arXiv subjects

Weiyuan Li

Publications and source records attributed to Weiyuan Li.

18 recordsLinked to original sources

arXiv2026

Verify Smarter, Evolve Further: Efficient Harness Evolution through Behavior-Aware Verification

Agent harnesses shape how language-model agents use instructions, tools, and runtime components, but adapting these harnesses requires costly verification. Existing propose-and-verify methods typically score every candidate on a fixed task set, wasting rollouts on unrelated behaviors and allowing aggregate scores to obscure specific regressions. We introduce HarnessLens, a budget-aware framework for automated harness evolution. HarnessLens jointly explores the task space and user-configurable components, derives candidate modifications from execution trajectories, and selectively verifies each candidate on behavior-relevant tasks using an attributable-evidence gate. Across three agent harnesses and four benchmarks, HarnessLens improves average held-out performance by 7.6-13.6% while consuming substantially less evaluation budget than competing baselines. These results demonstrate that behavior-aware verification with explicit attribution enables more reliable and sample-efficient harness evolution under constrained interaction budgets. Our code is available at https://github.com/jhxu5214/HarnessLens.

cs.AI↗

arXiv2026

Agentopia: Long-Term Life Simulation and Learning in Agent Societies

Humans learn from social life. Simulating this process with LLM-powered agents represents a promising research direction, raising a natural question: whether LLMs can learn from such simulated social experience to better understand and replicate human behavior. However, prior agent society simulations typically operate at the scale of days, limiting the depth of social interactions and long-term growth. In this paper, we study long-term life simulation and LLM learning in agent societies, with two goals: (1) investigating social behaviors that emerge from life-long simulation, and (2) developing anthropomorphic capabilities in LLMs, particularly intelligence in social life, through years of simulated social experience. Specifically, we present Agentopia, a comprehensive framework for long-term life simulation in multi-agent societies, where 100 agents autonomously pursue personal growth, develop social relationships, and fulfill their needs and goals over 10 simulated years. We define life reward to mirror human well-being, and leverage this reward to train LLMs via rejection sampling. Extensive experiments show that agents exhibit rich emergent social behaviors. Furthermore, life reward training effectively enhances the underlying LLM, which leads to improved agent well-being in simulation, and generalizes to downstream role-playing benchmarks with +15.6% improvement.

cs.CL↗

arXiv2026

HER: Human-like Reasoning and Reinforcement Learning for LLM Role-playing

LLM role-playing, i.e., using LLMs to simulate specific personas, has emerged as a key capability in various applications, such as companionship, content creation and digital games. While current models effectively capture character tones and knowledge, simulating the inner thoughts behind their behaviors remains a challenge. Towards cognitive simulation in LLM role-play, previous efforts mainly suffer from two deficiencies: lacking data with high-quality reasoning traces, and lacking reliable reward signals aligned with human preferences. In this paper, we propose HER, a unified framework for cognitive-level persona simulation. HER introduces dual-layer thinking, which distinguishes characters' first-person thinking from LLMs' third-person thinking. To bridge these gaps, we curate reasoning-augmented role-playing data via reverse engineering, and construct human-aligned principles and reward models. Leveraging these resources, we train HER models based on Qwen3-32B via supervised and reinforcement learning. Extensive experiments validate the effectiveness of our approach. Notably, our models significantly outperform the Qwen3-32B baseline, achieving a 30.26 improvement on the CoSER benchmark and a 14.97% gain on the Minimax Role-Play Bench. Our datasets, principles, and models are released to facilitate future research.

cs.LG↗

arXiv2026

Synthetic Data Powers Product Retrieval for Long-tail Knowledge-Intensive Queries in E-commerce Search

