God Header Files, large header files included by numerous other code files, present significant challenges for code comprehension and maintenance. Existing approaches leverage various code similarity metrics to decompose them, but these metrics do not always capture the code’s functional essence accurately. Large Language Models (LLMs), with their advanced capabilities in code understanding and generation, offer a promising alternative for producing more effective refactorings. However, LLMs face three critical limitations that hinder practical application: they struggle with lengthy header files due to token constraints, suffer from hallucination by generating incomplete or spurious results, and produce cyclic dependencies that violate architectural principles and cause compilation failures. To address these challenges, we propose HFDecomposer, a hybrid approach that enhances LLMs with staged grouping and dehallucination techniques to effectively decompose header files. Our approach introduces a two-stage grouping framework for lengthy header files: it first groups strongly related code entities using traditional similarity metrics, then feeds group summaries to the LLM for higher-level semantic aggregation. To mitigate LLM hallucinations, we enhance prompts with factual knowledge extracted from static analysis, detect errors in LLM output, and make necessary corrections by reassigning missing entities and resolving cyclic dependencies. Our evaluation on real-world header file decomposition refactorings demonstrates that our method effectively overcomes the limitations of purely LLM-based techniques and outperforms the traditional state-of-the-art approach by 11%, delivering more accurate and reliable decomposition results. Our approach enables LLMs to handle lengthy header files efficiently, significantly reduces hallucinations, and ensures the reliability and practicality of the final decomposition.

Reinforcement learning (RL) has shown promise in enhancing the general Chain-of-Thought (CoT) reasoning capabilities of multimodal large language models (MLLMs). However, when applied to improve general CoT reasoning, existing RL frameworks often struggle to generalize beyond the training distribution. To address this, we propose NoisyGRPO, a systematic multimodal RL framework that introduces controllable noise into visual inputs for enhanced exploration and explicitly models the advantage estimation process via a Bayesian framework. Specifically, NoisyGRPO improves RL training by: (1) \textbf{Noise-Injected Exploration Policy}: Perturbing visual inputs with Gaussian noise to encourage exploration across a wider range of visual scenarios; and (2) \textbf{Bayesian Advantage Estimation}: Formulating advantage estimation as a principled Bayesian inference problem, where the injected noise level serves as a prior and the observed trajectory reward as the likelihood. This Bayesian modeling fuses both sources of information to compute a robust posterior estimate of trajectory advantage, effectively guiding MLLMs to prefer visually grounded trajectories over noisy ones. Experiments on standard CoT quality, general capability, and hallucination benchmarks demonstrate that NoisyGRPO substantially improves generalization and robustness, especially in RL settings with small-scale MLLMs such as Qwen2.5-VL 3B.

Cross-modal hashing has gained considerable attention in cross-modal retrieval due to its low storage cost and prominent computational efficiency. However, preserving more semantic information in the compact hash codes to bridge the modality gap still remains challenging. Most existing methods unconsciously neglect the influence of modality-private information on semantic embedding discrimination, leading to unsatisfactory retrieval performance. In this paper, we propose a novel deep cross-modal hashing method, called Semantic Disentanglement Adversarial Hashing (SDAH), to tackle these challenges for cross-modal retrieval. Specifically, SDAH is designed to decouple the original features of each modality into modality-common features with semantic information and modality-private features with disturbing information. After the preliminary decoupling, the modality-private features are shuffled and treated as positive interactions to enhance the learning of modality-common features, which can significantly boost the discriminative and robustness of semantic embeddings. Moreover, the variational information bottleneck is introduced in the hash feature learning process, which can avoid the loss of a large amount of semantic information caused by the high-dimensional feature compression. Finally, the discriminative and compact hash codes can be computed directly from the hash features. A large number of comparative and ablation experiments show that SDAH achieves superior performance than other state-ofthe- art methods.