Researchers have developed GoCoMA, a novel multimodal framework designed to attribute code authorship to specific large language models. This system combines code stylometry, which analyzes structural and stylistic signatures, with image representations of binary pre-executable artifacts that capture lower-level byte semantics. By projecting these different data modalities into a hyperbolic space and fusing them using a specialized attention mechanism, GoCoMA aims to more accurately identify the LLM responsible for generating code. Experiments on established benchmarks demonstrated GoCoMA's superior performance compared to existing unimodal and Euclidean multimodal approaches. AI
Summary written by gemini-2.5-flash-lite from 1 source. How we write summaries →
IMPACT Introduces a new method for identifying LLM-generated code, potentially impacting code security and intellectual property.
RANK_REASON This is a research paper detailing a new method for code attribution using multimodal representations.