Deep Learning Neural Network Acceleration at the Edge
summary
Edge inference is practical today but fragmented: frameworks spike in 2015 and grow huge — TensorFlow: ~32k commits, 1.6M LOC, ~456 years equivalent — and each uses different model formats so converters are mandatory (protobuf → flatbuffer for TF Lite). Hardware ranges from Google Edge TPU (preview ASIC) to Arm Project Trillium NPUs and GPUs delivering full‑HD 60fps object detection; vendors ship custom runtimes. Linaro launches a machine‑intelligence push aiming for ONNX + Arm NN plugin runtime with CI checks for performance and accuracy.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 0:54Deep learning at the edge
- 4:35The edge ecosystem
- 6:25Machine learning frameworks
- 13:38Framework comparisons
- 19:49Hardware acceleration
- 27:01Addressing fragmentation
- 30:53Linaro's initiative
- 33:54Closing and outlook
related talks
55mUsing TensorFlow for Performing Deep Learning at the Edge
10mThe Rise and Impact of Edge AI
36mApplied AI on the edge
48mThe Power of Edge AI: Revolutionizing Future of Real-Time, Autonomous Processing
55mEdgeAI - AI in the Palm of Your Hand
28mML on the Edge for Industry 4.0 with Arm Ethos N-78 Neural Processing Unit
21mUnlocking the potential of low-power embedded devices with Edge AI software
33mHigh Performance Deep Learning on Edge Devices With