arXiv subjects

Licheng Pan

Publications and source records attributed to Licheng Pan.

28 records · Page 2Linked to original sources

arXiv2026

Robust Reward Modeling for Large Language Models via Causal Decomposition

Reward models are central to aligning large language models, yet they often overfit to spurious cues such as response length and overly agreeable tone. Most prior work weakens these cues directly by penalizing or controlling specific artifacts, but it does not explicitly encourage the model to ground preferences in the prompt's intent. We learn a decoder that maps a candidate answer to the latent intent embedding of the input. The reconstruction error is used as a signal to regularize the reward model training. We provide theoretical evidence that this signal emphasizes prompt-dependent information while suppressing prompt-independent shortcuts. Across math, helpfulness, and safety benchmarks, the decoder selects shorter and less sycophantic candidates with 0.877 accuracy. Incorporating this signal into RM training in Gemma-2-2B-it and Gemma-2-9B-it increases RewardBench accuracy from 0.832 to 0.868. For Best-of-N selection, our framework increases length-controlled win rates while producing shorter outputs, and remains robust to lengthening and mild off-topic drift in controlled rewrite tests.

cs.CL↗

arXiv2026

CausalRM: Causal-Theoretic Reward Modeling for RLHF from Observational User Feedbacks

Despite the success of reinforcement learning from human feedback (RLHF) in aligning language models, current reward modeling heavily relies on experimental feedback data collected from human annotators under controlled and costly conditions. In this work, we introduce observational reward modeling -- learning reward models with observational user feedback (e.g., clicks, copies, and upvotes) -- as a scalable and cost-effective alternative. We identify two fundamental challenges in this setting: (1) observational feedback is noisy due to annotation errors, which deviates it from true user preference; (2) observational feedback is biased by user preference, where users preferentially provide feedback on responses they feel strongly about, which creats a distribution shift between training and inference data. To address these challenges, we propose CausalRM, a causal-theoretic reward modeling framework that aims to learn unbiased reward models from observational feedback. To tackle challenge (1), CausalRM introduces a noise-aware surrogate loss term that is provably equivalent to the primal loss under noise-free conditions by explicitly modeling the annotation error generation process. To tackle challenge (2), CausalRM uses propensity scores -- the probability of a user providing feedback for a given response -- to reweight training samples, yielding a loss function that eliminates user preference bias. Extensive experiments across diverse LLM backbones and benchmark datasets validate that CausalRM effectively learns accurate reward signals from noisy and biased observational feedback and delivers substantial performance improvements on downstream RLHF tasks -- including a 49.2% gain on WildGuardMix and a 32.7% improvement on HarmBench. Code is available on our project website.

cs.LG↗

arXiv2026

Analyzing and Improving Diffusion Models for Time-Series Data Imputation: A Proximal Recursion Perspective

Diffusion models (DMs) have shown promise for Time-Series Data Imputation (TSDI); however, their performance remains inconsistent in complex scenarios. We attribute this to two primary obstacles: (1) non-stationary temporal dynamics, which can bias the inference trajectory and lead to outlier-sensitive imputations; and (2) objective inconsistency, since imputation favors accurate pointwise recovery whereas DMs are inherently trained to generate diverse samples. To better understand these issues, we analyze DM-based TSDI process through a proximal-operator perspective and uncover that an implicit Wasserstein distance regularization inherent in the process hinders the model's ability to counteract non-stationarity and dissipative regularizer, thereby amplifying diversity at the expense of fidelity. Building on this insight, we propose a novel framework called SPIRIT (Semi-Proximal Transport Regularized time-series Imputation). Specifically, we introduce entropy-induced Bregman divergence to relax the mass preserving constraint in the Wasserstein distance, formulate the semi-proximal transport (SPT) discrepancy, and theoretically prove the robustness of SPT against non-stationarity. Subsequently, we remove the dissipative structure and derive the complete SPIRIT workflow, with SPT serving as the proximal operator. Extensive experiments demonstrate the effectiveness of the proposed SPIRIT approach.

