arXiv subjects

Xuewei Wang

Publications and source records attributed to Xuewei Wang.

17 recordsLinked to original sources

arXiv2026

Convergence rates of Wasserstein gradient flows for nonlinear Fokker-Planck equations with mobility and related inequalities

For nonlinear Fokker-Planck equations with mobility, the Wasserstein gradient flow structure is described by the generalized relative entropy as the energy functional and the modified Wasserstein metric $W_h$ as the associated metric structure. This work investigates the nonlinear effects induced by mobility and establishes the corresponding inequalities. For nonlinear diffusion, we establish a logarithmic Sobolev inequality, which yields the convergence rate of the free energy functional and the Talagrand inequality. By further exploiting the relationship between the weighted homogeneous Sobolev norm and the $W_h$ metric, we derive an HWI inequality relating the relative entropy, the $W_h$ metric, and the Fisher information. In the case of mobility dependent drift and linear diffusion, the convergence rate in the $W_h$ metric is also obtained by applying the Girsanov theorem.

math.AP↗

arXiv2026

Nonlinear Vlasov-Fokker-Planck equations: From generalized Wasserstein gradient flow to GENERIC structure

We study the GENERIC (General Equation for Non-Equilibrium Reversible Irreversible Coupling) formulation of the nonlinear Vlasov-Fokker-Planck equation from the perspective of gradient flows along trajectories. After pulling back the reversible component, the evolution can be recast as a generalized Wasserstein gradient flow. The associated free energy functional consists of an entropy term and a conservative energy term, while the metric is induced by the Onsager operator. This trajectory based viewpoint shows that the trajectorial rate of free energy dissipation captures the influence of the nonlinear term, an effect that is not directly apparent at the macroscopic level. Finally, partial degeneracy yields a decomposition of the underlying metric space, which in turn enables the derivation of a partial HWI inequality.

math.AP↗

arXiv2026

Examining Reasoning LLMs-as-Judges in Non-Verifiable LLM Post-Training

Reasoning LLMs-as-Judges, which can benefit from inference-time scaling, provide a promising path for extending the success of reasoning models to non-verifiable domains where the output correctness/quality cannot be directly checked. However, while reasoning judges have shown better performance on static evaluation benchmarks, their effectiveness in actual policy training has not been systematically examined. Therefore, we conduct a rigorous study to investigate the actual impact of non-reasoning and reasoning judges in reinforcement-learning-based LLM alignment. Our controlled synthetic setting, where a "gold-standard" judge (gpt-oss-120b) provides preference annotations to train smaller judges, reveals key differences between non-reasoning and reasoning judges: non-reasoning judges lead to reward hacking easily, while reasoning judges can lead to policies that achieve strong performance when evaluated by the gold-standard judge. Interestingly, we find that the reasoning-judge-trained policies achieve such strong performance by learning to generate highly effective adversarial outputs that can also score well on popular benchmarks such as Arena-Hard by deceiving other LLM-judges. Combined with our further analysis, our study highlights both important findings and room for improvements for applying (reasoning) LLM-judges in non-verifiable LLM post-training.

cs.AI↗

arXiv2025

Deep Think with Confidence

Large Language Models (LLMs) have shown great potential in reasoning tasks through test-time scaling methods like self-consistency with majority voting. However, this approach often leads to diminishing returns in accuracy and high computational overhead. To address these challenges, we introduce Deep Think with Confidence (DeepConf), a simple yet powerful method that enhances both reasoning efficiency and performance at test time. DeepConf leverages model-internal confidence signals to dynamically filter out low-quality reasoning traces during or after generation. It requires no additional model training or hyperparameter tuning and can be seamlessly integrated into existing serving frameworks. We evaluate DeepConf across a variety of reasoning tasks and the latest open-source models, including Qwen 3 and GPT-OSS series. Notably, on challenging benchmarks such as AIME 2025, DeepConf@512 achieves up to 99.9% accuracy and reduces generated tokens by up to 84.7% compared to full parallel thinking.

