Artificial intelligence machine learning models used in aviation systems are growing with more use cases for edge and embedded use in real-time systems. Deployment for real-time systems edge use in aircraft and space systems requires rigorous performance and reliability testing of these models after initial architecture, training and validation is completed on scalable computing systems. Most often models must be quantized for embedded use, to reduce memory footprint and to reduce latency when used in avionics critical path solutions, and at the same time, the reliability of these models must be maintained.
To meet new demands for deployment of AI machine learning models in digital avionics, it is important to consider standards for deployment such as ONNX, reliability and performance testing, comparison to full-scale models to prevent reliability regression using standard measures such as precision, recall, receiver operator characteristics, F1 PR geometric mean, and latency for deployed models when integrated with edge sensor systems pipelines.
This tutorial will introduce you to the challenges, opportunities and the latest hardware and software tools and practices for machine learning model development with TensorFlow and PyTorch for convolutional neural networks and transformers and deployment of those models to embedded systems with acceleration by co-processing.
