arXiv subjects

Zhen Wu

Publications and source records attributed to Zhen Wu.

At least 55 records · Page 3Linked to original sources

arXiv2025

Persona-Aware Alignment Framework for Personalized Dialogue Generation

Personalized dialogue generation aims to leverage persona profiles and dialogue history to generate persona-relevant and consistent responses. Mainstream models typically rely on token-level language model training with persona dialogue data, such as Next Token Prediction, to implicitly achieve personalization, making these methods tend to neglect the given personas and generate generic responses. To address this issue, we propose a novel Persona-Aware Alignment Framework (PAL), which directly treats persona alignment as the training objective of dialogue generation. Specifically, PAL employs a two-stage training method including Persona-aware Learning and Persona Alignment, equipped with an easy-to-use inference strategy Select then Generate, to improve persona sensitivity and generate more persona-relevant responses at the semantics level. Through extensive experiments, we demonstrate that our framework outperforms many state-of-the-art personalized dialogue methods and large language models.

cs.CL↗

arXiv2025

Investigating Prosocial Behavior Theory in LLM Agents under Policy-Induced Inequities

As large language models (LLMs) increasingly operate as autonomous agents in social contexts, evaluating their capacity for prosocial behavior is both theoretically and practically critical. However, existing research has primarily relied on static, economically framed paradigms, lacking models that capture the dynamic evolution of prosociality and its sensitivity to structural inequities. To address these gaps, we introduce ProSim, a simulation framework for modeling the prosocial behavior in LLM agents across diverse social conditions. We conduct three progressive studies to assess prosocial alignment. First, we demonstrate that LLM agents can exhibit human-like prosocial behavior across a broad range of real-world scenarios and adapt to normative policy interventions. Second, we find that agents engage in fairness-based third-party punishment and respond systematically to variations in inequity magnitude and enforcement cost. Third, we show that policy-induced inequities suppress prosocial behavior, propagate norm erosion through social networks. These findings advance prosocial behavior theory by elucidating how institutional dynamics shape the emergence, decay, and diffusion of prosocial norms in agent-driven societies.

cs.SI↗

arXiv2025

PICNIC: Silicon Photonic Interconnected Chiplets with Computational Network and In-memory Computing for LLM Inference Acceleration

This paper presents a 3D-stacked chiplets based large language model (LLM) inference accelerator, consisting of non-volatile in-memory-computing processing elements (PEs) and Inter-PE Computational Network (IPCN), interconnected via silicon photonic to effectively address the communication bottlenecks. A LLM mapping scheme was developed to optimize hardware scheduling and workload mapping. Simulation results show it achieves $3.95\times$ speedup and $30\times$ efficiency improvement over the Nvidia A100 before chiplet clustering and power gating scheme (CCPG). Additionally, the system achieves further scalability and efficiency improvement with the implementation of CCPG to accommodate larger models, attaining $57\times$ efficiency improvement over Nvidia H100 at similar throughput.

cs.AR↗

arXiv2025

Outraged AI: Large language models prioritise emotion over cost in fairness enforcement

Emotions guide human decisions, but whether large language models (LLMs) use emotion similarly remains unknown. We tested this using altruistic third-party punishment, where an observer incurs a personal cost to enforce fairness, a hallmark of human morality and often driven by negative emotion. In a large-scale comparison of 4,068 LLM agents with 1,159 adults across 796,100 decisions, LLMs used emotion to guide punishment, sometimes even more strongly than humans did: Unfairness elicited stronger negative emotion that led to more punishment; punishing unfairness produced more positive emotion than accepting; and critically, prompting self-reports of emotion causally increased punishment. However, mechanisms diverged: LLMs prioritized emotion over cost, enforcing norms in an almost all-or-none manner with reduced cost sensitivity, whereas humans balanced fairness and cost. Notably, reasoning models (o3-mini, DeepSeek-R1) were more cost-sensitive and closer to human behavior than foundation models (GPT-3.5, DeepSeek-V3), yet remained heavily emotion-driven. These findings provide the first causal evidence of emotion-guided moral decisions in LLMs and reveal deficits in cost calibration and nuanced fairness judgements, reminiscent of early-stage human responses. We propose that LLMs progress along a trajectory paralleling human development; future models should integrate emotion with context-sensitive reasoning to achieve human-like emotional intelligence.

