Dear all, I am a part of a research group at the University Of Auckland. Our bigger vision is to create a set of tools for safety critical application of DNNs. This includes a formal approach for composition of multiple NNs, static timing analysis, and correct by construction approaches for SW/HW co-design.

Our first project is now live at https://www.aitransformer.com/ which is basically an enhanced Keras2C utility. Unlike the alternatives, the generated code doesn't depend on anything like TensorFlow C Library. This makes the code amenable to static timing analysis and perhaps more light-weight. If you use C code for AI application, or use AI in embedded systems and edge compute, we would love to hear from you. Leave comments. Thanks!