Research map / Label-efficient and robust learning
Recognition and long-tail: research map
6,306 papers on Recognition and long-tail in Label-efficient and robust learning, from ICML, NeurIPS, ICLR, CVPR, AAAI, ICCV and ECCV, and the journals TPAMI, IJCV and JMLR (2016–2026), grouped into 6 clusters and 27 approaches. The busiest year so far is 2026.
What it is
A broad topic about making recognition systems work in practice: recognising faces and people, explaining why a model decided something, and training when a few classes have many examples and most classes have very few (the "long tail").
Terms you will see on this page:
- Long-tailed recognition
- classification where a few head classes dominate the data and many tail classes are rare.
- Person re-identification
- matching the same person across different cameras.
- Concept bottleneck
- a model that first predicts human-readable concepts, then the label from them.
Start reading with the most cited papers here:
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · 7,708 citations)
- Unsupervised Feature Learning via Non-Parametric Instance Discrimination (CVPR 2018 · 3,635 citations)
Within Label-efficient and robust learning, its share held steady from 43.7% in 2023–24 to 42.9% in 2025–26 (1,253 → 1,312 papers at ICML, NeurIPS, ICLR, CVPR and AAAI, the venues with data for all four years).
Explore Recognition and long-tail in the interactive map
Working on something in this topic? Describe your idea in scime atlas to see which approach it falls under, the closest papers by meaning and how crowded the spot has become.
What changed recently
Approaches whose share of the topic moved most between 2023–24 and 2025–26 (ICML, NeurIPS, ICLR, CVPR and AAAI):
- Dense features and foundation priors rising: 12% → 15% of the topic (148 → 196 papers)
- Uncertainty-aware multi-view learning rising: 1% → 4% of the topic (12 → 51 papers)
- Contrastive representation theory rising: 10% → 12% of the topic (127 → 157 papers)
- Multilingual language representation pretraining falling: 9% → 5% of the topic (110 → 65 papers)
- Loss design and classifier theory falling: 5% → 3% of the topic (58 → 40 papers)
Papers from 2025–2026 that are already cited most:
- DAMMFND: Domain-Aware Multimodal Multi-view Fake News Detection (AAAI 2025 · 26 citations)
- CLOC: Contrastive Learning for Ordinal Classification with Multi-Margin N-pair Loss (CVPR 2025 · 15 citations)
- A Tale of Two Classes: Adapting Supervised Contrastive Learning to Binary Imbalanced Datasets (CVPR 2025 · 12 citations)
- Mixture-of-Attack-Experts with Class Regularization for Unified Physical-Digital Face Attack Detection (AAAI 2025 · 11 citations)
Approaches and key papers
Representation learning and transfer · 3,023 papers
Keywords: representations, transfer, unsupervised
Approaches in this cluster:
- Contrastive representation theory (558 papers)
Analyzes unsupervised and contrastive representation learning, disentanglement and invariance. - Correspondence and causal representation learning (690 papers)
Learns visual correspondence and transportable representations by matching multiple representations. - Dense features and foundation priors (559 papers)
Learns dense or tabular features using synthetic priors, hierarchical contrast and feature re-embedding. - Data augmentation and generalization bias (462 papers)
Studies augmentation, data properties and influence-based fixes for generalization and dataset bias. - Multilingual language representation pretraining (531 papers)
Pretrains cross-lingual language models and augmentation for transferable text representations. - Recommendation representation learning (223 papers)
Improves embeddings for recommender systems via multi-interest, inductive and collapse-aware modeling.
Most cited and most cited since 2024:
- Unsupervised Feature Learning via Non-Parametric Instance Discrimination (CVPR 2018 · 3,635 citations)
- Deep Metric Learning via Lifted Structured Feature Embedding (CVPR 2016 · 1,740 citations)
- Depth Anything V2 (NeurIPS 2024 · 525 citations)
- RoMa: Robust Dense Feature Matching (CVPR 2024 · 263 citations)
Classification with weak or partial labels · 1,640 papers
Keywords: label, classification, classifier
Approaches in this cluster:
- Partial and multi-label learning (449 papers)
Learns from partially annotated labels using class-aware, margin-based and representation blending approaches. - Loss design and classifier theory (367 papers)
Proposes margin losses, hierarchy-aware objectives and generative versus discriminative classifier analysis. - Group-robust classification (287 papers)
Improves worst-group and subclass performance via distribution balancing, priors and invariant augmentation. - Learning from unlabeled and positive-unlabeled data (304 papers)
Trains classifiers from positive-unlabeled or multiple unlabeled datasets with consistent risk estimators. - Confidence calibration objectives (143 papers)
Calibrates network confidence with soft objectives, pairwise constraints and ensemble methods. - Deep active learning (90 papers)
Selects informative samples for labeling using ensembles, submodular measures and supervision-level optimization.
