Designing Neural Networks for Real-Time Systems

TL;DR

This paper presents a pipeline to compile neural networks trained in Keras into C code suitable for static timing analysis, enabling real-time guarantees in safety-critical cyber-physical systems.

Contribution

It introduces a method to convert neural networks into analyzable C code for static timing analysis, improving safety guarantees in real-time systems.

Findings

Successfully derived worst-case execution timings for an autonomous vehicle neural network

Compiled neural networks to a time-predictable platform for static analysis

Enhanced safety assurance in cyber-physical systems through formal timing guarantees

Abstract

Artificial Neural Networks (ANNs) are increasingly being used within safety-critical Cyber-Physical Systems (CPSs). They are often co-located with traditional embedded software, and may perform advisory or control-based roles. It is important to validate both the timing and functional correctness of these systems. However, most approaches in the literature consider guaranteeing only the functionality of ANN based controllers. This issue stems largely from the implementation strategies used within common neural network frameworks -- their underlying source code is often simply unsuitable for formal techniques such as static timing analysis. As a result, developers of safety-critical CPS must rely on informal techniques such as measurement based approaches to prove correctness, techniques that provide weak guarantees at best. In this work we address this challenge. We propose a design pipeline whereby neural networks trained using the popular deep learning framework Keras are compiled to functionally equivalent C code. This C code is restricted to simple constructs that may be analysed by existing static timing analysis tools. As a result, if compiled to a suitable time-predictable platform all execution bounds may be statically derived. To demonstrate the benefits of our approach we execute an ANN trained to drive an autonomous vehicle around a race track. We compile the ANN to the Patmos time-predictable controller, and show that we can derive worst case execution timings.