CSE Doctoral Student Seminar: Son Dinh and Shali Jiang

Mar 24
12:30 p.m.
2 p.m.
Lopata Hall, Room 101

"Scheduling Platforms and Techniques for Parallel Soft Real-time Systems​"

Son Dinh
Adviser: Chris Gill

Multicore processors are becoming common nowadays. In order to take the advantage of multicore processors, programs need to be parallelized. Real-time applications are no exception from this trend, especially when real-time applications with high computation demand are emerging. However, the question of how to efficiently schedule parallel real-time tasks on multicore processors is still an open question. In this talk, we will experimentally investigate two common strategies for scheduling parallel applications, namely centralized scheduling and randomized work stealing. We also examine these scheduling approaches for soft real-time tasks when combining with federated scheduling, a real-time scheduling paradigm for parallel tasks which theoretically guarantees parallel tasks to meet their timing constraints (i.e., deadlines) when running on their own dedicated cores. 

"Efficient Nonmyopic Active Search"

Shali Jiang
Adviser: Roman Garnett

Active search is an active learning setting with the goal of identifying as many members of a given class as possible under a labeling budget. In this work, we first establish a theoretical hardness of active search, proving that no polynomial-time policy can achieve a constant factor approximation ratio with respect to the expected utility of the optimal policy. We also propose a novel, computationally efficient active search policy achieving exceptional performance on several real-world tasks. Our policy is nonmyopic, always considering the entire remaining search budget. It also automatically and dynamically balances exploration and exploitation consistent with the remaining budget, without relying on a parameter to control this tradeoff. We conduct experiments on diverse datasets from several domains: drug discovery, materials science, and a citation network. Our efficient nonmyopic policy recovers significantly more valuable points with the same budget than several alternatives from the literature, including myopic approximations to the optimal policy.