DSAI seminar with Arman Rahbar

Abstract

In this talk, I will focus on cost-efficient online decision making where the goal is making decisions (e.g., predicting labels) on a stream of data points with a low cost of acquiring feature values. The talk will be mainly based on our paper published in IJCAI2023 titled "Efficient Online Decision Tree Learning with Active Feature Acquisition". In this paper, we propose a novel framework for learning an online classifier (a decision tree) that can learn from streaming data points with a subset of feature values and thus with a low cost. Our framework consists of an active planning oracle embedded in an online learning scheme. Specifically, we employ surrogate information acquisition functions to actively query feature values with a minimal cost, while using a posterior sampling scheme to maintain a low regret for online prediction. At the end of the talk, I will briefly present a combinatorial multi-armed bandit formulation of this problem that leads to a tighter regret bound which is linear in the number of features.

About the speaker

Arman Rahbar is a WASP Ph.D. student at Chalmers University of Technology and is working under the supervision of Morteza Chehreghani. He received his bachelor's and master's degrees in computer engineering from Sharif University of Technology and Amirkabir University of Technology, respectively. Arman's research is mainly focused on representation learning problems that need to deal with the lack of sufficient available information.

This is a seminar from the DSAI seminars series usually held every Monday at 14:00 by the Data Science and AI division. The seminars are usually hybrid. No registration is required.