Edge AI & TinyML Application Development

Train, optimize, quantize, and deploy deep learning models directly onto resource-constrained microcontrollers.

5 Weeks / 40 Hours Hands-on Labs

Certificate of TinyML Specialty Awarded

Training Covering Area

Master the end-to-end pipeline of machine learning on the edge. Gather sensor data, build models in TensorFlow/PyTorch, quantize to INT8, and run inference locally on ARM Cortex-M or ESP32 cores.

Module 1: Machine Learning & DSP at the Edge

TinyML overview: Constraints of microcontrollers (kilobytes of RAM & Flash)
Time-series sensor preprocessing: Digital filtering, windowing, and overlap
Audio processing: FFT (Fast Fourier Transform), spectrograms, and MFCC feature extraction
Setting up the Python ML pipeline: Scikit-learn, PyTorch, and TensorFlow

Module 2: Model Training & Optimization

Designing lightweight Neural Network architectures: CNNs, DNNs, and Autoencoders
Model pruning: Eliminating redundant weights to optimize memory footprints
Quantization foundations: Float32 representation vs INT8 fixed-point scaling
Quantization-Aware Training (QAT) vs Post-Training Quantization (PTQ)

Module 3: Deploying to Microcontrollers

TFLite converter tools, FlatBuffers, and exporting C++ byte arrays
TFLM runtime: Allocating tensor arenas, registering operators, and configuring interpreters
Optimizing math kernels using CMSIS-DSP and CMSIS-NN libraries
Profiling RAM utilization, latency cycles, and power consumption of inference loops

Module 4: Practical Edge AI Deployments

Project 1: Real-time keyword spotting (KWS) using microphone sensors
Project 2: Machine vibration predictive maintenance anomaly detection
Project 3: Gesture recognition and motion analytics from IMU data
Deploying and updating models in the field without updating full firmware images