Research map / Label-efficient and robust learning / Recognition and long-tail
Stimulative Training of Residual Networks: A Social Psychology Perspective of Loafing
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 randomly sampled sub-network supervision with KL-divergence loss | MobileNetV3 · CIFAR-10 | Main-network Top-1 accuracy: 95.72 -> 96.88 | Helped |
| Add randomly sampled sub-network supervision with KL-divergence loss | MobileNetV3 · CIFAR-100 | Main-network Top-1 accuracy: 77.39 -> 81.07 | Helped |
| Add randomly sampled sub-network supervision with KL-divergence loss | ResNet50 · CIFAR-100 | Main-network Top-1 accuracy: 76.53 -> 81.06 | Helped |
| Add randomly sampled sub-network supervision with KL-divergence loss | ResNet50 · ImageNet validation | Main-network Top-1 accuracy: 76.1 -> 77.22 | Helped |
| Add randomly sampled sub-network supervision with KL-divergence loss | ResNet101 · ImageNet validation | Main-network Top-1 accuracy: 77.37 -> 78.58 | Helped |
| Add randomly sampled sub-network supervision with KL-divergence loss | MobileNetV3 · CIFAR-100 | Sub-network Top-1 accuracy (mean): 55.26 -> 80.01 | Helped |
| Add randomly sampled sub-network supervision with KL-divergence loss | MobileNetV3 · CIFAR-100 | Sub-network Top-1 accuracy standard deviation: 13.37 -> 0.59 | Helped |
| Use ordered rather than raw or stochastic residual sampling (ablation) | Stimulative Training on MobileNetV3 · CIFAR-100 | Top-1 accuracy: 5.77% and 0.58% | Helped |
| Increase the number of sampled sub-networks used during training (ablation) | Stimulative Training on MobileNetV3 · CIFAR-100 | Sub-network Top-1 accuracy mean | Helped |
| Increase the number of sampled sub-networks used during training (ablation) | Stimulative Training on MobileNetV3 · CIFAR-100 | Sub-network Top-1 accuracy standard deviation | Helped |
| Use stimulative training instead of common training | MobileNetV3 · CIFAR-100 test-time deletion of layers | Top-1 accuracy drop: no more than 1% | Helped |
| Use stimulative training instead of common training | MobileNetV3 · CIFAR-100 test-time deletion of multiple layers | Top-1 accuracy drop: no more than 3% | Helped |
Closest papers by meaning
- Boosting Residual Networks with Group Knowledge (AAAI 2024 · similarity 0.87)
- Theoretical and Empirical Insights into the Origins of Degree Bias in Graph Neural Networks (NeurIPS 2024 · similarity 0.84)
- Grokking in Linear Estimators -- A Solvable Model that Groks without Understanding (ICLR 2024 · similarity 0.83)
- Overcoming Sparsity Artifacts in Crosscoders to Interpret Chat-Tuning (NeurIPS 2025 · similarity 0.83)
- Ghidorah: Towards Robust Multi-Scale Information Diffusion Prediction via Test-Time Training (AAAI 2025 · similarity 0.83)
- Behavioral Geometric Supervision Aligns Video Foundation Models with Human Social Perception (NeurIPS 2026 · 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.