Session
Practical EdgeAI Infrastructure and Applications: Efficiency and Resilience in the Wild
Event TypeResearch Manuscript
AI
RISC-V
AI/ML Application and Infrastructure
TimeWednesday, July 12th10:30am - 12:00pm PDT
Location3003, 3rd Floor
DescriptionThis session presents four papers that look at the practical design aspects as well as applications of EdgeAI. To tackle the tremendous amount of data generated, the first paper presents a hardware-friendly neural autoregressive framework to improve lossless compression performance. The second paper formulates the microarchitecture design space exploration as Bayesian optimization, and captures the parameter dependencies using graph representation learning. The third paper addresses the burdensome chip HW design requirements extraction task and uses ML to translate timing-diagram images into a symbolic formal specification. The fourth paper proposes enhancing hardware malware detectors (HMD) by leveraging approximate computing techniques to achieve adversarial resiliency.
Presentations
| 10:30am - 10:40am PDT | Lightning Talk: Trends and Challenges in AI for Design: Prediction, Optimization and Reasoning | |
| 10:40am - 10:55am PDT | Faster and stronger Lossless Compression with Optimized Autoregressive framework | |
| 10:55am - 11:10am PDT | Graph Representation Learning for Microarchitecture Design Space Exploration | |
| 11:10am - 11:25am PDT | TD-Magic: From Pictures of Timing Diagrams to Formal Specifications | |
| 11:25am - 11:40am PDT | Stochastic-HMDs: Adversarial-Resilient Hardware Malware Detectors via Undervolting |
