REHASH: A Flexible, Developer Focused, Heuristic Adaptation Platform for Intermittently Powered ...
summary
A task-based adaptive runtime called REHASH runtime and a developer simulator REHASH Explorer adapt intermittently powered sensors by monitoring lightweight signals (on-time/off-time/task-count) and applying simple heuristics that compare current values to recent averages. The system boosts throughput: the ML workload runs up to 46% more classifications with ~5% accuracy loss, the activity-recognition app captures 76% more samples, and all evaluated apps show higher throughput versus non-adaptive baselines while keeping low memory and energy overhead.
AI-written summary · about
related talks
06mIntermittent Learning: On-Device Machine Learning on Intermittently Powered System
01h 12mPower, Energy and Analysis
51mAdventures in (Simulated) Asymmetric Scheduling
22mPreview: Hardware and Offloads
14mUnveiling Heisenbugs with Diversified Execution
20mALERT: Accurate Learning for Energy and Timeliness
SpaceExit: Enabling Efficient Adaptive Computing in Space with Early Exits
01h 01mIntermittent Computing