Zhen Yang

Fang Liu, Yang Liu, Lin Shi et al.

The rise of Large Language Models (LLMs) has significantly advanced many applications on software engineering tasks, particularly in code generation. Despite the promising performance, LLMs are prone to generate hallucinations, which means LLMs might produce outputs that deviate from users' intent, exhibit internal inconsistencies, or misalign with the factual knowledge, making the deployment of LLMs potentially risky in a wide range of applications. Existing work mainly focuses on investing the hallucination in the domain of natural language generation (NLG), leaving a gap in understanding the types and extent of hallucinations in the context of code generation. To bridge the gap, we conducted a thematic analysis of the LLM-generated code to summarize and categorize the hallucinations present in it. Our study established a comprehensive taxonomy of hallucinations in LLM-generated code, encompassing 5 primary categories of hallucinations depending on the conflicting objectives and varying degrees of deviation observed in code generation. Furthermore, we systematically analyzed the distribution of hallucinations, exploring variations among different LLMs and their correlation with code correctness. Based on the results, we proposed HalluCode, a benchmark for evaluating the performance of code LLMs in recognizing hallucinations. Hallucination recognition and mitigation experiments with HalluCode and HumanEval show existing LLMs face great challenges in recognizing hallucinations, particularly in identifying their types, and are hardly able to mitigate hallucinations. We believe our findings will shed light on future research about hallucination evaluation, detection, and mitigation, ultimately paving the way for building more effective and reliable code LLMs in the future.

Zhen Yang, Yansong Ma, Lei Chen

Traditional Evidence Deep Learning (EDL) methods rely on static hyperparameter for uncertainty calibration, limiting their adaptability in dynamic data distributions, which results in poor calibration and generalization in high-risk decision-making tasks. To address this limitation, we propose the Meta-Policy Controller (MPC), a dynamic meta-learning framework that adjusts the KL divergence coefficient and Dirichlet prior strengths for optimal uncertainty modeling. Specifically, MPC employs a bi-level optimization approach: in the inner loop, model parameters are updated through a dynamically configured loss function that adapts to the current training state; in the outer loop, a policy network optimizes the KL divergence coefficient and class-specific Dirichlet prior strengths based on multi-objective rewards balancing prediction accuracy and uncertainty quality. Unlike previous methods with fixed priors, our learnable Dirichlet prior enables flexible adaptation to class distributions and training dynamics. Extensive experimental results show that MPC significantly enhances the reliability and calibration of model predictions across various tasks, improving uncertainty calibration, prediction accuracy, and performance retention after confidence-based sample rejection.