Biosignal processing for devices like ECG patches, glucose monitors, fitness bands

“TinyML for Medical Wearables” is an international workshop that explores the intersection of embedded AI, digital health, and biomedical signal processing. It focuses on deploying ML models on ultra-low-power microcontrollers used in wearables such as smartwatches, fitness trackers, ECG patches, glucose monitors, and biosensors.

Participants will learn to build lightweight models for vital sign detection, arrhythmia classification, motion analysis, and disease prediction, using tools such as TensorFlow Lite for Microcontrollers, Edge Impulse, Arduino Nano BLE, and real-world physiological datasets.

• Bridge embedded ML with health-focused wearable design

• Train professionals in real-world, deployable AI for patient monitoring

• Encourage data efficiency, privacy-first design, and energy conservation

• Explore emerging regulatory frameworks for AI-enabled medical devices

• Drive innovation in personalized, always-on digital health monitoring

Day 1: Model Compression Basics

• Why Model Compression Matters

  • Challenges of deploying deep learning models on edge devices

  • Trade-offs: Accuracy vs. efficiency

• Compression Techniques Overview

  • Pruning (structured/unstructured)

  • Quantization (8-bit, binary networks)

  • Knowledge distillation basics

• Hands-on Compression Lab

  • Applying pruning and quantization using TensorFlow Lite or PyTorch Mobile

  • Benchmarking latency and model size

Day 2: On-Device Arrhythmia Classification – Case Study

• Introduction to ECG & Arrhythmia Detection

  • ECG signal properties and common arrhythmia types

  • Importance of real-time edge inference in cardiology

• End-to-End Model Workflow

  • Data preprocessing and labeling (MIT-BIH dataset)

  • Model selection (CNN, LSTM for time-series classification)

• Deployment & Evaluation

  • Model deployment on microcontrollers (Arduino, EdgeTPU)

  • Performance metrics: accuracy, latency, power use

Day 3: Privacy-Preserving Federated Updates

• • Fundamentals of Federated Learning

    • Decentralized model training across edge devices

    • Comparison with traditional cloud-based learning

  • Ensuring Data Privacy

    • Differential privacy in federated updates

    • Homomorphic encryption & secure aggregation techniques

  • Use Case Simulation

    • Implementing a federated training round using PySyft or Flower

    • Simulating data from multiple hospitals for arrhythmia classification

• Biomedical and electronics engineers

• Embedded system developers and IoT architects

• ML engineers focused on healthcare applications

• Clinical AI researchers and health-tech innovators

• UG/PG/PhD students in bioinformatics, medical electronics, or AI/IoT