cs.CL↗

arXiv2025

Extreme Capacities in Generalized Direct Sum Channels

Quantum channel capacities play a central role in quantum Shannon theory, a formalism built upon rigorous coding theorems for noisy channels. Evaluating exact capacity values for general quantum channels remains intractable due to superadditivity. As a step toward understanding this phenomenon, we construct the generalized direct sum (GDS) channel, extending conventional direct sum channels through a direct sum structure in their Kraus operators. This construction forms the basis of the GDS framework, encompassing classes of channels with single-letter formula for quantum capacities and others exhibiting striking capacity features. The quantum capacity can approach zero yet display unbounded superadditivity combined with erasure channels. Private and classical capacities coincide and can become arbitrarily large, resulting in an unbounded gap with the quantum capacity. Providing a simpler and more intuitive approach, the framework deepens our understanding of quantum channel capacities.

quant-ph↗

arXiv2025

On Code-Induced Reasoning in LLMs

Code data has been shown to enhance the reasoning capabilities of large language models (LLMs), but it remains unclear which aspects of code are most responsible. We investigate this question with a systematic, data-centric framework. We construct parallel instruction datasets in ten programming languages and apply controlled perturbations that selectively disrupt structural or semantic properties of code. We then finetune LLMs from five model families and eight scales on each variant and evaluate their performance on natural language, math, and code tasks. Across 3,331 experiments, our results show that LLMs are more vulnerable to structural perturbations than semantic ones, particularly on math and code tasks. Appropriate abstractions like pseudocode and flowcharts can be as effective as code, while encoding the same information with fewer tokens without adhering to original syntax can often retain or even improve performance. Remarkably, even corrupted code with misleading signals remains competitive when surface-level regularities persist. Finally, syntactic styles also shape task-specific gains with Python favoring natural language reasoning and lower-level languages such as Java and Rust favoring math. Through our systematic framework, we aim to provide insight into how different properties of code influence reasoning and inform the design of training data for enhancing LLM reasoning capabilities.

cs.CL↗

arXiv2025

TrustJudge: Inconsistencies of LLM-as-a-Judge and How to Alleviate Them

The adoption of Large Language Models (LLMs) as automated evaluators (LLM-as-a-judge) has revealed critical inconsistencies in current evaluation frameworks. We identify two fundamental types of inconsistencies: (1) Score-Comparison Inconsistency, where lower-rated responses outperform higher-scored ones in pairwise comparisons, and (2) Pairwise Transitivity Inconsistency, manifested through circular preference chains (A>B>C>A) and equivalence contradictions (A=B=C\neq A). We argue that these issues come from information loss in discrete rating systems and ambiguous tie judgments during pairwise evaluation. We propose TrustJudge, a probabilistic framework that addresses these limitations through two key innovations: 1) distribution-sensitive scoring that computes continuous expectations from discrete rating probabilities, preserving information entropy for more precise scoring, and 2) likelihood-aware aggregation that resolves transitivity violations using bidirectional preference probabilities or perplexity. We also formalize the theoretical limitations of current LLM-as-a-judge frameworks and demonstrate how TrustJudge's components overcome them. When evaluated with Llama-3.1-70B-Instruct as judge using our dataset, TrustJudge reduces Score-Comparison inconsistency by 8.43% (from 23.32% to 14.89%) and Pairwise Transitivity inconsistency by 10.82% (from 15.22% to 4.40%), while maintaining higher evaluation accuracy. Our work provides the first systematic analysis of evaluation framework inconsistencies in LLM-as-a-judge paradigms, offering both theoretical insights and practical solutions for reliable automated assessment. The framework demonstrates consistent improvements across various model architectures and scales, enabling more trustworthy LLM evaluation without requiring additional training or human annotations. The codes can be found at https://github.com/TrustJudge/TrustJudge.

cs.AI↗

arXiv2025

Learning to Ball: Composing Policies for Long-Horizon Basketball Moves

Learning a control policy for a multi-phase, long-horizon task, such as basketball maneuvers, remains challenging for reinforcement learning approaches due to the need for seamless policy composition and transitions between skills. A long-horizon task typically consists of distinct subtasks with well-defined goals, separated by transitional subtasks with unclear goals but critical to the success of the entire task. Existing methods like the mixture of experts and skill chaining struggle with tasks where individual policies do not share significant commonly explored states or lack well-defined initial and terminal states between different phases. In this paper, we introduce a novel policy integration framework to enable the composition of drastically different motor skills in multi-phase long-horizon tasks with ill-defined intermediate states. Based on that, we further introduce a high-level soft router to enable seamless and robust transitions between the subtasks. We evaluate our framework on a set of fundamental basketball skills and challenging transitions. Policies trained by our approach can effectively control the simulated character to interact with the ball and accomplish the long-horizon task specified by real-time user commands, without relying on ball trajectory references.

