Take It or Leave It: Intent-Controlled Partial Optimal Transport
Researchers have introduced a new method called intent-controlled partial optimal transport (IC-POT) to address limitations in existing optimal transport techniques. Unlike traditional methods that enforce exact matching or global rejection of data points, IC-POT allows for more nuanced, pointwise rejection based on specific criteria. This approach can be framed as a dual problem involving local acceptance thresholds and can be solved by reformulating it as a balanced Kantorovich optimal transport problem. The method has shown practical utility in areas like positive-unlabeled learning and open-partial domain adaptation, improving performance by incorporating side information into the rejection process. AI
IMPACT Introduces a novel method for handling data rejection in optimal transport, potentially improving performance in machine learning tasks like domain adaptation.