cs.LG↗

arXiv2026

A Causal Perspective for Enhancing Jailbreak Attack and Defense

Uncovering the mechanisms behind "jailbreaks" in large language models (LLMs) is crucial for enhancing their safety and reliability, yet these mechanisms remain poorly understood. Existing studies predominantly analyze jailbreak prompts by probing latent representations, often overlooking the causal relationships between interpretable prompt features and jailbreak occurrences. In this work, we propose Causal Analyst, a framework that integrates LLMs into data-driven causal discovery to identify the direct causes of jailbreaks and leverage them for both attack and defense. We introduce a comprehensive dataset comprising 35k jailbreak attempts across seven LLMs, systematically constructed from 100 attack templates and 50 harmful queries, annotated with 37 meticulously designed human-readable prompt features. By jointly training LLM-based prompt encoding and GNN-based causal graph learning, we reconstruct causal pathways linking prompt features to jailbreak responses. Our analysis reveals that specific features, such as "Positive Character" and "Number of Task Steps", act as direct causal drivers of jailbreaks. We demonstrate the practical utility of these insights through two applications: (1) a Jailbreaking Enhancer that targets identified causal features to significantly boost attack success rates on public benchmarks, and (2) a Guardrail Advisor that utilizes the learned causal graph to extract true malicious intent from obfuscated queries. Extensive experiments, including baseline comparisons and causal structure validation, confirm the robustness of our causal analysis and its superiority over non-causal approaches. Our results suggest that analyzing jailbreak features from a causal perspective is an effective and interpretable approach for improving LLM reliability. Our code is available at https://github.com/Master-PLC/Causal-Analyst.

cs.LG↗

arXiv2026

DeepFilter: A Transformer-style Framework for Accurate and Efficient Process Monitoring

The process monitoring task is characterized by stringent demands for accuracy and efficiency. Current transformer-based methods, characterized by self-attention for temporal fusion, exhibit limitations in accurately understanding the semantic context and efficiently processing monitoring logs, rendering them inadequate for process monitoring. To address these limitations, we introduce DeepFilter, which revises the self-attention mechanism to improve both accuracy and efficiency. As a straightforward yet versatile approach, DeepFilter provides an instrumental baseline for practitioners in process monitoring, whether initiating new projects or enhancing existing capabilities.

cs.AI↗

arXiv2025

Understanding and Mitigating Overrefusal in LLMs from an Unveiling Perspective of Safety Decision Boundary

Large language models (LLMs) have demonstrated remarkable capabilities across a wide range of tasks, yet they often refuse to answer legitimate queries--a phenomenon known as overrefusal. Overrefusal typically stems from over-conservative safety alignment, causing models to treat many reasonable prompts as potentially risky. To systematically understand this issue, we probe and leverage the models' safety decision boundaries to analyze and mitigate overrefusal. Our findings reveal that overrefusal is closely tied to misalignment at these boundary regions, where models struggle to distinguish subtle differences between benign and harmful content. Building on these insights, we present RASS, an automated framework for prompt generation and selection that strategically targets overrefusal prompts near the safety boundary. By harnessing steering vectors in the representation space, RASS efficiently identifies and curates boundary-aligned prompts, enabling more effective and targeted mitigation of overrefusal. This approach not only provides a more precise and interpretable view of model safety decisions but also seamlessly extends to multilingual scenarios. We have explored the safety decision boundaries of various LLMs and construct the MORBench evaluation set to facilitate robust assessment of model safety and helpfulness across multiple languages. Code and datasets are available at https://github.com/Master-PLC/RASS.

cs.AI↗

arXiv2025

Relaxing Probabilistic Latent Variable Models' Specification via Infinite-Horizon Optimal Control

