Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
This survey systematically maps label-efficient code vulnerability detection (CVD) approaches, categorizing them into five paradigm families based on their underlying mechanisms. The work connects these mechanisms to different code representations (token, graph, hybrid, knowledge-based) and highlights critical evaluation challenges, including label budget specification and leakage. The resulting Design Map and Decision Guide offer practical insights into trade-offs and failure modes for method selection in CVD when labeled data is scarce.
Navigating the CVD landscape with limited labels just got easier: a new Design Map and Decision Guide reveal the hidden trade-offs and failure modes of different approaches.
Machine-learning-based code vulnerability detection (CVD) has progressed rapidly, from deep program representations to pretrained code models and LLM-centered pipelines. Yet dependable vulnerability labeling remains expensive, noisy, and uneven across projects, languages, and CWE types, motivating approaches that reduce reliance on human labeling. This survey maps these approaches, synthesizing five paradigm families and the mechanisms they use. It connects mechanisms to token, graph, hybrid, and knowledgebased representations, and consolidates evaluation and reporting axes that limit comparison (label-budget specification, compute/cost assumptions, leakage, and granularity mismatches). A Design Map and constraintfirst Decision Guide distill trade-offs and failure modes for practical method selection.