Research map / Label-efficient and robust learning / Test-time adaptation
Predicting With Confidence on Unseen Distributions
Devin Guillory, Vaishaal Shankar, Sayna Ebrahimi, Trevor Darrell, Ludwig Schmidt
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 |
|---|---|---|---|
| predict accuracy with DoC instead of the average-confidence baseline AC | ImageNet classifiers · Natural shifts (calibrated on Syn1) | MAE: 0.087 -> 0.050 | Improved |
| use DoC instead of naive base accuracy | ImageNet classifiers · Natural shifts (calibrated on Syn1) | MAE: 0.234 -> 0.050 | Improved |
| keep only maximum softmax probability (DoC) instead of entropy (DoE) (ablation) | DoC · Natural shifts (calibrated on Syn1) | MAE: 0.075 -> 0.050 | Improved |
| use Frechet distance instead of AC | ImageNet classifiers · Natural shifts (calibrated on Syn1) | MAE: 0.087 -> 0.113 | Hurt |
| use discriminative A-proxy distance instead of AC | ImageNet classifiers · Natural shifts (calibrated on Syn1) | MAE: 0.087 -> 0.335 | Hurt |
| use DoC instead of AC on synthetic-to-synthetic | ImageNet classifiers · Synthetic Syn2 shifts (calibrated on Syn1) | MAE: 0.075 -> 0.039 | Improved |
| use DoC instead of Frechet distance on synthetic-to-synthetic | ImageNet classifiers · Synthetic Syn2 shifts (calibrated on Syn1) | MAE: 0.0384 -> 0.039 | No clear effect |
| use DoC instead of AC on adversarial ImageNet-A | ImageNet classifiers · ImageNet-A (calibrated on Syn1) | MAE: 0.476 -> 0.388 | Improved |
| calibrate DoC on natural shifts instead of Syn1 for ImageNet-A (ablation) | DoC · ImageNet-A | MAE: 0.388 -> 0.292 | Improved |
| calibrate DoE on natural shifts instead of Syn1 for ImageNet-A (ablation) | DoE · ImageNet-A | MAE: 0.524 -> 0.371 | Improved |
| calibrate DoC on Syn2 instead of Syn1 (ablation) | DoC · Natural shifts | MAE: 0.050 -> 0.066 | Hurt |
| calibrate DoC on natural shifts for synthetic target (ablation) | DoC · Synthetic Syn2 shifts | MAE: 0.039 -> 0.096 | Hurt |
Closest papers by meaning
- A Fine-Grained Analysis on Distribution Shift (ICLR 2022 · similarity 0.87)
- Measuring Robustness to Natural Distribution Shifts in Image Classification (NeurIPS 2020 · similarity 0.87)
- Leveraging unlabeled data to predict out-of-distribution performance (ICLR 2022 · similarity 0.86)
- How robust is unsupervised representation learning to distribution shift? (ICLR 2023 · similarity 0.86)
- Multiply Robust Estimation for Local Distribution Shifts with Multiple Domains (ICML 2024 · similarity 0.85)
- "What is Different Between These Datasets?" A Framework for Explaining Data Distribution Shifts (NeurIPS 2026 · 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.
