Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Learn about a novel machine learning approach for automatically generating function names in stripped binary code through this 16-minute conference presentation from USENIX Security '25. Discover how BLens revolutionizes binary reverse engineering by adapting automated image captioning techniques to the domain of binary function analysis, enabling different parts of a binary function to be associated with corresponding parts of its name. Explore the innovative ensemble embedding methodology that combines multiple binary function representations and aligns them with name representation latent space using contrastive learning approaches. Understand how the transformer architecture has been specifically tailored for function name generation, addressing the critical challenge of generalizing to projects unrelated to training sets. Examine the significant performance improvements achieved by BLens, including F1 scores of 0.79 versus 0.70 in standard per-binary splitting scenarios, 0.46 versus 0.29 in cross-project settings emphasizing generalizability, and 0.32 versus 0.19 in experimental configurations with reduced shared components across projects. Gain insights into how this breakthrough approach outperforms existing state-of-the-art methods and provides valuable assistance to human reverse engineers working with stripped binaries.
