We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Machine Learning

Title: AP-Perf: Incorporating Generic Performance Metrics in Differentiable Learning

Abstract: We propose a method that enables practitioners to conveniently incorporate custom non-decomposable performance metrics into differentiable learning pipelines, notably those based upon neural network architectures. Our approach is based on the recently developed adversarial prediction framework, a distributionally robust approach that optimizes a metric in the worst case given the statistical summary of the empirical distribution. We formulate a marginal distribution technique to reduce the complexity of optimizing the adversarial prediction formulation over a vast range of non-decomposable metrics. We demonstrate how easy it is to write and incorporate complex custom metrics using our provided tool. Finally, we show the effectiveness of our approach various classification tasks on tabular datasets from the UCI repository and benchmark datasets, as well as image classification tasks. The code for our proposed method is available at this https URL
Comments: Appears in the Proceedings of the 23rd International Conference on Artificial Intelligence and Statistics (AISTATS) 2020
Subjects: Machine Learning (cs.LG); Machine Learning (stat.ML)
Cite as: arXiv:1912.00965 [cs.LG]
  (or arXiv:1912.00965v2 [cs.LG] for this version)

Submission history

From: Rizal Fathony [view email]
[v1] Mon, 2 Dec 2019 17:53:05 GMT (99kb)
[v2] Tue, 3 Mar 2020 13:58:44 GMT (127kb,D)

Link back to: arXiv, form interface, contact.