Awesome ML Positive Unlabeled learning

Papers

• Papers \w code
  • Overview
    • Theoretical Comparisons of Positive-Unlabeled Learning against Positive-Negative Learning(2016)
    • Learning from positive and unlabeled data: a survey(2018)
    • Positive and unlabeled learning algorithms and applications: A survey(2019)
    • Exploring Positive Unlabeled Machine Learning(2021)
    • Positive Unlabeled Learning - Synthesis Lectures on AI and ML(2022)
    • A recent survey on instance-dependent positive and unlabeled learning(2025)
  • Cost-sensitive methods
    • Learning Classifiers from Only Positive and Unlabeled Data(2008)[pulearn]
    • Analysis of Learning from Positive and Unlabeled Data(2014)
    • A Modified Logistic Regression for Positive and Unlabeled Learning(2019) [video]
    • DEDPUL: Difference-of-Estimated-Densities-based Positive-Unlabeled Learning(2019)[source]
    • PULSNAR: Positive Unlabeled Learning Selected Not At Random(2023)
    • PUe: Biased Positive-Unlabeled Learning Enhancement by Causal Inference(2023)[source]
    • Contrastive Approach to Prior Free Positive Unlabeled Learning(2024)
    • Positive and Unlabeled Learning with Controlled Probability Boundary Fence(2024)
    • A boosting framework for positive-unlabeled learning - AdaPU(2024)
    • Class prior estimation for positive-unlabeled learning when label shift occurs(2025)
  • Sample-selection/Two-step methods
    • Positive-unlabeled learning for disease gene identification(2012)
    • A bagging SVM to learn from positive and unlabeled examples(2013) [pulearn]
    • Revisiting Sample Selection Approach to Positive-Unlabeled Learning: Turning Unlabeled Data into Positive rather than Negative(2019)
    • Improving Positive Unlabeled Learning: Practical AUL Estimation and New Training Method for Extremely Imbalanced Data Sets(2020)
    • PULNS: Positive-Unlabeled Learning with Effective Negative Sample Selector(2021)[slides]
    • Positive-Unlabeled Learning by Latent Group-Aware Meta Disambiguation(2024)
  • Deep Learning methods
    • Positive-Unlabeled Learning with Non-Negative Risk Estimator - nnPU(2017)[source]
    • Variational approach for learning from positive and unlabeled data - VPU(2020)[source]
    • Improving Non-Negative Positive-Unlabeled Learning for News Headline Classification - FLPU(2023)
    • Multiscale Positive-Unlabeled Framework for AI-Generated Text Detection - MPU(2024)
  • Graph Neural Networks
    • Learning Graph Neural Networks with Positive and Unlabeled Nodes - LSDAN(2021)
    • PU-GNN: Polypharmacy Side-Effects Detection Based on Graph Neural Networks(2024)
    • Bootstrap Latent Prototypes for graph positive-unlabeled learning(2024)
    • Unraveling the impact of heterophilic structures on graph positive-unlabeled learning(2024)
    • Graph neural networks for positive and unlabeled learning: a rewiring approach(2025)
  • Anomaly Detection
    • Boosting Positive and Unlabeled Learning for Anomaly Detection With Multi-Features(2017)
    • Deep Positive-Unlabeled Anomaly Detection for Contaminated Unlabeled Data(2024)
    • Few-shot anomaly detection using positive unlabeled learning with cycle consistency - PUFAD(2024)

Implementations

• Frameworks
  • Python
    • nnPUlearning - Non-negative PU learning (Chainer)
    • pulearn - Elkan & Noto, Bagging-based methods
    • dedpul - Density-based PU learning
    • vpu - Variational PU learning (PyTorch)
    • PUe - Causal Inference PU (NeurIPS 2023)
    • pulse - Audio Signal Enhancement (ICASSP 2023 Best Paper)
    • Active_PU_Learning - Class Prior Estimation in Active PU
    • pu-learning - PyTorch implementation of nnPU/uPU
  • Spark
    • pu4spark

Tutorials

• An introductory tutorial to the "Learning from Positive and Unlabeled Data" field.
• tinyML Talks Phoenix: Positive Unlabeled Learning for Tiny ML
• Semi-Supervised Classification of Unlabeled Data (PU Learning)

Concepts

• Class prior and label frequency
• Cost-sensitive and sample-selection methods
• Inductive vs Transductive PU learning. a.k.a(?) (Single-training set vs case-control scenario)
• Labelling mechanism
• Assumptions
  • Partial Seperability
  • SCAR(Selected Completely At Random) - assumes all positive examples have equal probability of being labeled
  • Non-SCAR / Instance-dependent labeling - labeling probability depends on instance features
    • In healthcare: severe cases more likely to be diagnosed than mild ones
    • In e-commerce: popular items more likely to receive reviews
    • Examples far from decision boundary have higher labeling probability
    • Instance-dependent PU learning survey(2025)

Cases/Applications

• Positive Unlabeled Learing, MegaFon, ODS.ai(2020)[RU]
• FedPU: Federated Learning with Positive and Unlabeled Data(2022)
• PULSE: Positive-Unlabelled Learning for audio Signal Enhancement(2023) - ICASSP 2023 Best Paper Award
• Fairness-Aware Online Positive-Unlabeled Learning(2024)
• Applications of PU and NU Learning in Cybersecurity(2024)
• Positive-Unlabeled Learning in Implicit Feedback Recommendation Systems(2025)