AI and Machine Learning BoF
summary
Conference participants map the stack for deep learning inference on ARM and identify two concrete targets: define a ONNX/NEF common model format so any framework (TensorFlow, Caffe, PyTorch, MXNet) can export once, and standardize the runtime/operator layer with shared math libraries (ARM Compute Library, CMSIS‑NN) and clear back ends. Demos reported include a Cortex‑M live object recognition prototype and a PCIe card with eight AI chips, both working, while most SoC accelerators still ship as proprietary user‑space blobs and require per‑vendor drivers, so CI and benchmarks are proposed to track regressions.
AI-written summary · about
chapters (12)
- 0:00Introduction and AI history
- 3:01Deep learning focus and accelerators
- 7:18Software frameworks and fragmentation
- 10:47Hardware architectures comparison
- 14:50Server-side AI challenges
- 20:27Common model format efforts
- 23:47Automotive AI stack mapping
- 27:07Open vs closed stack debate
- 31:48Fragmentation and standardization
- 36:03Deployment and benchmarking
- 38:57Community and framework focus
- 41:52Closing and Q&A
related talks
14mArm: One Platform for Agentic AI, from Cloud to the Edge
16mEdge AI needs to be for everyone: breaking barriers to adoption
16mAccelerating Web AI on
37mDeep Learning Neural Network Acceleration at the Edge
37mDeep Learning Neural Network Acceleration at the Edge
33mAI and SDR: Software Meets Hardware Again...
36mApplied AI on the edge
20mArchitecting AI: Efficient Models, Everywhere