Research map / Learning theory and algorithms / Prediction and decision losses
Wisdom of the Ensemble: Improving Consistency of Deep Learning Models
Lijing Wang, Dipanjan Ghosh, Maria Gonzalez Diaz, Ahmed Farahat, Mahbubul Alam, Chetan Gupta, Jiangzhuo Chen, Madhav Marathe
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 |
|---|---|---|---|
| Combine accuracy-selected snapshot learners across resampled training runs using dynamic pruning | DynSnap-cyc · CIFAR100 | ACC-CON (%) | Improved |
| Use DynSnap-cyc instead of unpruned Snapshot (β=1) | ResNet20 · CIFAR10 | CON (%): 89.88 -> 93.04 | Improved |
| Use DynSnap-cyc instead of unpruned Snapshot (β=1) | ResNet56 · CIFAR100 | ACC-CON (%): 66.14 -> 71.37 | Improved |
| Use DynSnap-cyc instead of unpruned Snapshot (β=1) | fastText · YAHOO!Answers | CON (%): 91.87 -> 92.88 | Improved |
| Replace SingleBase with an ensemble of component learners | deep classifiers · CIFAR10, CIFAR100, and YAHOO!Answers | CON (%): 4.5%-8.2%, 9.8%-19.3%, and 3.4%-4.5% | Improved |
| Replace SingleBase with MCDropout | base classifiers · CIFAR10, CIFAR100, and YAHOO!Answers | ACC (%) | Hurt |
| Increase ensemble size m (ablation) | DynSnap-cyc with AVG · CIFAR100 | performance | Improved |
| Increase pruning factor β, reducing pruning and admitting relatively poorer learners (ablation) | DynSnap-cyc with AVG · CIFAR100 | performance | Hurt |
| Increase snapshot count N (ablation) | DynSnap-cyc with AVG · CIFAR100 | performance | Improved |
| Increase snapshot count beyond N=20 (ablation) | DynSnap-cyc with AVG · CIFAR100 | performance: N=20 | No clear effect |
| Vary snapshot window size T/N (ablation) | DynSnap-cyc with AVG · CIFAR100 | performance | No clear effect |
| Remove the learning-rate schedule (ablation) | DynSnap-step, top-N snapshot · CIFAR100 | performance | Hurt |
Closest papers by meaning
- Selective Ensembles for Consistent Predictions (ICLR 2022 · similarity 0.87)
- Certifying Ensembles: A General Certification Theory with S-Lipschitzness (ICML 2023 · similarity 0.86)
- Improving self-training under distribution shifts via anchored confidence with theoretical guarantees (NeurIPS 2024 · similarity 0.85)
- Quantifying Uncertainty in Error Consistency: Towards Reliable Behavioral Comparison of Classifiers (NeurIPS 2025 · similarity 0.85)
- United We Stand: Using Epoch-Wise Agreement of Ensembles to Combat Overfit (AAAI 2024 · similarity 0.85)
- Improved Techniques for Training Consistency Models (ICLR 2024 · similarity 0.85)
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.
