arXiv subjects

Tangjie Lv

Publications and source records attributed to Tangjie Lv.

At least 19 recordsLinked to original sources

arXiv2026

DynSess: Dynamic Session-Level Evaluation and Optimization Framework for Role-Playing Agents

Role-playing with large language models is fundamentally a session-level task, requiring agents to sustain character identity and interaction quality across extended multi-turn conversations. Yet existing evaluation and optimization methods remain largely turn-level, failing to capture long-horizon quality. We propose DynSess, a unified session-level framework for role-playing agents. DynSess-Eval scores complete dialogue sessions via rubrics targeting long-horizon behaviors. Leveraging its session-level rewards, we construct high-quality training trajectories through multi-turn lookahead search and train DynSess-Character with two complementary variants: DSPO (off-policy) and GSRPO (on-policy). Experiments show that DynSess-Eval aligns with human judgments substantially better than prior evaluators, and blind human evaluation further shows that DynSess-Character matches the strongest character model despite using substantially fewer parameters, while maintaining strong role consistency and interactive ability. Our dataset and code will be released to facilitate future research.

cs.CL↗

arXiv2026

From Facts to Insights: A Persona-Driven Dual Memory Framework and Dataset for Role-Playing Agents

While role-playing agents excel in short-term interactions, long-term conversations overwhelm context windows, motivating external memory frameworks. Current systems typically rely on persona-agnostic summarization, which records facts without persona-specific interpretation, yielding generic responses that compromise persona fidelity. To bridge this gap, we introduce RoleMemo, a dataset featuring four reasoning tasks where the factual fragments must be interpreted through the persona to reach the correct answer. Evaluation on RoleMemo exposes critical limitations of persona-agnostic frameworks. We thus propose DualMem, which decouples memory into two streams: factual cognition and persona-conditioned insight. Trained through Supervised Fine-Tuning (SFT) and Reinforcement Learning (RL), our framework with a 4B-parameter model outperforms zero-shot persona-agnostic frameworks powered by DeepSeek-V3.2 for sustained persona fidelity. Our resources are available at https://github.com/role2026/rolememo.

cs.CL↗

arXiv2026

Preferences Order, Ratings Anchor: From Fused Expert Aesthetic Ground Truth to Self-Distillation

Pairwise preferences and pointwise ratings are the two dominant annotation protocols in image aesthetic assessment (IAA), yet existing benchmarks adopt only one, leaving their complementarity unmeasured under controlled conditions. We introduce PPaint, a matched dual-protocol benchmark in which 15 domain experts, 5 per category, annotate 150 Chinese paintings under both protocols across five aesthetic dimensions, collecting 45,900 pairwise expert judgments through a locally dense preference design alongside the matched ratings. The matched design reveals complementary strengths: preferences yield more consistent ordinal rankings, while ratings anchor the absolute score scale. Fusing both signals via two independent preference-to-score methods yields a fused expert ground truth on which the two constructions converge to nearly identical scores. The same preference-to-score principle extends to label-free VLM training. PSDistill converts VLM pairwise judgments into calibrated pseudo-scores via an Elo reference pool, and trains the same VLM with confidence-weighted ranking optimization to produce a single-pass aesthetic scorer. Trained on a single painting category, the distilled Qwen3-VL-8B improves mean SRCC from 0.504 to 0.709 across all three categories, outperforming all open-source baselines including the dedicated aesthetic model ArtiMuse and matching closed-source Gemini-3.1-Pro within 0.04 SRCC at single-pass inference cost, with cross-domain transfer further validated on APDDv2. We will release the full PPaint dataset and training code.

