Research map / Label-efficient and robust learning / Semi-supervised learning
Adaptive Conformal Guidance for Learning under Uncertainty
Rui Liu, Peng Gao, Yu Shen, Ming Lin, Pratap Tokekar
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 |
|---|---|---|---|
| add AdaConG conformal weighting to KD with a noisy teacher | KD, ResNet110 teacher to ResNet20 · CIFAR-100 with Gaussian noise domain shift | top-1 accuracy: 57.23 -> 66.53 | Improved |
| plain KD from an underperforming teacher | ResNet20 student · CIFAR-100 noisy | top-1 accuracy: 66.51 (scratch) -> 57.23 (KD) | Hurt |
| add AdaConG to feature-based KD (FitNet, PKT, FT) | KD methods · CIFAR-100 noisy homogeneous | top-1 accuracy: FT ResNet110: +0.11 (below 0.15 threshold); others +0.26 to +2.41 | Improved |
| add AdaConG to semi-supervised methods | FixMatch · CIFAR-10 40 labels | top-1 accuracy: 64.18 -> 70.16 | Improved |
| add AdaConG to FlexMatch | FlexMatch · CIFAR-100 2500 labels | top-1 accuracy: 63.59 -> 69.22 | Improved |
| use MSE instead of cross-entropy as guidance loss in SSL (ablation) | AdaConG in UDA/FixMatch/FlexMatch · CIFAR-100 | top-1 accuracy: FixMatch 400 labels 8.56 -> 9.82 | No clear effect |
| use hard weighting (w=1 only when prediction set is a singleton) (ablation) | AdaConG · CIFAR-100 KD ResNet110/ResNet20 | top-1 accuracy: 66.53 (soft) vs 66.52 (hard) for KD | No clear effect |
| use conformal quantile prediction sets instead of raw nonconformity score (ablation) | AdaConG KD · CIFAR-100 ResNet110/ResNet20 | top-1 accuracy: 64.17 -> 66.53 | Improved |
| use conformal prediction uncertainty instead of heuristic MC dropout, MSP or entropy | KD · CIFAR-100 ResNet110/ResNet20 | top-1 accuracy: 63.59 (MC dropout) -> 66.53 | Improved |
| increase temperature gamma in exponential weight up to 10 (ablation) | AdaConG KD · CIFAR-100 ResNet110/ResNet20 | top-1 accuracy | Improved |
| vary conformal error rate alpha (ablation) | AdaConG · CIFAR-100 KD and CIFAR-10 FixMatch | top-1 accuracy | No clear effect |
| add AdaConG to KD for steering prediction | KD · SullyChen steer prediction with noise | mean accuracy: 73.5 -> 76.8 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- FlexMatch — FlexMatch: Boosting Semi-Supervised Learning with Curriculum Pseudo Labeling (NeurIPS 2021) · better on 2 of 2 shared results
- FixMatch — FixMatch: Simplifying Semi-Supervised Learning with Consistency and Confidence (NeurIPS 2020) · better on 2 of 2 shared results
- EA-KD — EA-KD: Entropy-based Adaptive Knowledge Distillation (ICCV 2025) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Adaptive Classifier-Free Guidance via Dynamic Low-Confidence Masking (NeurIPS 2025 · similarity 0.86)
- Conformal Prediction in The Loop: A Feedback-Based Uncertainty Model for Trajectory Optimization (NeurIPS 2025 · similarity 0.85)
- Learning Polyhedral Conformal Sets for Robust Optimization (NeurIPS 2026 · similarity 0.84)
- Gradient Descent as Loss Landscape Navigation: a Normative Framework for Deriving Learning Rules (NeurIPS 2025 · similarity 0.84)
- AdaptNC: Adaptive Nonconformity Scores for Conformal Prediction under Distribution Shift (NeurIPS 2026 · similarity 0.84)
- Trust Guided Decision Transformer (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.
