Lightning Talk: Orchestrating Machine Learning on Edge Devices...
summary
Claim: PyTorch models can run efficiently on heterogeneous edge devices by compiling inference to WebAssembly, scheduling them via a Wasm broker, and treating sensors/cameras as Kubernetes resources with Akri. Implementation details: PyTorch + WebAssembly, Akri as Kubernetes device discovery, a Rust discovery handler compiled to Wasm registers edge nodes, and mobile NeRF/instant NGP inference is packaged as Wasm modules. Result: near‑native performance with small binaries, fast startup, and almost real‑time rendering on constrained devices.
AI-written summary · about
chapters (7)
- 0:00WebAssembly and edge ML
- 1:25WebAssembly suits edge devices
- 2:31WebAssembly vs containers
- 3:13Arri for Kubernetes on edge
- 5:21WebAssembly with ML workflows
- 7:08Demo: ML on edge
- 11:01Q&A: Module linking
related talks
25mOrchestrating Machine Learning on Edge Devices Using Akri and WebAssembly
07mLightning Talk: AI at the Edge: ONNX Inference in WASM on Featherweight K0s
10mLightning Talk: Scaling AI With Wasm and Edge Computing
36mApplied AI on the edge
21mConnecting to devices at the edge with minimal footprint using AKS Edge Essentials and Akri on WASMs
33mWebAssembly based AI as a Service with Kubernetes
25mLeveraging PyTorch for Generative AI in Distributed Edge Clouds
29mLiving on the Edge: Using IoT Devices on Kubernetes WebAssembly Applications