cs.CV↗

arXiv2026

EvoTale: Continual Character Customization for Expanding Story Worlds

Character-centric story visualization aims to synthesize coherent image sequences that depict narrative events and interactions while preserving recurring character identities. In expanding story worlds, new user-specified characters must be continually incorporated despite varying customization difficulty and identity conflicts in multi-character scenes, without disrupting previously learned identities. In this paper, we propose EvoTale, a continual character customization framework for expanding stylized story worlds. We first introduce an All-in-One-World Character Integrator, which accumulates character-specific residual components within a unified LoRA branch using sparsely overlapping subspaces spanned by a shared orthonormal basis, while freezing previously learned components to limit cross-character coupling. We then develop a Character Quality Gate that uses rubric-guided MLLM feedback as a bounded controller to adapt the optimization budget based on the assessed customization quality. Finally, we propose Character-Aware Region-Focus Sampling, which combines bounding-box-guided regional denoising with identity-aware global denoising to preserve character identities within their designated regions while maintaining global narrative coherence. Experimental results show that EvoTale achieves a favorable balance across character fidelity, continual identity retention, multi-character generation quality, and story-text alignment compared with representative story visualization and customization methods.

cs.CV↗

arXiv2025

EvoCurr: Self-evolving Curriculum with Behavior Code Generation for Complex Decision-making

Complex decision-making often requires agents to progress through intermediate tasks rather than solve the final target directly. Existing LLM self-refinement methods typically iterate on a fixed target, while curriculum-learning methods often rely on hand-designed schedules, training-time optimization, or domain-specific difficulty metrics. To smooth out the learning curve with adaptive curriculum desgin, we introduce EvoCurr, a general inference-time framework that co-evolves curricula and executable policies. A Designer proposes verifiable intermediate tasks from the latest accepted progress and recent failures, while a Solver generates or trains policies for these tasks. A task-policy pair is accepted only when it passes hard feasibility checks and reaches a specified performance threshold meanwhile an accepted-floor constraint preserves the latest verified checkpoint and prevents failed harder attempts from overwriting mastered behavior. We evaluate EvoCurr across code-as-policy and closed-loop MARL settings on StarCraft II super-late-game micromanagement, hardest stress test, where EvoCurr achieves 96.7--99.2% macro-average win rates across three LLM backbones without LLM parameter training. The performance improvments on other experiments, Flatland, POGEMA, Overcooked, and LiveCodeBench Hard, reflect the generality of our EvoCurr.

cs.AI↗

arXiv2025

Empowering Economic Simulation for Massively Multiplayer Online Games through Generative Agent-Based Modeling

Within the domain of Massively Multiplayer Online (MMO) economy research, Agent-Based Modeling (ABM) has emerged as a robust tool for analyzing game economics, evolving from rule-based agents to decision-making agents enhanced by reinforcement learning. Nevertheless, existing works encounter significant challenges when attempting to emulate human-like economic activities among agents, particularly regarding agent reliability, sociability, and interpretability. In this study, we take a preliminary step in introducing a novel approach using Large Language Models (LLMs) in MMO economy simulation. Leveraging LLMs' role-playing proficiency, generative capacity, and reasoning aptitude, we design LLM-driven agents with human-like decision-making and adaptability. These agents are equipped with the abilities of role-playing, perception, memory, and reasoning, addressing the aforementioned challenges effectively. Simulation experiments focusing on in-game economic activities demonstrate that LLM-empowered agents can promote emergent phenomena like role specialization and price fluctuations in line with market rules.

cs.AI↗

arXiv2025

The Effects of Data Augmentation on Confidence Estimation for LLMs

Confidence estimation is crucial for reflecting the reliability of large language models (LLMs), particularly in the widely used closed-source models. Utilizing data augmentation for confidence estimation is viable, but discussions focus on specific augmentation techniques, limiting its potential. We study the impact of different data augmentation methods on confidence estimation. Our findings indicate that data augmentation strategies can achieve better performance and mitigate the impact of overconfidence. We investigate the influential factors related to this and discover that, while preserving semantic information, greater data diversity enhances the effectiveness of augmentation. Furthermore, the impact of different augmentation strategies varies across different range of application. Considering parameter transferability and usability, the random combination of augmentations is a promising choice.

cs.LG↗

arXiv2025

{\Psi}-Arena: Interactive Assessment and Optimization of LLM-based Psychological Counselors with Tripartite Feedback