cs.LG↗

arXiv2025

Boosting LLM Reasoning via Spontaneous Self-Correction

While large language models (LLMs) have demonstrated remarkable success on a broad range of tasks, math reasoning remains a challenging one. One of the approaches for improving math reasoning is self-correction, which designs self-improving loops to let the model correct its own mistakes. However, existing self-correction approaches treat corrections as standalone post-generation refinements, relying on extra prompt and system designs to elicit self-corrections, instead of performing real-time, spontaneous self-corrections in a single pass. To address this, we propose SPOC, a spontaneous self-correction approach that enables LLMs to generate interleaved solutions and verifications in a single inference pass, with generation dynamically terminated based on verification outcomes, thereby effectively scaling inference time compute. SPOC considers a multi-agent perspective by assigning dual roles -- solution proposer and verifier -- to the same model. We adopt a simple yet effective approach to generate synthetic data for fine-tuning, enabling the model to develop capabilities for self-verification and multi-agent collaboration. We further improve its solution proposal and verification accuracy through online reinforcement learning. Experiments on mathematical reasoning benchmarks show that SPOC significantly improves performance. Notably, SPOC boosts the accuracy of Llama-3.1-8B and 70B Instruct models, achieving gains of 8.8% and 11.6% on MATH500, 10.0% and 20.0% on AMC23, and 3.3% and 6.7% on AIME24, respectively.

cs.AI↗

arXiv2025

Reinforcement Learning-based Token Pruning in Vision Transformers: A Markov Game Approach

Vision Transformers (ViTs) have computational costs scaling quadratically with the number of tokens, calling for effective token pruning policies. Most existing policies are handcrafted, lacking adaptivity to varying inputs. Moreover, they fail to consider the sequential nature of token pruning across multiple layers. In this work, for the first time (as far as we know), we exploit Reinforcement Learning (RL) to data-adaptively learn a pruning policy. Formulating token pruning as a sequential decision-making problem, we model it as a Markov Game and utilize Multi-Agent Proximal Policy Optimization (MAPPO) where each agent makes an individualized pruning decision for a single token. We also develop reward functions that enable simultaneous collaboration and competition of these agents to balance efficiency and accuracy. On the well-known ImageNet-1k dataset, our method improves the inference speed by up to 44% while incurring only a negligible accuracy drop of 0.4%. The source code is available at https://github.com/daashuai/rl4evit.

cs.CV↗

arXiv2025

A trajectorial approach to the gradient flow of McKean-Vlasov SDEs with mobility

We establish the gradient flow representation of diffusion with mobility $b$ with respect to the modified Wasserstein quasi-metric $W_h$, where $h(r)=rb(r)$. The appropriate selection of the free energy functional depends on the specific form of the generalized entropy. Different from the JKO scheme, we derive the trajectorial version of the relative entropy dissipation identity for the McKean-Vlasov stochastic differential equation (SDE) with Nemytskii-type coefficients, utilizing techniques from stochastic analysis. Based on this, we demonstrate that the trajectorial average of the solution process to the McKean-Vlasov SDE, with respect to the underlying measure, corresponds to the rate of dissipation of the free energy. As an application, we present the energy dissipation of the Fermi-Dirac-Fokker-Planck equation, a model widely used in physics and biology to describe saturation effects. Inspired by numerical simulations, we propose two questions on condensation phenomena and non-exponential convergence rate.

math.PR↗

arXiv2024

Self-Generated Critiques Boost Reward Modeling for Language Models

Reward modeling is crucial for aligning large language models (LLMs) with human preferences, especially in reinforcement learning from human feedback (RLHF). However, current reward models mainly produce scalar scores and struggle to incorporate critiques in a natural language format. We hypothesize that predicting both critiques and the scalar reward would improve reward modeling ability. Motivated by this, we propose Critic-RM, a framework that improves reward models using self-generated critiques without extra supervision. Critic-RM employs a two-stage process: generating and filtering high-quality critiques, followed by joint fine-tuning on reward prediction and critique generation. Experiments across benchmarks show that Critic-RM improves reward modeling accuracy by 3.7%-7.3% compared to standard reward models and LLM judges, demonstrating strong performance and data efficiency. Additional studies further validate the effectiveness of generated critiques in rectifying flawed reasoning steps with 2.5%-3.2% gains in improving reasoning accuracy.

cs.CL↗

arXiv2024

Law of the Weakest Link: Cross Capabilities of Large Language Models

The development and evaluation of Large Language Models (LLMs) have largely focused on individual capabilities. However, this overlooks the intersection of multiple abilities across different types of expertise that are often required for real-world tasks, which we term cross capabilities. To systematically explore this concept, we first define seven core individual capabilities and then pair them to form seven common cross capabilities, each supported by a manually constructed taxonomy. Building on these definitions, we introduce CrossEval, a benchmark comprising 1,400 human-annotated prompts, with 100 prompts for each individual and cross capability. To ensure reliable evaluation, we involve expert annotators to assess 4,200 model responses, gathering 8,400 human ratings with detailed explanations to serve as reference examples. Our findings reveal that, in both static evaluations and attempts to enhance specific abilities, current LLMs consistently exhibit the "Law of the Weakest Link," where cross-capability performance is significantly constrained by the weakest component. Specifically, across 58 cross-capability scores from 17 models, 38 scores are lower than all individual capabilities, while 20 fall between strong and weak, but closer to the weaker ability. These results highlight the under-performance of LLMs in cross-capability tasks, making the identification and improvement of the weakest capabilities a critical priority for future research to optimize performance in complex, multi-dimensional scenarios.