cs.GR↗

arXiv2025

VideoJudge: Bootstrapping Enables Scalable Supervision of MLLM-as-a-Judge for Video Understanding

Precisely evaluating video understanding models remains challenging: commonly used metrics such as BLEU, ROUGE, and BERTScore fail to capture the fineness of human judgment, while obtaining such judgments through manual evaluation is costly. Recent work has explored using large language models (LLMs) or multimodal LLMs (MLLMs) as evaluators, but their extension to video understanding remains relatively unexplored. In this work, we introduce VideoJudge, a 3B and 7B-sized MLLM judge specialized to evaluate outputs from video understanding models (\textit{i.e.}, text responses conditioned on videos). To train VideoJudge, our recipe builds on the interplay between a generator and an evaluator: the generator is prompted to produce responses conditioned on a target rating, and responses not matching the evaluator's rating are discarded. Across three out of four meta-evaluation benchmarks, VideoJudge-7B outperforms larger MLLM judge baselines such as Qwen2.5-VL (32B and 72B). Notably, we find that LLM judges (Qwen3) models perform worse than MLLM judges (Qwen2.5-VL) and long chain-of-thought reasoning does not improve performance, indicating that providing video inputs is crucial for evaluation of video understanding tasks.

cs.CV↗

arXiv2025

RealBench: A Chinese Multi-image Understanding Benchmark Close to Real-world Scenarios

While various multimodal multi-image evaluation datasets have been emerged, but these datasets are primarily based on English, and there has yet to be a Chinese multi-image dataset. To fill this gap, we introduce RealBench, the first Chinese multimodal multi-image dataset, which contains 9393 samples and 69910 images. RealBench distinguishes itself by incorporating real user-generated content, ensuring high relevance to real-world applications. Additionally, the dataset covers a wide variety of scenes, image resolutions, and image structures, further increasing the difficulty of multi-image understanding. Ultimately, we conduct a comprehensive evaluation of RealBench using 21 multimodal LLMs of different sizes, including closed-source models that support multi-image inputs as well as open-source visual and video models. The experimental results indicate that even the most powerful closed-source models still face challenges when handling multi-image Chinese scenarios. Moreover, there remains a noticeable performance gap of around 71.8\% on average between open-source visual/video models and closed-source models. These results show that RealBench provides an important research foundation for further exploring multi-image understanding capabilities in the Chinese context.

cs.CL↗

arXiv2025

CS-FLEURS: A Massively Multilingual and Code-Switched Speech Dataset

We present CS-FLEURS, a new dataset for developing and evaluating code-switched speech recognition and translation systems beyond high-resourced languages. CS-FLEURS consists of 4 test sets which cover in total 113 unique code-switched language pairs across 52 languages: 1) a 14 X-English language pair set with real voices reading synthetically generated code-switched sentences, 2) a 16 X-English language pair set with generative text-to-speech 3) a 60 {Arabic, Mandarin, Hindi, Spanish}-X language pair set with the generative text-to-speech, and 4) a 45 X-English lower-resourced language pair test set with concatenative text-to-speech. Besides the four test sets, CS-FLEURS also provides a training set with 128 hours of generative text-to-speech data across 16 X-English language pairs. Our hope is that CS-FLEURS helps to broaden the scope of future code-switched speech research. Dataset link: https://huggingface.co/datasets/byan/cs-fleurs.

cs.CL↗

arXiv2025

LLM Bazaar: A Service Design for Supporting Collaborative Learning with an LLM-Powered Multi-Party Collaboration Infrastructure

For nearly two decades, conversational agents have played a critical role in structuring interactions in collaborative learning, shaping group dynamics, and supporting student engagement. The recent integration of large language models (LLMs) into these agents offers new possibilities for fostering critical thinking and collaborative problem solving. In this work, we begin with an open source collaboration support architecture called Bazaar and integrate an LLM-agent shell that enables introduction of LLM-empowered, real time, context sensitive collaborative support for group learning. This design and infrastructure paves the way for exploring how tailored LLM-empowered environments can reshape collaborative learning outcomes and interaction patterns.

cs.HC↗

arXiv2025

Robust optimal stopping with regime switching

In this paper, we study an optimal stopping problem in the presence of model uncertainty and regime switching. The max-min formulation for robust control and the dynamic programming approach are adopted to establish a general theoretical framework for such kind of problem. First, based on the dynamic programming principle, the value function of the optimal stopping problem is characterized as the unique viscosity solution to the associated Hamilton-Jacobi-Bellman equation. Then, the so-called smooth-fit principle for optimal stopping problems is proved in the current context, and a verification theorem consisting of a set of sufficient conditions for robust optimality is established. Moreover, when the Markov chain has a large state space and exhibits a two-time-scale structure, a singular perturbation approach is utilized to reduce the complexity involved and obtain an asymptotically optimal solution. Finally, an example of choosing the best time to sell a stock is provided, in which numerical experiments are reported to illustrate the implications of model uncertainty and regime switching.

