Zelei Cheng

Chia-Hsuan Lee, Mingyang Zhou, Renkun Ni et al.

Preference optimization methods such as DPO and KTO are widely used for aligning language models, yet little is understood about what properties of preference data drive downstream reasoning gains. We ask: what aspects of a preference pair improve a reasoning model's performance on general reasoning tasks? We investigate two distinct notions of quality delta in preference data: generator-level delta, arising from the differences in capability between models that generate chosen and rejected reasoning traces, and sample-level delta, arising from differences in judged quality differences within an individual preference pair. To study generator-level delta, we vary the generator's scale and model family, and to study sample-level delta, we employ an LLM-as-a-judge to rate the quality of generated traces along multiple reasoning-quality dimensions. We find that increasing generator-level delta steadily improves performance on out-of-domain reasoning tasks and filtering data by sample-level delta can enable more data-efficient training. Our results suggest a twofold recipe for improving reasoning performance through preference optimization: maximize generator-level delta when constructing preference pairs and exploit sample-level delta to select the most informative training examples.

Xiaoguang Li, Hui Li, Haonan Yan et al.

Public intelligent services enabled by machine learning algorithms are vulnerable to model extraction attacks that can steal confidential information of the learning models through public queries. Differential privacy (DP) has been considered a promising technique to mitigate this attack. However, we find that the vulnerability persists when regression models are being protected by current DP solutions. We show that the adversary can launch a query-flooding parameter duplication (QPD) attack to infer the model information by repeated queries. To defend against the QPD attack on logistic and linear regression models, we propose a novel High-Dimensional Gaussian (HDG) mechanism to prevent unauthorized information disclosure without interrupting the intended services. In contrast to prior work, the proposed HDG mechanism will dynamically generate the privacy budget and random noise for different queries and their results to enhance the obfuscation. Besides, for the first time, HDG enables an optimal privacy budget allocation that automatically determines the minimum amount of noise to be added per user-desired privacy level on each dimension. We comprehensively evaluate the performance of HDG using real-world datasets and shows that HDG effectively mitigates the QPD attack while satisfying the privacy requirements. We also prepare to open-source the relevant codes to the community for further research.

Zelei Cheng, Xian Wu, Jiahao Yu et al.

Deep reinforcement learning (DRL) is playing an increasingly important role in real-world applications. However, obtaining an optimally performing DRL agent for complex tasks, especially with sparse rewards, remains a significant challenge. The training of a DRL agent can be often trapped in a bottleneck without further progress. In this paper, we propose RICE, an innovative refining scheme for reinforcement learning that incorporates explanation methods to break through the training bottlenecks. The high-level idea of RICE is to construct a new initial state distribution that combines both the default initial states and critical states identified through explanation methods, thereby encouraging the agent to explore from the mixed initial states. Through careful design, we can theoretically guarantee that our refining scheme has a tighter sub-optimality bound. We evaluate RICE in various popular RL environments and real-world applications. The results demonstrate that RICE significantly outperforms existing refining schemes in enhancing agent performance.

Zelei Cheng, Jiahao Yu, Xinyu Xing

Deep Reinforcement Learning (DRL) has achieved remarkable success in sequential decision-making tasks across diverse domains, yet its reliance on black-box neural architectures hinders interpretability, trust, and deployment in high-stakes applications. Explainable Deep Reinforcement Learning (XRL) addresses these challenges by enhancing transparency through feature-level, state-level, dataset-level, and model-level explanation techniques. This survey provides a comprehensive review of XRL methods, evaluates their qualitative and quantitative assessment frameworks, and explores their role in policy refinement, adversarial robustness, and security. Additionally, we examine the integration of reinforcement learning with Large Language Models (LLMs), particularly through Reinforcement Learning from Human Feedback (RLHF), which optimizes AI alignment with human preferences. We conclude by highlighting open research challenges and future directions to advance the development of interpretable, reliable, and accountable DRL systems.

Zelei Cheng, Xian Wu, Jiahao Yu et al.

Toxicity classification in textual content remains a significant problem. Data with labels from a single annotator fall short of capturing the diversity of human perspectives. Therefore, there is a growing need to incorporate crowdsourced annotations for training an effective toxicity classifier. Additionally, the standard approach to training a classifier using empirical risk minimization (ERM) may fail to address the potential shifts between the training set and testing set due to exploiting spurious correlations. This work introduces a novel bi-level optimization framework that integrates crowdsourced annotations with the soft-labeling technique and optimizes the soft-label weights by Group Distributionally Robust Optimization (GroupDRO) to enhance the robustness against out-of-distribution (OOD) risk. We theoretically prove the convergence of our bi-level optimization algorithm. Experimental results demonstrate that our approach outperforms existing baseline methods in terms of both average and worst-group accuracy, confirming its effectiveness in leveraging crowdsourced annotations to achieve more effective and robust toxicity classification.

Jiahao Yu, Zelei Cheng, Xian Wu et al.

Large language models (LLMs) are increasingly used in various domains, showing impressive potential on different tasks. Recently, reasoning LLMs have been proposed to improve the \textit{reasoning} or \textit{thinking} capabilities of LLMs to solve complex problems. Despite the promising results of reasoning LLMs, enhancing the multi-step reasoning capabilities of LLMs still remains a significant challenge. While existing optimization methods have advanced the LLM reasoning capabilities, they often treat reasoning trajectories as a whole, without considering the underlying critical steps within the trajectory. In this paper, we introduce \textbf{G}uided \textbf{P}ivotal \textbf{O}ptimization (GPO), a novel fine-tuning strategy that dives into the reasoning process to enable more effective improvements. GPO first identifies the `critical step' within a reasoning trajectory - a point that the model must carefully proceed to succeed at the problem. We locate the critical step by estimating the advantage function. GPO then resets the policy to the critical step, samples the new rollout and prioritizes the learning process on those rollouts. This focus allows the model to learn more effectively from pivotal moments within the reasoning process to improve the reasoning performance. We demonstrate that GPO is a general strategy that can be integrated with various optimization methods to improve reasoning performance. Besides theoretical analysis, our experiments across challenging reasoning benchmarks show that GPO can consistently and significantly enhance the performance of existing optimization methods, showcasing its effectiveness and generalizability in improving LLM reasoning by concentrating on pivotal moments within the generation process.