The talk opens by providing a sketch of machine learning and how it relates to functional programming. We subsequently introduce ONNX and ONNX-Scala, looking at design goals and features, discussing backends, the generated fine-grained API, and model code generation. Along the way we compare an ONNX model visualization with the representation of the model in ONNX-Scala. This is followed by a brief discussion of the type-safe/functional features offered by ONNX-Scala and the Dotty features they are built on. Finally, we close on some remarks on future directions, particularly the use of Dotty metaprogramming and other new features.

Topics: Machine Learning, Interoperability