Large language models (LLMs) have shown promise in providing scalable mental health support, while evaluating their counseling capability remains crucial to ensure both efficacy and safety. Existing evaluations are limited by the static assessment that focuses on knowledge tests, the single perspective that centers on user experience, and the open-loop framework that lacks actionable feedback. To address these issues, we propose {\Psi}-Arena, an interactive framework for comprehensive assessment and optimization of LLM-based counselors, featuring three key characteristics: (1) Realistic arena interactions that simulate real-world counseling through multi-stage dialogues with psychologically profiled NPC clients, (2) Tripartite evaluation that integrates assessments from the client, counselor, and supervisor perspectives, and (3) Closed-loop optimization that iteratively improves LLM counselors using diagnostic feedback. Experiments across eight state-of-the-art LLMs show significant performance variations in different real-world scenarios and evaluation perspectives. Moreover, reflection-based optimization results in up to a 141% improvement in counseling performance. We hope PsychoArena provides a foundational resource for advancing reliable and human-aligned LLM applications in mental healthcare.

cs.CL↗

arXiv2025

Crisp: Cognitive Restructuring of Negative Thoughts through Multi-turn Supportive Dialogues

Cognitive Restructuring (CR) is a psychotherapeutic process aimed at identifying and restructuring an individual's negative thoughts, arising from mental health challenges, into more helpful and positive ones via multi-turn dialogues. Clinician shortage and stigma urge the development of human-LLM interactive psychotherapy for CR. Yet, existing efforts implement CR via simple text rewriting, fixed-pattern dialogues, or a one-shot CR workflow, failing to align with the psychotherapeutic process for effective CR. To address this gap, we propose CRDial, a novel framework for CR, which creates multi-turn dialogues with specifically designed identification and restructuring stages of negative thoughts, integrates sentence-level supportive conversation strategies, and adopts a multi-channel loop mechanism to enable iterative CR. With CRDial, we distill Crisp, a large-scale and high-quality bilingual dialogue dataset, from LLM. We then train Crispers, Crisp-based conversational LLMs for CR, at 7B and 14B scales. Extensive human studies show the superiority of Crispers in pointwise, pairwise, and intervention evaluations.

cs.CL↗

arXiv2024

CharacterBench: Benchmarking Character Customization of Large Language Models

Character-based dialogue (aka role-playing) enables users to freely customize characters for interaction, which often relies on LLMs, raising the need to evaluate LLMs' character customization capability. However, existing benchmarks fail to ensure a robust evaluation as they often only involve a single character category or evaluate limited dimensions. Moreover, the sparsity of character features in responses makes feature-focused generative evaluation both ineffective and inefficient. To address these issues, we propose CharacterBench, the largest bilingual generative benchmark, with 22,859 human-annotated samples covering 3,956 characters from 25 detailed character categories. We define 11 dimensions of 6 aspects, classified as sparse and dense dimensions based on whether character features evaluated by specific dimensions manifest in each response. We enable effective and efficient evaluation by crafting tailored queries for each dimension to induce characters' responses related to specific dimensions. Further, we develop CharacterJudge model for cost-effective and stable evaluations. Experiments show its superiority over SOTA automatic judges (e.g., GPT-4) and our benchmark's potential to optimize LLMs' character customization. Our repository is at https://github.com/thu-coai/CharacterBench.

cs.CL↗

arXiv2024

StoryWeaver: A Unified World Model for Knowledge-Enhanced Story Character Customization

Story visualization has gained increasing attention in artificial intelligence. However, existing methods still struggle with maintaining a balance between character identity preservation and text-semantics alignment, largely due to a lack of detailed semantic modeling of the story scene. To tackle this challenge, we propose a novel knowledge graph, namely Character Graph (\textbf{CG}), which comprehensively represents various story-related knowledge, including the characters, the attributes related to characters, and the relationship between characters. We then introduce StoryWeaver, an image generator that achieve Customization via Character Graph (\textbf{C-CG}), capable of consistent story visualization with rich text semantics. To further improve the multi-character generation performance, we incorporate knowledge-enhanced spatial guidance (\textbf{KE-SG}) into StoryWeaver to precisely inject character semantics into generation. To validate the effectiveness of our proposed method, extensive experiments are conducted using a new benchmark called TBC-Bench. The experiments confirm that our StoryWeaver excels not only in creating vivid visual story plots but also in accurately conveying character identities across various scenarios with considerable storage efficiency, \emph{e.g.}, achieving an average increase of +9.03\% DINO-I and +13.44\% CLIP-T. Furthermore, ablation experiments are conducted to verify the superiority of the proposed module. Codes and datasets are released at https://github.com/Aria-Zhangjl/StoryWeaver.