Product retrieval is the backbone of e-commerce search: for each user query, it identifies a high-recall candidate set from billions of items, laying the foundation for high-quality ranking and user experience. Despite extensive optimization for mainstream queries, existing systems still struggle with long-tail queries, especially knowledge-intensive ones. These queries exhibit diverse linguistic patterns, often lack explicit purchase intent, and require domain-specific knowledge reasoning for accurate interpretation. They also suffer from a shortage of reliable behavioral logs, which makes such queries a persistent challenge for retrieval optimization. To address these issues, we propose an efficient data synthesis framework tailored to retrieval involving long-tail, knowledge-intensive queries. The key idea is to implicitly distill the capabilities of a powerful offline query-rewriting model into an efficient online retrieval system. Leveraging the strong language understanding of LLMs, we train a multi-candidate query rewriting model with multiple reward signals and capture its rewriting capability in well-curated query-product pairs through a powerful offline retrieval pipeline. This design mitigates distributional shift in rewritten queries, which might otherwise limit incremental recall or introduce irrelevant products. Experiments demonstrate that without any additional tricks, simply incorporating this synthetic data into retrieval model training leads to significant improvements. Online Side-By-Side (SBS) human evaluation results indicate a notable enhancement in user search experience.

cs.IR↗

arXiv2026

HumanLLM: Benchmarking and Improving LLM Anthropomorphism via Human Cognitive Patterns

Large Language Models (LLMs) have demonstrated remarkable capabilities in reasoning and generation, serving as the foundation for advanced persona simulation and Role-Playing Language Agents (RPLAs). However, achieving authentic alignment with human cognitive and behavioral patterns remains a critical challenge for these agents. We present HumanLLM, a framework treating psychological patterns as interacting causal forces. We construct 244 patterns from $\sim$12,000 academic papers and synthesize 11,359 scenarios where 2-5 patterns reinforce, conflict, or modulate each other, with multi-turn conversations expressing inner thoughts, actions, and dialogue. Our dual-level checklists evaluate both individual pattern fidelity and emergent multi-pattern dynamics, achieving strong human alignment ($r=0.90$) while revealing that holistic metrics conflate simulation accuracy with social desirability. HumanLLM-8B outperforms Qwen3-32B on multi-pattern dynamics despite 4$\times$ fewer parameters, demonstrating that authentic anthropomorphism requires cognitive modeling -- simulating not just what humans do, but the psychological processes generating those behaviors. Our dataset, code, and model are available at:https://github.com/YJGoodbye2024/HumanLLM

cs.CL↗

arXiv2026

Letter Of Intent for a future $μ^+ \to \mathrm{e}^+ γ$ experiment at the High Intensity Muon Beam facility at PSI

Searches for charged lepton flavor violation in the muon sector stand out among the most sensitive and clean probes for physics beyond the Standard Model. Currently, $μ^+ \to \mathrm{e}^+ γ$ experiments provide the best constraints in this field for a wide range of models while, in the coming years, new experiments investigating the processes of $μ^+ \to \mathrm{e}^+ \mathrm{e}^+ \mathrm{e}^-$ and $μ\to \mathrm{e}$ conversion in the nuclear field are anticipated to reach comparable or higher sensitivities. The High-Intensity Muon Beam (HIMB) facility at PSI, which is expected to deliver muon beam intensities up to two orders of magnitude higher than the existing beam lines, offers a unique opportunity to significantly enhance the sensitivity of $μ^+ \to \mathrm{e}^+ γ$ searches. The discovery potential could be substantially boosted and a sensitivity comparable to that of all the other projects could be reestablished, which is essential for discriminating among competing new-physics scenarios should an observation occur in any of the channels. In this document, we express our interest in developing a $μ^+ \to \mathrm{e}^+ γ$ experimental program at HIMB, with the goal of improving, within the next decade, the sensitivity of the $μ^+ \to \mathrm{e}^+ γ$ search by more than one order of magnitude relative to the expected final result of the current leading experiment, MEG II. This effort would ensure that PSI retains its leadership in this field.

hep-ex↗

arXiv2025

EulerESG: Automating ESG Disclosure Analysis with LLMs

Environmental, Social, and Governance (ESG) reports have become central to how companies communicate climate risk, social impact, and governance practices, yet they are still published primarily as long, heterogeneous PDF documents. This makes it difficult to systematically answer seemingly simple questions. Existing tools either rely on brittle rule-based extraction or treat ESG reports as generic text, without explicitly modelling the underlying reporting standards. We present \textbf{EulerESG}, an LLM-powered system for automating ESG disclosure analysis with explicit awareness of ESG frameworks. EulerESG combines (i) dual-channel retrieval and LLM-driven disclosure analysis over ESG reports, and (ii) an interactive dashboard and chatbot for exploration, benchmarking, and explanation. Using four globally recognised companies and twelve SASB sub-industries, we show that EulerESG can automatically populate standard-aligned metric tables with high fidelity (up to 0.95 average accuracy) while remaining practical in end-to-end runtime, and we compare several recent LLM models in this setting. The full implementation, together with a demonstration video, is publicly available at https://github.com/UNSW-database/EulerESG.

