Leveraging Machine Learning for Classifying Assembly Language Snippets to Aid Programmers in Understanding Assembly Code

Abstract - In this study, we propose a machine learning-based approach to help programmers learn and understand assembly language by classifying assembly code snippets. Assembly language, while powerful, is often difficult to interpret due to its low-level nature and the lack of high-level context. Our model trains on assembly language snippets labeled with common programming patterns to classify unseen assembly code. By doing so, we offer a tool that can assist in reverse engineering, optimization, and the understanding of compiled programs. We generate a dataset by pairing C code snippets with corresponding assembly instructions, then train a recurrent neural network to classify these snippets. Our results demonstrate the feasibility of this approach, with the model showing strong performance in classifying assembly code types. We also discuss future directions, including generating pseudocode from assembly. Key Words: Machine learning, compilers, pseudocode, assembly language, deep learning, RNN, LSTM