LLMs tackle smart contract vulnerabilities with new continual learning framework

Researchers have developed a novel framework to address the challenges of using Large Language Models (LLMs) for smart contract vulnerability detection. The proposed three-stage pipeline tackles parameter-efficient adaptation, catastrophic forgetting, and adapter consolidation. The adaptation stage utilizes Frequency-Aware Low-Rank Adaptation (FA-LoRA) with per-frequency importance gates, requiring minimal trainable parameters. The continual learning stage employs Forget-Aware Replay (FAR) to prioritize vulnerable knowledge for rehearsal, achieving a Micro-F1 score of 0.8022. Finally, Anchor-Protected Progressive Merging (APPM) consolidates multiple adapters into a single model, maintaining high performance with minimal overhead. AI

IMPACT This framework could improve the efficiency and effectiveness of LLMs in identifying evolving security threats in smart contracts.

RANK_REASON Academic paper detailing a new framework and methods for LLM application. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Anchor-Protected Progressive Merging
• DIVE
• FA-LoRA
• Forget-Aware Replay
• Frequency-Aware Continual Learning for Smart Contract Vulnerability Detection with Large Language Models
• Frequency-Aware Low-Rank Adaptation
• Large Language Models

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →