pu-learning

Non-negative Positive-Unlabeled (nnPU) and unbiased Positive-Unlabeled (uPU) learning reproductive code on MNIST and CIFAR10

[ICML2020] "Self-PU: Self Boosted and Calibrated Positive-Unlabeled Training" by Xuxi Chen, Wuyang Chen, Tianlong Chen, Ye Yuan, Chen Gong, Kewei Chen, Zhangyang Wang

Code and results accompanying our paper titled Mixture Proportion Estimation and PU Learning: A Modern Approach at Neurips 2021 (Spotlight)

A collection of notebooks that implement algorithms introduced in "Learning from positive and unlabeled data: a survey"

[MICCAI 2022] ShapePU: A New PU Learning Framework Regularized by Global Consistency for Scribble Supervised Cardiac Segmentation

uPU, nnPU and PN learning with Extra Trees classifier.

Official Tensorflow implementation for Deep Generative Positive-Unlabeled Learning under Selection Bias (VAE-PU) in CIKM 2020.

Source code & appendices accompanying the AAAI2022 paper "Unifying Knowledge Base Completion with PU Learning to Mitigate the Observation Bias"

A PU-learning tool on spark

Code to generate SenSet, a novel human lung senescence cell gene signature.

Software implementation of the manuscript "A Two-step Anomaly detection Based Method for PU Classification in Imbalanced Data Sets".

Implementation of the paper: Towards Improved Illicit Node Detection with Positive-Unlabelled Learning

PU Hellinger Trees is a technique for positive and unlabeled imbalanced data.

Mastercard Data Quest: detecting hidden micro-businesses among consumer cardholders with PU-learning and LightGBM on transaction data.

Semi-supervised learning using positive and unlabeled learning for hypertension risk prediction.

National Technological Olympiad AI 25/26 Finals solution. Recovering lost implicit feedback in a Recommender System using a Two-Stage pipeline and PU-learning. Score: 0,1442

ML pipeline that scores companies against an Ideal Customer Profile using firmographic signals: industry, headcount, seniority, and job function.

Finding fraud leads in 11.4M public PPP loans, validated against real DOJ prosecutions — honest about what works and what doesn't.

Production-grade financial crime intelligence platform: temporal graph ML, PU learning, governed GenAI investigation agents, context engineering, eval gates, OTel tracing, and decision lineage

Which UK companies' ownership disclosure is shaped like known concealment, from the disclosure alone. ICIJ + Companies House, PR-AUC 6.1x over base rate. FTI on Hopsworks.