Formal description of ML models for unambiguous implementation

TL;DR

This paper proposes an extension to the nnef language to enable unambiguous, traceable implementation of neural networks in safety-critical systems, demonstrated through CUDA implementation on Xavier hardware.

Contribution

It introduces a formal extension to nnef for precise model specification and demonstrates its practical implementation on Xavier hardware for safety-critical applications.

Findings

Extended nnef allows traceable distribution and parallelization.

Implementation on Xavier demonstrates feasibility.

Supports safety-critical system requirements.

Abstract

Implementing deep neural networks in safety critical systems, in particular in the aeronautical domain, will require to offer adequate specification paradigms to preserve the semantics of the trained model on the final hardware platform. We propose to extend the nnef language in order to allow traceable distribution and parallelisation optimizations of a trained model. We show how such a specification can be implemented in cuda on a Xavier platform.