CMSIS-NN
summary
Shows how to run neural nets on Cortex‑M microcontrollers using ARM tools and CMSIS‑NN to squeeze models to fit and run efficiently. Uses a keyword‑spotting example, compares architectures and finds a CRNN at about 80 KB model, 3.3M ops gives best accuracy/ops tradeoff vs a depth‑wise separable CNN with half the memory but nearly double ops. Demonstrates 8‑bit quantization retains accuracy and claims roughly 5× speed and energy improvement with CMSIS‑NN.
AI-written summary · about
chapters (5)
- 0:00Introduction to CMSIS-NN
- 3:13Why run ML at the edge
- 7:53Choosing neural network models
- 12:05Model compression and deployment
- 15:21Performance with CMSIS-NN
related talks
24mRunning AI/Neural networks on microcontrollers made simple with the STM32Cube.AI
26mMemory-Efficient LLM Inference on Edge Devices With NNTrainer
01h 57mIntroduction to Neuromorphic Computing
19mOptimizing CNN Model Inference on CPUs
38mNEURAL NETWORKS ON YOUR MOBILE PHONE. DIVING INTO THE NN API
21mThings I learned running neural networks on microcontrollers
01h 02mMulti-Dimension Mapping of Embedded Applications
19mWhy Machine Learning on Microcontrollers?