cs.CL↗

arXiv2025

Curse of Knowledge: When Complex Evaluation Context Benefits yet Biases LLM Judges

As large language models (LLMs) grow more capable, they face increasingly diverse and complex tasks, making reliable evaluation challenging. The paradigm of LLMs as judges has emerged as a scalable solution, yet prior work primarily focuses on simple settings. Their reliability in complex tasks--where multi-faceted rubrics, unstructured reference answers, and nuanced criteria are critical--remains understudied. In this paper, we constructed ComplexEval, a challenge benchmark designed to systematically expose and quantify Auxiliary Information Induced Biases. We systematically investigated and validated 6 previously unexplored biases across 12 basic and 3 advanced scenarios. Key findings reveal: (1) all evaluated models exhibit significant susceptibility to these biases, with bias magnitude scaling with task complexity; (2) notably, Large Reasoning Models (LRMs) show paradoxical vulnerability. Our in-depth analysis offers crucial insights for improving the accuracy and verifiability of evaluation signals, paving the way for more general and robust evaluation models.

cs.CL↗

arXiv2025

Development of a high-rate capable DLC-RPC based on a current evacuation pattern

A Resistive Plate Chamber using Diamond-Like Carbon electrodes (DLC-RPC) has been developed as a background tagging detector in the MEG$~$II experiment. The DLC-RPC is planned to be installed in a high-intensity and low-momentum muon beam. This detector is required to have a detection efficiency above 90 % with four active gaps in the muon beam due to the limitation of the material budget. In such an environment, the high current flowing through the resistive electrodes causes a voltage drop, which reduces the performance of the DLC-RPC. This voltage drop can be suppressed by implementing a current evacuation pattern, though discharges are more likely to occur near the pattern. Therefore the pattern must be covered by a protection cover made of an insulator. In this study, electrode samples with a current evacuation pattern and different widths of protection cover (0.2 mm and 0.8 mm) have been produced, and their performance and stability were measured. The detection efficiency of a single-gap chamber for $β$-rays from a $^{90}$Sr source was measured to be up to approximately 60 % in both electrode samples. The target efficiency can be achieved even with a drop of 100 $-$ 150 V. On the other hand, after more than a dozen hours of operation, discharges suddenly occurred and the detector was prevented from further operation. These discharges created current paths on the spacing pillars. This serious problem must be investigated and solved in the future.

physics.ins-det↗

arXiv2025

Enhancing Persona Consistency for LLMs' Role-Playing using Persona-Aware Contrastive Learning

In recent years, large language models (LLMs) have achieved breakthrough progress in many dialogue generation tasks. However, their lack of emotion and fine-grained role awareness limits the model's ability to provide personalized and diverse interactions further. Current methods face high costs in collecting high-quality annotated data for scenarios such as role-playing, and traditional human alignment methods are difficult to deploy due to the inherent diversity of model behavior in role-playing scenarios. Inspired by the alignment of models for safety behaviors through RLHF (Reinforcement Learning from Human Feedback), in this paper, we revisit model role-playing behavior from the perspective of persona alignment and propose a novel annotation-free framework named \textbf{\underline{P}}ersona-Aware \textbf{\underline{C}}ontrastive \textbf{\underline{L}}earning (PCL) to align LLMs' behavior during role-playing, enhancing the model's role consistency. Specifically, we first design a role chain method to encourage the model to self-question based on the role characteristics and dialogue context to adjust personality consistency. Then, we further enhance the model's role-playing strategy through iterative contrastive learning between the use of role characteristics and not. Experiments on both black-box and white-box LLMs show that LLMs equipped with PCL significantly outperform vanilla LLMs under automatic evaluation methods (CharEval \& GPT-4) and human expert evaluation.