math.OC↗

arXiv2025

Human-Object Interaction from Human-Level Instructions

Intelligent agents must autonomously interact with the environments to perform daily tasks based on human-level instructions. They need a foundational understanding of the world to accurately interpret these instructions, along with precise low-level movement and interaction skills to execute the derived actions. In this work, we propose the first complete system for synthesizing physically plausible, long-horizon human-object interactions for object manipulation in contextual environments, driven by human-level instructions. We leverage large language models (LLMs) to interpret the input instructions into detailed execution plans. Unlike prior work, our system is capable of generating detailed finger-object interactions, in seamless coordination with full-body movements. We also train a policy to track generated motions in physics simulation via reinforcement learning (RL) to ensure physical plausibility of the motion. Our experiments demonstrate the effectiveness of our system in synthesizing realistic interactions with diverse objects in complex environments, highlighting its potential for real-world applications.

cs.AI↗

arXiv2025

Indefinite Linear-Quadratic Partially Observed Mean-Field Game

This paper investigates an indefinite linear-quadratic partially observed mean-field game with common noise, incorporating both state-average and control-average effects. In our model, each agent's state is observed through both individual and public observations, which are modeled as general stochastic processes rather than Brownian motions. {It is noteworthy that} the weighting matrices in the cost functional are allowed to be indefinite. We derive the optimal decentralized strategies using the Hamiltonian approach and establish the well-posedness of the resulting Hamiltonian system by employing a relaxed compensator. The associated consistency condition and the feedback representation of decentralized strategies are also established. Furthermore, we demonstrate that the set of decentralized strategies form an $\varepsilon$-Nash equilibrium. As an application, we solve a mean-variance portfolio selection problem.

math.OC↗

arXiv2025

Super-additivity of quantum capacity in simple channels

The super-additivity of quantum channel capacity is an important feature of quantum information theory different from classical theory, which has been attracting attention. Recently a special channel called ``platypus channel'' exhibits super-additive quantum capacity when combined with qudit erasure channels. Here we consider the ``generalized platypus channel'', prove that it has computable channel capacities, such as both private and classical capacity equal to $1$, and in particular, the generalized platypus channel still displays the super-additivity of quantum capacity when combined with qudit erasure channels and multilevel amplitude damping channels respectively.

quant-ph↗

arXiv2025

PersuasiveToM: A Benchmark for Evaluating Machine Theory of Mind in Persuasive Dialogues

The ability to understand and predict the mental states of oneself and others, known as the Theory of Mind (ToM), is crucial for effective social scenarios. Although recent studies have evaluated ToM in Large Language Models (LLMs), existing benchmarks focus on simplified settings (e.g., Sally-Anne-style tasks) and overlook the complexity of real-world social interactions. To mitigate this gap, we propose PersuasiveToM, a benchmark designed to evaluate the ToM abilities of LLMs in persuasive dialogues. Our framework contains two core tasks: ToM Reasoning, which tests tracking of evolving desires, beliefs, and intentions; and ToM Application, which assesses the use of inferred mental states to predict and evaluate persuasion strategies. Experiments across eight leading LLMs reveal that while models excel on multiple questions, they struggle with the tasks that need tracking the dynamics and shifts of mental states and understanding the mental states in the whole dialogue comprehensively. Our aim with PersuasiveToM is to allow an effective evaluation of the ToM reasoning ability of LLMs with more focus on complex psychological activities. Our code is available at https://github.com/Yu-Fangxu/PersuasiveToM.

cs.CL↗

arXiv2025

Recursive Optimal Stopping with Poisson Stopping Constraints

This paper solves a recursive optimal stopping problem with Poisson stopping constraints using the penalized backward stochastic differential equation (PBSDE) with jumps. Stopping in this problem is only allowed at Poisson random intervention times, and jumps play a significant role not only through the stopping times but also in the recursive objective functional and model coefficients. To solve the problem, we propose a decomposition method based on Jacod-Pham that allows us to separate the problem into a series of sub-problems between each pair of consecutive Poisson stopping times. To represent the value function of the recursive optimal stopping problem when the initial time falls between two consecutive Poisson stopping times and the generator is concave/convex, we leverage the comparison theorem of BSDEs with jumps. We then apply the representation result to American option pricing in a nonlinear market with Poisson stopping constraints.

math.OC↗