cs.AI↗

arXiv2024

Correlation between unconventional superconductivity and strange metallicity revealed by operando superfluid density measurements

Strange-metal behavior has been observed in superconductors ranging from cuprates to pressurized nickelates, but its relationship to unconventional superconductivity remains elusive. Here, we perform operando superfluid density measurements on ion-gated FeSe films. We observe for the first time a synchronized evolution of superconducting condensate and the strange-metal phase with electron doping. A linear scaling between zero-temperature superfluid density and the strange-metal resistivity coefficient is further established, which nails down a direct link between the formation of superfluid in the superconducting state and the scattering of carriers in the strange-metal normal state. Remarkably, the scaling also applies for different iron-based and cuprate superconductors despite their distinct electronic structures and pairing symmetries. Such a correlation can be reproduced in a theoretical calculation on the two-dimensional Yukawa-Sachdev-Ye-Kitaev model by considering a cooperative effect of quantum critical fluctuation and disorder. These findings indicate a fundamental principle governing superconducting condensation and strange-metal scattering in unconventional superconductors.

cond-mat.supr-con↗

arXiv2024

Scaling User Modeling: Large-scale Online User Representations for Ads Personalization in Meta

Effective user representations are pivotal in personalized advertising. However, stringent constraints on training throughput, serving latency, and memory, often limit the complexity and input feature set of online ads ranking models. This challenge is magnified in extensive systems like Meta's, which encompass hundreds of models with diverse specifications, rendering the tailoring of user representation learning for each model impractical. To address these challenges, we present Scaling User Modeling (SUM), a framework widely deployed in Meta's ads ranking system, designed to facilitate efficient and scalable sharing of online user representation across hundreds of ads models. SUM leverages a few designated upstream user models to synthesize user embeddings from massive amounts of user features with advanced modeling techniques. These embeddings then serve as inputs to downstream online ads ranking models, promoting efficient representation sharing. To adapt to the dynamic nature of user features and ensure embedding freshness, we designed SUM Online Asynchronous Platform (SOAP), a latency free online serving system complemented with model freshness and embedding stabilization, which enables frequent user model updates and online inference of user embeddings upon each user request. We share our hands-on deployment experiences for the SUM framework and validate its superiority through comprehensive experiments. To date, SUM has been launched to hundreds of ads ranking models in Meta, processing hundreds of billions of user requests daily, yielding significant online metric gains and improved infrastructure efficiency.

cs.IR↗

arXiv2023

Towards the Better Ranking Consistency: A Multi-task Learning Framework for Early Stage Ads Ranking

Dividing ads ranking system into retrieval, early, and final stages is a common practice in large scale ads recommendation to balance the efficiency and accuracy. The early stage ranking often uses efficient models to generate candidates out of a set of retrieved ads. The candidates are then fed into a more computationally intensive but accurate final stage ranking system to produce the final ads recommendation. As the early and final stage ranking use different features and model architectures because of system constraints, a serious ranking consistency issue arises where the early stage has a low ads recall, i.e., top ads in the final stage are ranked low in the early stage. In order to pass better ads from the early to the final stage ranking, we propose a multi-task learning framework for early stage ranking to capture multiple final stage ranking components (i.e. ads clicks and ads quality events) and their task relations. With our multi-task learning framework, we can not only achieve serving cost saving from the model consolidation, but also improve the ads recall and ranking consistency. In the online A/B testing, our framework achieves significantly higher click-through rate (CTR), conversion rate (CVR), total value and better ads-quality (e.g. reduced ads cross-out rate) in a large scale industrial ads ranking system.

cs.IR↗

arXiv2020

A stability-driven protocol for drug response interpretable prediction (staDRIP)

Modern cancer -omics and pharmacological data hold great promise in precision cancer medicine for developing individualized patient treatments. However, high heterogeneity and noise in such data pose challenges for predicting the response of cancer cell lines to therapeutic drugs accurately. As a result, arbitrary human judgment calls are rampant throughout the predictive modeling pipeline. In this work, we develop a transparent stability-driven pipeline for drug response interpretable predictions, or staDRIP, which builds upon the PCS framework for veridical data science (Yu and Kumbier, 2020) and mitigates the impact of human judgment calls. Here we use the PCS framework for the first time in cancer research to extract proteins and genes that are important in predicting the drug responses and stable across appropriate data and model perturbations. Out of the 24 most stable proteins we identified using data from the Cancer Cell Line Encyclopedia (CCLE), 18 have been associated with the drug response or identified as a known or possible drug target in previous literature, demonstrating the utility of our stability-driven pipeline for knowledge discovery in cancer drug response prediction modeling.