Most cited and most cited since 2024:
- ChestX-ray8: Hospital-Scale Chest X-Ray Database and Benchmarks on Weakly-Supervised Classification and Localization of Common Thorax Diseases (CVPR 2017 · 3,529 citations)
- Multi-Task Learning Using Uncertainty to Weigh Losses for Scene Geometry and Semantics (CVPR 2018 · 3,007 citations)
- Adaptive Multi-Modal Cross-Entropy Loss for Stereo Matching (CVPR 2024 · 36 citations)
- Multi-Label Supervised Contrastive Learning (AAAI 2024 · 35 citations)
Face recognition and person re-identification · 788 papers
Keywords: face, person identification, facial
Approaches in this cluster:
- Margin-based face recognition losses (223 papers)
Shapes face embeddings with angular margins, adaptive margins and quality-aware losses. - Supervised person re-identification (221 papers)
Learns discriminative person features with similarity learning, part features and bias removal. - Cross-modality and occlusion re-identification (146 papers)
Handles infrared, occluded and training-free matching for person re-identification. - Fair and quality-aware face recognition (128 papers)
Uses quality-adaptive margins, unified thresholds and fairness adapters in face recognition. - Uncertainty-aware facial expression recognition (70 papers)
Models label ambiguity and bias for facial expression recognition.
Most cited and most cited since 2024:
- ArcFace: Additive Angular Margin Loss for Deep Face Recognition (CVPR 2019 · 7,708 citations)
- SphereFace: Deep Hypersphere Embedding for Face Recognition (CVPR 2017 · 3,010 citations)
- Implicit Discriminative Knowledge Learning for Visible-Infrared Person Re-Identification (CVPR 2024 · 76 citations)
- Shallow-Deep Collaborative Learning for Unsupervised Visible-Infrared Person Re-Identification (CVPR 2024 · 69 citations)
Interpretability and explanations · 459 papers
Keywords: explanations, concept, interpretable
Approaches in this cluster:
- Self-explaining and visual explanation networks (147 papers)
Builds networks that produce explanations through spatial attention, self-explaining design and explanation-based augmentation. - Prototype and concept dictionary models (110 papers)
Uses prototypical parts and sparse concept dictionaries for interpretable predictions. - Post-hoc explanation analysis (100 papers)
Evaluates and improves post-hoc explanations to expose spurious features and user needs. - Feature attribution methods (64 papers)
Formalizes and tests feature attribution through necessity, sufficiency and gradient-based tests. - Concept bottleneck models (38 papers)
Predicts through human-interpretable concept layers, including interactive, probabilistic and language-guided variants.
Most cited and most cited since 2024:
- Network Dissection: Quantifying Interpretability of Deep Visual Representations (CVPR 2017 · 1,404 citations)
- This Looks Like That: Deep Learning for Interpretable Image Recognition (NeurIPS 2019 · 560 citations)
- Deep Unfolded Network with Intrinsic Supervision for Pan-Sharpening (AAAI 2024 · 37 citations)
- SLICE: Stabilized LIME for Consistent Explanations for Image Classification (CVPR 2024 · 17 citations)
Long-tailed recognition · 222 papers
Keywords: long tailed, classes, tailed recognition
Approaches in this cluster:
- Decoupled and calibrated long-tail training (118 papers)
Rebalances optimization and calibrates statistics between head and tail classes. - Classifier rebalancing and margin losses (71 papers)
Separates representation and classifier learning and applies balancing margins or weight rebalancing. - Contrastive learning for imbalanced data (33 papers)
Uses contrastive and anchor-based regularization for imbalanced and multi-label long-tailed data.
Most cited and most cited since 2024:
- Class-Balanced Loss Based on Effective Number of Samples (CVPR 2019 · 2,852 citations)
- Large-Scale Long-Tailed Recognition in an Open World (CVPR 2019 · 1,159 citations)
- Feature Fusion from Head to Tail for Long-Tailed Visual Recognition (AAAI 2024 · 52 citations)
- SFC: Shared Feature Calibration in Weakly Supervised Semantic Segmentation (AAAI 2024 · 51 citations)
Multi-view learning · 174 papers
Keywords: multi view, views, view clustering
Approaches in this cluster:
- Uncertainty-aware multi-view learning (128 papers)
Fuses views using uncertainty estimates, consensus networks and openness-aware designs. - Incomplete multi-view multi-label learning (46 papers)
Imputes embeddings and shares label-guided representations for incomplete multi-view multi-label data.
Most cited and most cited since 2024:
- COMPLETER: Incomplete Multi-View Clustering via Contrastive Prediction (CVPR 2021 · 441 citations)
- Multi-Level Feature Learning for Contrastive Multi-View Clustering (CVPR 2022 · 388 citations)
- Incomplete Contrastive Multi-View Clustering with High-Confidence Guiding (AAAI 2024 · 109 citations)
- Decoupled Contrastive Multi-View Clustering with High-Order Random Walks (AAAI 2024 · 92 citations)
Related topics in Label-efficient and robust learning
- Noisy labels (576)
- Zero-/few-shot and meta-learning (1,029)
- Test-time adaptation (437)
- OOD and anomaly detection (836)
- Semi-supervised learning (897)
- Self-supervised and contrastive (1,267)
- Domain adaptation and generalization (2,012)
- Knowledge distillation (783)