In this paper, we address the issue of model specification in probabilistic latent variable models (PLVMs) using an infinite-horizon optimal control approach. Traditional PLVMs rely on joint distributions to model complex data, but introducing latent variables results in an ill-posed parameter learning problem. To address this issue, regularization terms are typically introduced, leading to the development of the expectation-maximization (EM) algorithm, where the latent variable distribution is restricted to a predefined normalized distribution family to facilitate the expectation step. To overcome this limitation, we propose representing the latent variable distribution as a finite set of instances perturbed via an ordinary differential equation with a control policy. This approach ensures that the instances asymptotically converge to the true latent variable distribution as time approaches infinity. By doing so, we reformulate the distribution inference problem as an optimal control policy determination problem, relaxing the model specification to an infinite-horizon path space. Building on this formulation, we derive the corresponding optimal control policy using the Pontryagin's maximum principle and provide a closed-form expression for its implementation using the reproducing kernel Hilbert space. After that, we develop a novel, convergence-guaranteed EM algorithm for PLVMs based on this infinite-horizon-optimal-control-based inference strategy. Finally, extensive experiments are conducted to validate the effectiveness and superiority of the proposed approach.

eess.SY↗

arXiv2025

Mixture of Low Rank Adaptation with Partial Parameter Sharing for Time Series Forecasting

Multi-task forecasting has become the standard approach for time-series forecasting (TSF). However, we show that it suffers from an Expressiveness Bottleneck, where predictions at different time steps share the same representation, leading to unavoidable errors even with optimal representations. To address this issue, we propose a two-stage framework: first, pre-train a foundation model for one-step-ahead prediction; then, adapt it using step-specific LoRA modules.This design enables the foundation model to handle any number of forecast steps while avoiding the expressiveness bottleneck. We further introduce the Mixture-of-LoRA (MoLA) model, which employs adaptively weighted LoRA experts to achieve partial parameter sharing across steps. This approach enhances both efficiency and forecasting performance by exploiting interdependencies between forecast steps. Experiments show that MoLA significantly improves model expressiveness and outperforms state-of-the-art time-series forecasting methods. Code is available at https://anonymous.4open.science/r/MoLA-BC92.

cs.LG↗

arXiv2023

Modeling Task Relationships in Multi-variate Soft Sensor with Balanced Mixture-of-Experts

Accurate estimation of multiple quality variables is critical for building industrial soft sensor models, which have long been confronted with data efficiency and negative transfer issues. Methods sharing backbone parameters among tasks address the data efficiency issue; however, they still fail to mitigate the negative transfer problem. To address this issue, a balanced Mixture-of-Experts (BMoE) is proposed in this work, which consists of a multi-gate mixture of experts (MMoE) module and a task gradient balancing (TGB) module. The MoE module aims to portray task relationships, while the TGB module balances the gradients among tasks dynamically. Both of them cooperate to mitigate the negative transfer problem. Experiments on the typical sulfur recovery unit demonstrate that BMoE models task relationship and balances the training process effectively, and achieves better performance than baseline models significantly.

cs.LG↗

arXiv2023

TMoE-P: Towards the Pareto Optimum for Multivariate Soft Sensors

Multi-variate soft sensor seeks accurate estimation of multiple quality variables using measurable process variables, which have emerged as a key factor in improving the quality of industrial manufacturing. The current progress stays in some direct applications of multitask network architectures; however, there are two fundamental issues remain yet to be investigated with these approaches: (1) negative transfer, where sharing representations despite the difference of discriminate representations for different objectives degrades performance; (2) seesaw phenomenon, where the optimizer focuses on one dominant yet simple objective at the expense of others. In this study, we reformulate the multi-variate soft sensor to a multi-objective problem, to address both issues and advance state-of-the-art performance. To handle the negative transfer issue, we first propose an Objective-aware Mixture-of-Experts (OMoE) module, utilizing objective-specific and objective-shared experts for parameter sharing while maintaining the distinction between objectives. To address the seesaw phenomenon, we then propose a Pareto Objective Routing (POR) module, adjusting the weights of learning objectives dynamically to achieve the Pareto optimum, with solid theoretical supports. We further present a Task-aware Mixture-of-Experts framework for achieving the Pareto optimum (TMoE-P) in multi-variate soft sensor, which consists of a stacked OMoE module and a POR module. We illustrate the efficacy of TMoE-P with an open soft sensor benchmark, where TMoE-P effectively alleviates the negative transfer and seesaw issues and outperforms the baseline models.

cs.AI↗