Data science in a C programming environment for AI-assisted programming in embedded systems

Abstract

This paper explores the feasibility of using a Large Language Model (LLM) to automate the translation of PyTorch models into deployable C code for embedded systems. With the increasing demand for machine learning on resource-constrained devices, manual implementation of neural networks in low-level languages such as C remains a significant bottleneck. In this paper, we evaluate the ability of LLMs to generate numerically consistent, functionally correct C implementations of various deep learning architectures, ranging from simple fully connected networks to convolutional neural networks (CNN).