cs.CL↗

arXiv2025

SmartRAG: Jointly Learn RAG-Related Tasks From the Environment Feedback

RAG systems consist of multiple modules to work together. However, these modules are usually separately trained. We argue that a system like RAG that incorporates multiple modules should be jointly optimized to achieve optimal performance. To demonstrate this, we design a specific pipeline called \textbf{SmartRAG} that includes a policy network and a retriever. The policy network can serve as 1) a decision maker that decides when to retrieve, 2) a query rewriter to generate a query most suited to the retriever, and 3) an answer generator that produces the final response with/without the observations. We then propose to jointly optimize the whole system using a reinforcement learning algorithm, with the reward designed to encourage the system to achieve the best performance with minimal retrieval cost. When jointly optimized, all the modules can be aware of how other modules are working and thus find the best way to work together as a complete system. Empirical results demonstrate that the jointly optimized SmartRAG can achieve better performance than separately optimized counterparts.

cs.IR↗

arXiv2024

Revenue Management with Calendar-Aware and Dependent Demands: Asymptotically Tight Fluid Approximations

When modeling the demand in revenue management systems, a natural approach is to focus on a canonical interval of time, such as a week, so that we forecast the demand over each week in the selling horizon. Ideally, we would like to use random variables with general distributions to model the demand over each week. The current demand can give a signal for the future demand, so we also would like to capture the dependence between the demands over different weeks. Prevalent demand models in the literature, which are based on a discrete-time approximation to a Poisson process, are not compatible with these needs. In this paper, we focus on revenue management models that are compatible with a natural approach for forecasting the demand. Building such models through dynamic programming is not difficult. We divide the selling horizon into multiple stages, each stage being a canonical interval of time on the calendar. We have random number of customer arrivals in each stage, whose distribution is arbitrary and depends on the number of arrivals in the previous stage. The question we seek to answer is the form of the corresponding fluid approximation. We give the correct fluid approximation in the sense that it yields asymptotically optimal policies. The form of our fluid approximation is surprising as its constraints use expected capacity consumption of a resource up to a certain time period, conditional on the demand in the stage just before the time period in question. As the resource capacities and number of stages increase with the same rate, our performance guarantee converges to one. To our knowledge, this result gives the first asymptotically optimal policy under dependent demands with arbitrary distributions. Our computational experiments indicate that using the correct fluid approximation can make a dramatic impact in practice.

math.OC↗

arXiv2024

Radiation Hardness Studies of RPC Based on Diamond-Like Carbon Electrodes for MEG II Experiment

A novel type of resistive plate chamber, based on diamond-like carbon (DLC) electrodes is under development for background identification in the MEG II experiment. The DLC-RPC is required to have a radiation hardness to mass irradiation since it is planned to be placed in a high-intensity and low-momentum muon beam. In this study, the aging test using a high-intensity X-ray beam was conducted to evaluate the radiation hardness of the DLC-RPC. The accumulated charge due to X-ray irradiation reached about 54 C/cm$^2$, which is approximately half of the one-year irradiation dose expected in the MEG II experiment. As a result, the degradation of the gas gain was observed due to fluorine deposition and insulators formed on the DLC electrodes. In addition, discharges via spacers were also observed repeatedly and interrupted the DLC-RPC operation.

physics.ins-det↗

arXiv2024

Prototype study of $0.1\%\,X_0$ and $\mathrm{MHz/cm^2}$ tolerant Resistive Plate Chamber with Diamond-Like Carbon electrodes

A novel Resistive Plate Chamber (RPC) was designed with Diamond-Like Carbon (DLC) electrodes and performance studies were carried out for 384$\,\mathrm{μm}$ gap configuration with a $2\,\mathrm{cm}\times2\,\mathrm{cm}$ prototype. The use of thin films coated with DLC enables an ultra-low mass design of $< 0.1\%\,X_0$ with up to a four-layer configuration. At the same time, 42% MIP efficiency, and 180 ps timing resolution per layer were achieved in a measurement performed under a $1\,\mathrm{MHz/cm^2}$ non-MIP charged particle beam. In addition, we propose a further improved design for a $20\,\mathrm{cm}$-scale detector that can achieve 90% four-layer efficiency in an even higher $4\,\mathrm{MHz/cm^2}$ beam. In this paper, we describe the detector design, present the results of performance measurements, and characterize the rate capability of the DLC-based RPCs with a performance projection for an improved design.

