Embedded Systems Design
summary
Amphis claims a semi‑supervised adversarial controller can directly generate Pareto‑optimal hardware configurations for reconfigurable processors, avoiding expensive trial evaluations. It trains two MLP generators and two discriminators as a duality between configuration→performance and performance→configuration, runs on the Sniper x86 simulator, and uses schedule biasing to focus on Pareto fronts. The implemented manager produces configurations within 3.02% energy of the optimum, is robust with scarce labeled data, and replaces slow per‑query predictors with a generative adversarial controller.
AI-written summary · about
chapters (23)
- 0:00Start
- 37:08Baseline Methods for Comparison
- 37:59Sensitivity Analysis
- 41:59Background of Fpj and Non-Volatile Memories
- 43:05Non-Volatile Memories
- 48:46Approximation Method
- 56:23Srp versus Ssl
- 1:07:38What Is an Assertion Miner
- 1:08:40Limitations
- 1:32:40Gradient Upload
- 1:33:11Modeled Synchronization
- 1:34:47Poisson Model
- 1:35:57Computational Overhead
- 1:36:58Conclusion
- 1:40:55Ensemble Model
- 1:50:58Per Kernel Fine-Grained Analysis
- 1:53:19Computer Memory Costs
- 1:56:10Introduction
- 1:56:50Temporal Action Localization
- 1:58:53On Device Video Learning Framework
- 2:02:06Self-Adaptive Video Dividing Approach
- 2:03:36Interest Based Sampling Strategy
- 2:04:18Experimental Result
related talks
19mBoosting confidence in tiny AI deployment in embedded control systems
22mPreview: Hardware and Offloads
22mAlgorithm-Aware Hardware Optimization using E-Graph Rewriting: how should we marry(…)
51mAdventures in (Simulated) Asymmetric Scheduling
20mSLEIGHT: Real-time Closed-Loop Attack Analytics Framework for Holistic Intrusion...
47mEmbedded Systems Architecture
10mNative Parity for Development & Testing of SOAFEE Software-Defined Vehicle
56mSafety and Security