Code for Domain Adaptation under Open Set Label Shift
This repository contains code and results for our paper titled Mixture Proportion Estimation and PU Learning: A Modern Approach which was accepted at NeurIPS 2021. Given only Positive (P) and Unlabeled (U) data, containing both positives and negative samples, we propose new approaches to estimate the fraction of positives in unlabeled and learn positive vs negatives classifiers.
This repository houses the dataset described in the paper Learning the Difference that Makes a Difference with Counterfactually-Augmented Data. Given documents and their initial labels, we tasked humans to (i) revise each document to accord with a counterfactual target label, subject to producing revisions that (ii) result in internally consistent documents and (iii) avoid any gratuitous changes to facts that are semantically unrelated to the applicability of the label.
An interactive open-source book teaching deep learning from the basics through to advanced topics, taught entirely through Jupyter notebooks. We set out to create a resource that could (i) be freely available for everyone; (ii) offer sufficient technical depth to provide a starting point on the path to actually becoming an applied machine learning scientist; (iii) include runnable code, showing readers how to solve problems in practice; (iv) allow for rapid updates, both by us and also by the community at large; and (v) be complemented by a forum for interactive discussion of technical details and to answer questions.