Compiler fuzzing faces a fundamental dilemma: traditional rule-based generators easily fall into coverage bottlenecks due to rigid grammar constraints, while Large Language Models (LLMs), despite synthesizing semantically rich code, suffer from extremely low validity rates in their raw outputs, causing a severe waste of inference resources. To resolve this contradiction, this paper proposes RL-LLMfuzzer, a reinforcement learning and LLM-based differential fuzzing framework. The core innovation lies in its pioneering dual-model decoupled architecture: it utilizes a Multi-Armed Bandit (MAB) algorithm to drive a high-capacity prompt evolution model, dynamically optimizing high-level testing intents based on coverage feedback; subsequently, a local code generation model instantiates these intents. To overcome the inherent syntactic fragility and inference latency of LLMs, the framework integrates a lightweight heuristic repair module and a deterministic AST mutation engine, structurally amplifying valid seeds before differential evaluation. Extensive experiments on GCC and Clang demonstrate superior testing efficacy: verified across multiple independent 24-hour trials, RL-LLMfuzzer achieved a statistically significant 32.1% increase in average compiler backend line coverage in the C-language benchmark compared to the state-of-the-art LLM fuzzer Fuzz4All (p<0.001\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$p < 0.001$$\end{document}). The generated corpus significantly outperforms baseline tools like Csmith, YARPGen, and GrayC in multi-dimensional diversity. To date, RL-LLMfuzzer has successfully unearthed 28 officially confirmed vulnerabilities in GCC and LLVM/Clang, establishing a highly efficient and scalable paradigm for LLM-driven compiler validation.
Donghui Li, Yingying Li, Bo Zhao et al.· Journal of King Saud Univers...· 0 citations
CHORUS is presented, a post-training framework that pushes performance beyond what a conventional supervised fine-tuning (SFT)-to-reinforcement learning (RL) pipeline achieves, and consolidates the resulting specialists into a single 4B model.
Hejia Zhang, Sheng Lu, Zhongming Yu et al.· 0 citations
Fault Localization (FL) has evolved with the emergence of the Transformer architecture. Traditionally, FL relied on generating a ranking of suspiciousness scores for each code unit using information obtained during program execution from a set of test cases. However, Language Models trained with source code have enabled approaches that avoid the need for test suites, which are, sometimes, difficult or time-consuming to obtain. In this work, we present a refined training strategy tested on DeBERTa-v3-base, leveraging an encoder-based approach, to address two key challenges in FL: (1) the accurate classification of lines of code according to their suspiciousness scores, and (2) the class imbalance inherent in this type of problem. To this end, we extract the last hidden state corresponding to the embeddings of the new line tokens to generate a suspiciousness score for each line. This score then classifies each line as defective or not. The evaluation of our approach on the ConDefects dataset, repeated over ten random seeds, shows promising results compared to existing approaches, with ROC AUC gains up to 0.27 in cross-dataset comparisons. Thereby, we offer a competitive baseline for future test-free fault localization research while using language models that are approximately 200 times smaller than those used in prior work.
Jesus-Angel del-Hoyo-Gabaldon, Antonio Garcia-Cabot, Eva García-López et al.· Applied Sciences· 0 citations
Detecting machine-generated code is crucial for maintaining software security, quality and academic integrity. Traditional approaches often rely on stylistic or statistical features, which are increasingly circumvented by advanced code generation models. This paper introduces a novel approach leveraging Graph Neural Networks (GNNs) to capture the structural characteristics of code, specifically modeling source snippets as Abstract Syntax Trees (ASTs). To enhance semantic comprehension, we integrate pre-trained CodeBERT embeddings into the architecture, creating a hybrid model that incorporates structural and semantic information. We evaluate our approach on SemEval-2026 Task 13, covering binary detection, multi-class authorship attribution, and hybrid code classification. Experimental results demonstrate that our GNN-based structural analysis outperforms standalone stylistic and transformer-based base-lines across all subtasks, particularly in multi-class and adversarial settings. This work highlights the potential of GNNs for a more structural understanding of code authorship.
LLMs have demonstrated strong capabilities in code generation and automated program repair, but migrating an entire repository rarely produces a runnable application because long-horizon translation challenges LLM-based agents'ability to maintain repository-level migration objectives. In this work, we propose Entropy-based Code Adversarial Translation (ECAT), a multi-agent framework for automated Android-to-HarmonyOS repository migration. ECAT formulates repository migration as adversarial entropy minimization through a generator-discriminator architecture. The discriminator measures migration quality using a unified metric called Code Entropy and produces text gradients that specify both file-level generation directives and the skills needed to execute them. Guided by these optimization signals, the generator iteratively updates the repository, and each update is accepted only if it reduces Code Entropy. Repeated generator--discriminator interactions progressively drive the migration from an initial template toward a functionally complete HarmonyOS repository. Successful low-entropy trajectories are further distilled into a self-evolving memory tree, enabling transferable migration knowledge across repositories. We also introduce A2H-RepoBench, the first real-world benchmark for Android-to-HarmonyOS repository migration, covering applications from tens of thousands to hundreds of thousands of lines of code. Evaluated by node alignment and an agent-based functional judge, ECAT achieves 74.7% overall migration quality and consistently outperforms existing agent-based methods across repositories of different scales.
Yushun Tang, Yisen Cao, Zhicheng Chen et al.· 0 citations
Large language models are widely used for code generation, but they can also produce insecure programs due to patterns learned from their pretraining data. Decoding-time steering has become an important solution to this problem: a small expert model is combined with the target model at each step to generate more secure code, which is referred to as co-decoding. However, the acceptance rule for existing co-decoding approaches does not consider the expert model's confidence. When the security expert is unconfident due to unseen patterns or out-of-distribution (OOD) contexts, its guidance can therefore be misleading. To address the challenge, we propose CoGate, a confidence-gated co-decoding approach that controls the expert's influence on the co-decoding process based on its confidence. We implement our approach and evaluate it across multiple LLM backends (CodeGen, DeepSeek-Coder, Qwen-Coder, StarCoder) on several code generation benchmarks (HumanEval, security suite, and CWEval). Our approach outperforms existing co-decoding methods (CoSec+) across multiple benchmarks, achieving up to a 12.6% gain of Func-Sec@10 on CWEval.
Minghao Hu, Lannan Luo, Allen G. Roush et al.· 0 citations