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
- Key characteristics of deploying TinyML models
- Limitations inherent to microcontroller environments
- Survey of embedded AI toolchains
Foundations of Model Optimization
- Comprehending computational bottlenecks
- Recognizing operations with high memory demands
- Establishing baseline performance profiles
Quantization Methodologies
- Post-training quantization approaches
- Quantization-aware training techniques
- Analyzing the trade-off between accuracy and resource usage
Pruning and Compression
- Structured versus unstructured pruning methods
- Weight sharing and introducing model sparsity
- Algorithms for compressing models for lightweight inference
Hardware-Specific Optimization
- Deploying models on ARM Cortex-M architectures
- Leveraging DSP and accelerator extensions for optimization
- Considerations for memory mapping and dataflow
Benchmarking and Validation
- Analysis of latency and throughput
- Measurement of power and energy consumption
- Testing for accuracy and system robustness
Deployment Workflows and Tooling
- Utilizing TensorFlow Lite Micro for embedded implementation
- Integrating TinyML models into Edge Impulse pipelines
- Testing and debugging processes on physical hardware
Advanced Optimization Tactics
- Applying Neural Architecture Search to TinyML
- Combining quantization and pruning strategies
- Model distillation techniques for embedded inference
Conclusion and Future Directions
Requirements
- A solid grasp of machine learning workflows
- Hands-on experience with embedded systems or microcontroller development
- Proficiency in Python programming
Target Audience
- AI researchers
- Embedded ML engineers
- Professionals developing inference systems under resource constraints
Testimonials (1)
The exercises