physics.ins-det↗

arXiv2023

MDSC: Towards Evaluating the Style Consistency Between Music and Dance

We propose MDSC(Music-Dance-Style Consistency), the first evaluation metric that assesses to what degree the dance moves and music match. Existing metrics can only evaluate the motion fidelity and diversity and the degree of rhythmic matching between music and dance. MDSC measures how stylistically correlated the generated dance motion sequences and the conditioning music sequences are. We found that directly measuring the embedding distance between motion and music is not an optimal solution. We instead tackle this through modeling it as a clustering problem. Specifically, 1) we pre-train a music encoder and a motion encoder, then 2) we learn to map and align the motion and music embedding in joint space by jointly minimizing the intra-cluster distance and maximizing the inter-cluster distance, and 3) for evaluation purposes, we encode the dance moves into embedding and measure the intra-cluster and inter-cluster distances, as well as the ratio between them. We evaluate our metric on the results of several music-conditioned motion generation methods, combined with user study, we found that our proposed metric is a robust evaluation metric in measuring the music-dance style correlation.

cs.SD↗

arXiv2023

Controlling Character Motions without Observable Driving Source

How to generate diverse, life-like, and unlimited long head/body sequences without any driving source? We argue that this under-investigated research problem is non-trivial at all, and has unique technical challenges behind it. Without semantic constraints from the driving sources, using the standard autoregressive model to generate infinitely long sequences would easily result in 1) out-of-distribution (OOD) issue due to the accumulated error, 2) insufficient diversity to produce natural and life-like motion sequences and 3) undesired periodic patterns along the time. To tackle the above challenges, we propose a systematic framework that marries the benefits of VQ-VAE and a novel token-level control policy trained with reinforcement learning using carefully designed reward functions. A high-level prior model can be easily injected on top to generate unlimited long and diverse sequences. Although we focus on no driving sources now, our framework can be generalized for controlled synthesis with explicit driving sources. Through comprehensive evaluations, we conclude that our proposed framework can address all the above-mentioned challenges and outperform other strong baselines very significantly.

cs.AI↗

arXiv2020

DoveNet: Deep Image Harmonization via Domain Verification

Image composition is an important operation in image processing, but the inconsistency between foreground and background significantly degrades the quality of composite image. Image harmonization, aiming to make the foreground compatible with the background, is a promising yet challenging task. However, the lack of high-quality publicly available dataset for image harmonization greatly hinders the development of image harmonization techniques. In this work, we contribute an image harmonization dataset iHarmony4 by generating synthesized composite images based on COCO (resp., Adobe5k, Flickr, day2night) dataset, leading to our HCOCO (resp., HAdobe5k, HFlickr, Hday2night) sub-dataset. Moreover, we propose a new deep image harmonization method DoveNet using a novel domain verification discriminator, with the insight that the foreground needs to be translated to the same domain as background. Extensive experiments on our constructed dataset demonstrate the effectiveness of our proposed method. Our dataset and code are available at https://github.com/bcmi/Image_Harmonization_Datasets.

cs.CV↗

arXiv2020

Image Harmonization Dataset iHarmony4: HCOCO, HAdobe5k, HFlickr, and Hday2night

Image composition is an important operation in image processing, but the inconsistency between foreground and background significantly degrades the quality of composite image. Image harmonization, which aims to make the foreground compatible with the background, is a promising yet challenging task. However, the lack of high-quality public dataset for image harmonization, which significantly hinders the development of image harmonization techniques. Therefore, we contribute an image harmonization dataset iHarmony4 by generating synthesized composite images based on existing COCO (resp., Adobe5k, day2night) dataset, leading to our HCOCO (resp., HAdobe5k, Hday2night) sub-dataset. To enrich the diversity of our dataset, we also generate synthesized composite images based on our collected Flick images, leading to our HFlickr sub-dataset. The image harmonization dataset iHarmony4 is released at https://github.com/bcmi/Image_Harmonization_Datasets.

cs.CV↗