Target-Independent Code Generation for Embedded AI Deployment
Overview
Embedded AI applications are increasingly deployed across a wide range of hardware targets, from desktop CPUs and GPUs to microprocessors and edge devices. Achieving strong on-target performance often requires integrating vendor-specific deep learning libraries, but these libraries can add setup complexity, hardware dependencies, and version compatibility challenges.
In this webinar, we will demonstrate how to generate target-independent C, C++, and CUDA code from deep learning networks using MATLAB. Target-independent code generation enables engineers to deploy AI models that do not depend on 3rd party deep learning libraries, helping simplify deployment while maintaining a workflow for simulation, validation, profiling, and benchmarking.
We will use an industrial visual inspection example to show the end-to-end workflow: preparing image data, fine-tuning a pretrained SqueezeNet model, validating inference on the host, running software-in-the-loop simulation, generating CUDA code, and deploying to an NVIDIA Jetson AGX Orin. Along the way, we will compare the target-independent generated CUDA code with cuDNN and TensorRT workflows to examine deployment tradeoffs in code generation time, dynamic memory usage, accuracy, and on-target latency.
Highlights
- Generate target-independent code for complete applications, including deep learning networks along with pre- and postprocessing code, for NVIDIA Jetson boards.
- Configure for target-independent deep learning code generation.
- Validate generated code on host and target using software-in-the-loop simulation.
- Analyze memory usage, inference timing, and performance bottlenecks using postprocessing script and profiling tools.
Please allow approximately 45 minutes to attend the presentation and Q&A session. We will be recording this webinar, so if you can't make it for the live broadcast, register and we will send you a link to watch it on-demand.
About the Presenter
Wenzhe Li is a Product Manager for Embedded AI at MathWorks. He works with customers and internal teams to shape product strategy, prioritize features based on customer insights, and enable efficient deployment of AI models to embedded systems. Wenzhe holds a Ph.D. in Mechanical Engineering from the University of Cincinnati and brings research and industry experience in machine learning, predictive maintenance, and advanced process control.
