Departmental Papers (CIS)

Date of this Version

2-2020

Document Type

Conference Paper

Comments

International Conference on Learning Representations (ICLR 2020), Addis Ababa, Ethiopia, April 26-30, 2020

Abstract

We propose an algorithm combining calibrated prediction and generalization bounds from learning theory to construct confidence sets for deep neural networks with PAC guarantees---i.e., the confidence set for a given input contains the true label with high probability. We demonstrate how our approach can be used to construct PAC confidence sets on ResNet for ImageNet, a visual object tracking model, and a dynamics model for the half-cheetah reinforcement learning problem.

Subject Area

CPS Safe Autonomy

Publication Source

International Conference on Learning Representations (ICLR 2020)

Keywords

PAC, confidence sets, classification, regression, reinforcement learning

Bib Tex

@inproceedings{ Park2020PAC, title={PAC Confidence Sets for Deep Neural Networks via Calibrated Prediction}, author={Sangdon Park and Osbert Bastani and Nikolai Matni and Insup Lee}, booktitle={International Conference on Learning Representations}, year={2020}, url={https://openreview.net/forum?id=BJxVI04YvB} }

Share

COinS
 

Date Posted: 03 March 2020

This document has been peer reviewed.