event

Ph.D. Proposal Oral Exam - Jamin Seo

Title:  Workload-aware Cross-stack Co-optimization for Multi-task Multi-model Edge Machine Learning

Committee:

Dr. Krishna, Advisor

Dr. Kwon, Co-Advisor

Dr. Hao, Chair

Dr. Lin

Media

No media selected

Summary

The objective of the proposed research is to explore cross-stack co-optimization on designing edge accelerators tailored for multi-task multi-model (MTMM) machine learning (ML) in edge, namely AR/VR and drone use cases. These workloads introduce unique application-level characteristics and constraints that distinguish the optimization goals than traditional ML accelerator designs. The proposed research aims to investigate workload-aware cross-stack co-optimization across key accelerator components: GEMM compute with constrained dataflow flexibility at minimal hardware overhead; non-GEMM compute with hardware-friendly numerical approximation algorithms; and on-chip memory management with dynamicity-aware placement strategies.

Details

Thursday
May 1 2025
09:00am - 11:00am

Location: Room 1212, Klaus

In campus calendar: No

Sidebar Content

No sidebar content

Groups

Status

  • Workflow status: Published
  • Created by: Daniela Staiculescu
  • Created: 04/26/2025
  • Modified By: Daniela Staiculescu
  • Modified: 04/26/2025

Related Links

  • No related links.

Categories

Keywords

Target Audience