EdgeCraft: Automated Model Crafting for Edge IoT
EdgeCraft uses an LLM to craft deployable edge ML models that meet quality, latency, and energy goals.
EdgeCraft is an LLM-driven system that turns a high-level intent into a deployable edge ML artifact, covering representation, model design, domain training, and runtime customization. A constraint-aware synthesis tree uses measured gaps in quality, latency, and energy to guide candidates, while a multi-fidelity verifier mixes cheap checks with full device verification and reuses recorded failures. A multi-tenant runtime runs cloud training and device checks in parallel. Across 50 public tasks it beat the task-specific reference on 40 and found an SLO-feasible artifact on 45, and it was competitive on a self-collected SEN sensing dataset.
- LLM system turns high-level intent into deployable edge ML artifacts.
- Constraint-aware synthesis tree uses measured SLO gaps to guide search.
- Multi-fidelity verifier cuts full target-device checks while keeping reliability.
- Beat task-specific references on 40 of 50 public tasks; SLO-feasible on 45.
From the original article
1 min read at arxiv.orgMachine learning (ML) increasingly powers Internet of Things (IoT) applications at the edge. Yet producing a deployable edge ML artifact for a specific scenario requires navigating a huge search space spanning data representation, model design, training on domain-specific data, and runtime customization. This workflow is fragmented and difficult to scale across diverse edge applications.
By Genglin Wang · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.