A Deep Learning framework for Interoperable Machine Learning

Abstract

In this paper, we introduce an opinionated, extensible, Python framework that transpiles variety of classical Statistical and Machine Learning models onto Open Neural Network Exchange (ONNX) format via an underlying Deep Learning model. We achieve this by exploiting the compositionality of Deep Learning technology. By appropriately choosing the features, architecture, loss functions, and regularizers, among others, the fidelity between the source model and the target model can be specified. Depending on the model being transpiled, the fidelity can be exact or approximate. We present the design details, APIs of the framework, reference implementations, road map for development, and guidelines for contributions. A reference implementation is available for the popular scikit-learn APIs.