Research map / Architectures and efficiency / Neural architecture search
Neural Ensemble Search for Uncertainty Estimation and Dataset Shift
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 |
|---|---|---|---|
| Select an ensemble from varying architectures using NES-RE | Deep ensembles over the DARTS search space · CIFAR-10, CIFAR-100, and Tiny ImageNet; severity 0–5 | classification error (%): 10.8 -> 9.4; 33.2 -> 30.4; 37.5 -> 37.0 | Helped |
| Select an ensemble from varying architectures using NES-RS | Deep ensembles over NAS-Bench-201 · CIFAR-10, severity 3 and 5 | classification error (%): 27.1 -> 22.0; 46.3 -> 32.5 | Helped |
| Choose varying-architecture ensembles with NES-RE instead of NES-RS | NES over DARTS · CIFAR-10 and CIFAR-100 | predictive NLL | Helped |
| Select the final ensemble using shifted validation data | NES-RS and NES-RE · CIFAR-10, CIFAR-100, and Tiny ImageNet under dataset shift | robustness to dataset shift | Helped |
| Use varying architectures rather than a fixed architecture with ensemble selection (ablation) | DeepEns + ES versus NES on Tiny ImageNet · Tiny ImageNet | NLL | Helped |
| Add ensemble selection over initializations of a fixed architecture (ablation) | Deep ensembles on Tiny ImageNet · Tiny ImageNet | NLL | Helped |
| Use NES-RE with half the training budget of DeepEns + ES (RS) (ablation) | Tiny ImageNet · Tiny ImageNet | NLL: 200; 400 | Helped |
| Use NES-RS with half the training budget of DeepEns + ES (RS) (ablation) | Tiny ImageNet · Tiny ImageNet | NLL: 200; 400 | No clear effect |
| Use only unshifted validation data during NES-RE evolution (ablation) | NES-RE-0 · CIFAR-10 under dataset shift | NLL | Mixed |
| Reduce the validation set size, including to 10 samples (ablation) | NES · Dataset shift severities 0–5 | test loss: 10 | No clear effect |
| Replace greedy forward selection with alternative ensemble selection methods (ablation) | NES · Experiments reported across datasets | test performance | Hurt |
| Use weighted averaging instead of forward step-wise selection (ablation) | NES · Experiments reported across datasets | test performance | No clear effect |
Closest papers by meaning
- Deep Ensembles Work, But Are They Necessary? (NeurIPS 2022 · similarity 0.91)
- Hyperparameter Ensembles for Robustness and Uncertainty Quantification (NeurIPS 2020 · similarity 0.90)
- Packed Ensembles for efficient uncertainty estimation (ICLR 2023 · similarity 0.90)
- Disentangling the Predictive Variance of Deep Ensembles through the Neural Tangent Kernel (NeurIPS 2022 · similarity 0.89)
- DIBS: Diversity Inducing Information Bottleneck in Model Ensembles (AAAI 2021 · similarity 0.89)
- Theoretical Limitations of Ensembles in the Age of Overparameterization (ICML 2025 · 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.