Thank you for sending your enquiry! One of our team members will contact you shortly.
Thank you for sending your booking! One of our team members will contact you shortly.
Duration 21 hours
Course Outline
Introduction to TinyML and Embedded AI for government applications
- Key attributes of TinyML model integration
- Operational constraints within microcontroller environments
- Summary of embedded AI software stacks
Foundations of Model Optimization
- Analysis of computational constraints
- Identification of operations with high memory demand
- Initial performance assessment and profiling
Quantization Methodologies
- Strategies for post-training quantization
- Quantization-aware training processes
- Assessment of precision versus resource utilization trade-offs
Pruning and Data Compression
- Structured and unstructured pruning techniques
- Utilization of weight sharing and model sparsity
- Compression algorithms designed for lightweight inference
Hardware-Specific Optimization
- Model deployment on ARM Cortex-M architectures
- Optimization for DSP and accelerator extensions
- Considerations for memory mapping and data flow
Benchmarking and Verification
- Evaluation of latency and throughput
- Measurement of power and energy usage
- Testing for accuracy and system robustness
Deployment Processes and Tooling
- Implementation using TensorFlow Lite Micro for embedded systems
- Integration of TinyML models with Edge Impulse workflows
- Testing and debugging procedures on physical hardware
Advanced Optimization Tactics
- Neural architecture search tailored for TinyML
- Combined quantization and pruning methodologies
- Model distillation for embedded inference tasks
Conclusion and Recommended Next Steps
Requirements
- Proficiency in machine learning workflows
- Experience with embedded systems or microcontroller-based development
- Familiarity with Python programming
Target Audience
- AI researchers
- Embedded ML engineers
- Professionals involved in resource-constrained inference systems
