Advances in Compiler Optimization and Smart Contract Security

The field of compiler optimization and smart contract security is rapidly evolving, with a strong focus on leveraging machine learning and large language models to improve performance and security. Researchers are exploring new approaches to compiler auto-tuning, including the use of reinforcement learning and large language models to optimize compiler pass sequences. In the area of smart contract security, novel frameworks are being developed to detect vulnerabilities and phishing attacks, often using large language models and machine learning techniques to analyze contract bytecode and identify malicious patterns. Additionally, new compilation pipelines are being proposed to improve the performance and security of smart contracts, including the use of WebAssembly and MLIR. Notable papers in this area include Compiler-R1, which introduces a reinforcement learning-driven framework for compiler auto-tuning, and Decompiling Smart Contracts with a Large Language Model, which presents a pioneering decompilation pipeline that leverages large language models to transform Ethereum Virtual Machine bytecode into human-readable Solidity code. Another notable work is PhishingHook, which applies machine learning techniques to detect phishing activities in smart contracts by analyzing the contract's bytecode and its constituent opcodes.

Sources

Compiler-R1: Towards Agentic Compiler Auto-tuning with Reinforcement Learning

ETrace:Event-Driven Vulnerability Detection in Smart Contracts via LLM-Based Trace Analysis

WAMI: Compilation to WebAssembly through MLIR without Losing Abstraction

Dataset of Yul Contracts to Support Solidity Compiler Research

PhishingHook: Catching Phishing Ethereum Smart Contracts leveraging EVM Opcodes

Decompiling Smart Contracts with a Large Language Model

JsDeObsBench: Measuring and Benchmarking LLMs for JavaScript Deobfuscation

Built with on top of