q-bio.QM↗

arXiv2020

Unified model selection approach based on minimum description length principle in Granger causality analysis

Granger causality analysis (GCA) provides a powerful tool for uncovering the patterns of brain connectivity mechanism using neuroimaging techniques. Conventional GCA applies two different mathematical theories in a two-stage scheme: (1) the Bayesian information criterion (BIC) or Akaike information criterion (AIC) for the regression model orders associated with endogenous and exogenous information; (2) F-statistics for determining the causal effects of exogenous variables. While specifying endogenous and exogenous effects are essentially the same model selection problem, this could produce different benchmarks in the two stages and therefore degrade the performance of GCA. In this course, we present a unified model selection approach based on the minimum description length (MDL) principle for GCA in the context of the general regression model paradigm. Compared with conventional methods, our approach emphasize that a single mathematical theory should be held throughout the GCA process. Under this framework, all candidate models within the model space might be compared freely in the context of the code length, without the need for an intermediate model. We illustrate its advantages over conventional two-stage GCA approach in a 3-node network and a 5-node network synthetic experiments. The unified model selection approach is capable of identifying the actual connectivity while avoiding the false influences of noise. More importantly, the proposed approach obtained more consistent results in a challenge fMRI dataset for causality investigation, mental calculation network under visual and auditory stimulus, respectively. The proposed approach has potential to accommodate other Granger causality representations in other function space. The comparison between different GC representations in different function spaces can also be naturally deal with in the framework.

stat.ME↗

arXiv2020

Effects of Persuasive Dialogues: Testing Bot Identities and Inquiry Strategies

Intelligent conversational agents, or chatbots, can take on various identities and are increasingly engaging in more human-centered conversations with persuasive goals. However, little is known about how identities and inquiry strategies influence the conversation's effectiveness. We conducted an online study involving 790 participants to be persuaded by a chatbot for charity donation. We designed a two by four factorial experiment (two chatbot identities and four inquiry strategies) where participants were randomly assigned to different conditions. Findings showed that the perceived identity of the chatbot had significant effects on the persuasion outcome (i.e., donation) and interpersonal perceptions (i.e., competence, confidence, warmth, and sincerity). Further, we identified interaction effects among perceived identities and inquiry strategies. We discuss the findings for theoretical and practical implications for developing ethical and effective persuasive chatbots. Our published data, codes, and analyses serve as the first step towards building competent ethical persuasive chatbots.

cs.HC↗

arXiv2020

Persuasion for Good: Towards a Personalized Persuasive Dialogue System for Social Good

Developing intelligent persuasive conversational agents to change people's opinions and actions for social good is the frontier in advancing the ethical development of automated dialogue systems. To do so, the first step is to understand the intricate organization of strategic disclosures and appeals employed in human persuasion conversations. We designed an online persuasion task where one participant was asked to persuade the other to donate to a specific charity. We collected a large dataset with 1,017 dialogues and annotated emerging persuasion strategies from a subset. Based on the annotation, we built a baseline classifier with context information and sentence-level features to predict the 10 persuasion strategies used in the corpus. Furthermore, to develop an understanding of personalized persuasion processes, we analyzed the relationships between individuals' demographic and psychological backgrounds including personality, morality, value systems, and their willingness for donation. Then, we analyzed which types of persuasion strategies led to a greater amount of donation depending on the individuals' personal backgrounds. This work lays the ground for developing a personalized persuasive dialogue system.

cs.CL↗

arXiv2019

How to Build User Simulators to Train RL-based Dialog Systems

User simulators are essential for training reinforcement learning (RL) based dialog models. The performance of the simulator directly impacts the RL policy. However, building a good user simulator that models real user behaviors is challenging. We propose a method of standardizing user simulator building that can be used by the community to compare dialog system quality using the same set of user simulators fairly. We present implementations of six user simulators trained with different dialog planning and generation methods. We then calculate a set of automatic metrics to evaluate the quality of these simulators both directly and indirectly. We also ask human users to assess the simulators directly and indirectly by rating the simulated dialogs and interacting with the trained systems. This paper presents a comprehensive evaluation framework for user simulator study and provides a better understanding of the pros and cons of different user simulators, as well as their impacts on the trained systems.

cs.CL↗