PhD Simon Friedberger | NXP Semiconductors | Belgium
Matthias Gohlke | Toradex AG | Germany
These are the topics that will be covered in this half-day seminar covering both theoretical and hands-on explanations:
- Theoretical aspects of deploying machine learning at the edge, focused on using a variety of open source frameworks, libraries and tools.
- How to target a model for CPU, GPU, DSP, or dedicated ML accelerator.
- Transfer learning - how a pre-trained model can be adapted to a new classification application
- Demonstrate an application doing simultaneous face and voice recognition
- How to compress a model for smaller size and higher performance
- Quantization techniques including pre- and post-training quantization.
- Overview of a vision pipeline - signal input, algorithm processing, inference/detection, encoding and streaming.
- -Processing requirements at each stage of the vision pipeline
- -Understanding and optimizing inference performance on Arm cores by using a quantized version of a state-of-the-art model
- System-level application (e.g. abandoned package detection) to illustrate a tangible application use case and associated challenges.
- Step-by-Step AWS walkthrough from training in the cloud to the optimized edge inference
- Technology demonstration integrated in an Industry 4.0 use case including industrial-grade MIPI CSI Camera
- Security Hardening for Machine Learning Systems
- Introduce model inversion, model cloning and adversarial examples
- Effectiveness of attacks and the state-of-the-art in security for machine learning