Seminars

View all Seminars  |  Download ICal for this event

Adversary-Resilient Asynchronous Learning in Distributed Systems with Heterogeneous Measurements

Series: CSA Faculty Colloquium

Speaker: Prof. Gugan Thoppe, Assistant Professor, Dept. of CSA

Date/Time: Mar 01 16:00:00

Location: CSA Seminar Hall (Room No. 254, First Floor)

Abstract:
In distributed optimization , particularly within a framework involving a parameter server with multiple worker nodes, the presence of adversaries poses a significant challenge. Conventionally, resilience against such adversarial behavior is achieved by mandating all worker nodes to synchronously compute and communicate identical quantities (such as gradients at a shared point) to a central server. The server then employs robust aggregation methods, such as median-based techniques, to exclude malicious inputs and update the solution estimate. Although this methodology has been validated through various studies, it proves inadequate for applications with inherent asynchronous and heterogeneous operations, such as network tomography. For instance, in network tomography, worker nodes might independently assess packet delays across diverse network paths. In this talk, we will introduce an innovative algorithm that utilizes observation matrices to support online learning with asynchronous and heterogeneous measurements in adversarial settings. We will explore the algorithms operational framework, alongside its theoretical underpinnings, demonstrating convergence and providing guarantees on the convergence rate. This algorithm not only addresses the limitations of traditional methods in complex, real-world applications but also broadens the horizon for distributed optimization under adversarial conditions.

Speaker Bio:
Gugan Thoppe is an Assistant Professor in the Computer Science and Automation department at the Indian Institute of Science since 2019. He is also an Associate Researcher at the Robert Bosch Centre, IIT Madras. He received his Ph.D. in 2016 from the Tata Institute of Fundamental Research (TIFR), Mumbai. Following this, he completed postdoctoral research at two places: Technion Institute of Technology, Israel (2015-17) and Duke University, USA (2017-19). His research is supported by the Cefipra Indo-French grant, the Walmart CSR grant, and the DST-SERBs core research grant. He is also involved in collaborative research with NPCI, the national payments corporation of India. He is also the winner of the Pratiksha trusts young investigator award, the IISc Award for Excellence in Teaching, the TIFR award for the best Ph.D. thesis, and two IBM Ph.D. fellowships. His research interests include reinforcement learning, online learning, stochastic approximation, and random topology.

Host Faculty: Prof. Sumit Kumar Mandal