Research map / Architectures and efficiency / RNNs and normalization
On Power Laws in Deep Ensembles
Ekaterina Lobacheva, Nadezhda Chirkova, Maxim Kodryan, Dmitry P Vetrov
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 |
|---|---|---|---|
| apply temperature before averaging instead of after averaging (ablation) | deep ensemble calibration · CIFAR-10/100, VGG, WideResNet | power-law fit of (C)NLL vs ensemble size | Improved |
| calibrate NLL with optimal temperature instead of tau=1 (ablation) | single VGG on CIFAR-100 · network size sweep | NLL vs network size | Improved |
| replace a single large network with a memory split ensemble of medium networks of same total parameters | VGG/WideResNet · CIFAR-100 | CNLL | Improved |
| increase member network size when ensembling | WideResNet/VGG ensembles · CIFAR-100 | CNLL | Hurt |
| fit power law on ensembles of size 1-4 to predict larger ensembles | CNLL power law · CIFAR-100 | prediction RMSE | Improved |
| increase temperature in non-calibrated NLL ensemble (ablation) | NLL_n(tau) · VGG CIFAR-100 | ensembling gain parameter b | Hurt |
Closest papers by meaning
- Scaling Laws for Hyperparameter Optimization (NeurIPS 2023 · similarity 0.87)
- Analyzing Neural Scaling Laws in Two-Layer Networks with Power-Law Data Spectra (ICLR 2025 · similarity 0.87)
- Deep Linear Network Training Dynamics from Random Initialization: Data, Width, Depth, and Hyperparameter Transfer (ICML 2025 · similarity 0.85)
- Theoretical Limitations of Ensembles in the Age of Overparameterization (ICML 2025 · similarity 0.85)
- A Dynamical Model of Neural Scaling Laws (ICML 2024 · similarity 0.85)
- Investigating the Overlooked Hessian Structure: From CNNs to LLMs (ICML 2025 · similarity 0.84)
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.
