Research map / Label-efficient and robust learning / Semi-supervised learning
HybridNet: Classification and Reconstruction Cooperation for Semi-Supervised Learning
Thomas Robert, Nicolas Thome, Matthieu Cord
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 the full HybridNet training loss to the two-branch architecture (ablation) | HybridNet architecture · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 80.3 -> 86.6 (4000 labels); 63.2 -> 74.1 (1000 labels) | Improved |
| use the two-branch HybridNet architecture instead of the plain auto-encoder with only the basic losses (ablation) | Auto-encoder baseline · CIFAR-10 (1000 labels, ConvLarge) | accuracy (%): 65.0 -> 63.2 | Hurt |
| add stability loss to the classification baseline (ablation) | Classification baseline · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 71.5 -> 74.6 | Improved |
| add stability loss to the auto-encoder baseline (ablation) | Auto-encoder · CIFAR-10 (1000 labels, ConvLarge) | accuracy (%): 65.0 -> 71.8 | Improved |
| add final hybrid reconstruction loss only (ablation) | Classification baseline · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 71.5 -> 72.4 | Improved |
| add intermediate reconstruction losses (ablation) | HybridNet with stability and reconstruction · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 77.7 -> 80.8 | Improved |
| add branch-balanced reconstruction without intermediate reconstruction (ablation) | HybridNet with stability and reconstruction · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 77.7 -> 77.4 | No clear effect |
| add branch-balanced reconstruction on top of intermediate reconstructions (ablation) | HybridNet with stability and intermediate reconstruction · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 80.8 -> 81.6 | Improved |
| add balancing and intermediate reconstruction terms without stability (ablation) | HybridNet without stability · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 72.4 -> 75.2 | Improved |
| add stability loss to full reconstruction losses (ablation) | HybridNet reconstruction losses · CIFAR-10 (2000 labels, ConvLarge) | accuracy (%): 75.2 -> 81.6 | Improved |
| add full HybridNet loss on STL-10 (ablation) | Classification baseline · STL-10 (1000 labels, ConvLarge-like) | accuracy (%): 65.6 -> 74.1 | Improved |
| add stability loss to classification on STL-10 (ablation) | Classification baseline · STL-10 (1000 labels, ConvLarge-like) | accuracy (%): 65.6 -> 69.8 | Improved |
Later papers that compared with this method
1 later accepted paper reports HybridNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Repetitive Reprediction Deep Decipher for Semi-Supervised Learning (AAAI 2020 · better on 1 of 1 shared result)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Big Self-Supervised Models are Strong Semi-Supervised Learners (NeurIPS 2020 · similarity 0.88)
- Semi-Supervised Vision Transformers (ECCV 2022 · similarity 0.87)
- DualNet: Learn Complementary Features for Image Recognition (ICCV 2017 · similarity 0.87)
- Barely-Supervised Learning: Semi-supervised Learning with Very Few Labeled Images (AAAI 2022 · similarity 0.87)
- SimPLE: Similar Pseudo Label Exploitation for Semi-Supervised Classification (CVPR 2021 · similarity 0.87)
- Diffusion Models and Semi-Supervised Learners Benefit Mutually with Few Labels (NeurIPS 2023 · similarity 0.86)
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.
