- Target-Embedding Autoencoders for Supervised Representation Learning (ICLR 2020 · similarity 0.90)
- Vector Quantization-Based Regularization for Autoencoders (AAAI 2020 · similarity 0.89)
- Understanding and Improving Interpolation in Autoencoders via an Adversarial Regularizer (ICLR 2019 · similarity 0.88)
- Scaling and evaluating sparse autoencoders (ICLR 2025 · similarity 0.87)
- To Grok or not to Grok: Disentangling Generalization and Memorization on Corrupted Algorithmic Datasets (ICLR 2024 · similarity 0.87)
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.