Showing 1–2 of 2 results for author: Su, A

ConQUR: Mitigating Delusional Bias in Deep Q-learning

Authors: Andy Su, Jayden Ooi, Tyler Lu, Dale Schuurmans, Craig Boutilier

Abstract: Delusional bias is a fundamental source of error in approximate Q-learning. To date, the only techniques that explicitly address delusion require comprehensive search using tabular value estimates. In this paper, we develop efficient methods to mitigate delusional bias by training Q-approximators with labels that are "consistent" with the underlying greedy policy class. We introduce a simple penal… ▽ More Delusional bias is a fundamental source of error in approximate Q-learning. To date, the only techniques that explicitly address delusion require comprehensive search using tabular value estimates. In this paper, we develop efficient methods to mitigate delusional bias by training Q-approximators with labels that are "consistent" with the underlying greedy policy class. We introduce a simple penalization scheme that encourages Q-labels used across training batches to remain (jointly) consistent with the expressible policy class. We also propose a search framework that allows multiple Q-approximators to be generated and tracked, thus mitigating the effect of premature (implicit) policy commitments. Experimental results demonstrate that these methods can improve the performance of Q-learning in a variety of Atari games, sometimes dramatically. △ Less

Submitted 27 February, 2020; originally announced February 2020.

Branch: An interactive, web-based tool for testing hypotheses and develo** predictive models

Authors: Karthik Gangavarapu, Vyshakh Babji, Tobias Meißner, Andrew I. Su, Benjamin M. Good

Abstract: Branch is a web application that provides users with no programming with the ability to interact directly with large biomedical datasets. The interaction is mediated through a collaborative graphical user interface for building and evaluating decision trees. These trees can be used to compose and test sophisticated hypotheses and to develop predictive models. Decision trees are evaluated based on… ▽ More Branch is a web application that provides users with no programming with the ability to interact directly with large biomedical datasets. The interaction is mediated through a collaborative graphical user interface for building and evaluating decision trees. These trees can be used to compose and test sophisticated hypotheses and to develop predictive models. Decision trees are evaluated based on a library of imported datasets and can be stored in a collective area for sharing and re-use. Branch is hosted at http://biobranch.org/ and the open source code is available at http://bitbucket.org/sulab/biobranch/. △ Less

Submitted 30 September, 2015; v1 submitted 22 September, 2015; originally announced September 2015.