cs.CV↗

arXiv2024

Reinforcement Learning From Imperfect Corrective Actions And Proxy Rewards

In practice, reinforcement learning (RL) agents are often trained with a possibly imperfect proxy reward function, which may lead to a human-agent alignment issue (i.e., the learned policy either converges to non-optimal performance with low cumulative rewards, or achieves high cumulative rewards but in undesired manner). To tackle this issue, we consider a framework where a human labeler can provide additional feedback in the form of corrective actions, which expresses the labeler's action preferences although this feedback may possibly be imperfect as well. In this setting, to obtain a better-aligned policy guided by both learning signals, we propose a novel value-based deep RL algorithm called Iterative learning from Corrective actions and Proxy rewards (ICoPro), which cycles through three phases: (1) Solicit sparse corrective actions from a human labeler on the agent's demonstrated trajectories; (2) Incorporate these corrective actions into the Q-function using a margin loss to enforce adherence to labeler's preferences; (3) Train the agent with standard RL losses regularized with a margin loss to learn from proxy rewards and propagate the Q-values learned from human feedback. Moreover, another novel design in our approach is to integrate pseudo-labels from the target Q-network to reduce human labor and further stabilize training. We experimentally validate our proposition on a variety of tasks (Atari games and autonomous driving on highway). On the one hand, using proxy rewards with different levels of imperfection, our method can better align with human preferences and is more sample-efficient than baseline methods. On the other hand, facing corrective actions with different types of imperfection, our method can overcome the non-optimality of this feedback thanks to the guidance from proxy reward.

cs.LG↗

arXiv2024

StyleTalk++: A Unified Framework for Controlling the Speaking Styles of Talking Heads

Individuals have unique facial expression and head pose styles that reflect their personalized speaking styles. Existing one-shot talking head methods cannot capture such personalized characteristics and therefore fail to produce diverse speaking styles in the final videos. To address this challenge, we propose a one-shot style-controllable talking face generation method that can obtain speaking styles from reference speaking videos and drive the one-shot portrait to speak with the reference speaking styles and another piece of audio. Our method aims to synthesize the style-controllable coefficients of a 3D Morphable Model (3DMM), including facial expressions and head movements, in a unified framework. Specifically, the proposed framework first leverages a style encoder to extract the desired speaking styles from the reference videos and transform them into style codes. Then, the framework uses a style-aware decoder to synthesize the coefficients of 3DMM from the audio input and style codes. During decoding, our framework adopts a two-branch architecture, which generates the stylized facial expression coefficients and stylized head movement coefficients, respectively. After obtaining the coefficients of 3DMM, an image renderer renders the expression coefficients into a specific person's talking-head video. Extensive experiments demonstrate that our method generates visually authentic talking head videos with diverse speaking styles from only one portrait image and an audio clip.

cs.CV↗

arXiv2024

TalkCLIP: Talking Head Generation with Text-Guided Expressive Speaking Styles

Audio-driven talking head generation has drawn growing attention. To produce talking head videos with desired facial expressions, previous methods rely on extra reference videos to provide expression information, which may be difficult to find and hence limits their usage. In this work, we propose TalkCLIP, a framework that can generate talking heads where the expressions are specified by natural language, hence allowing for specifying expressions more conveniently. To model the mapping from text to expressions, we first construct a text-video paired talking head dataset where each video has diverse text descriptions that depict both coarse-grained emotions and fine-grained facial movements. Leveraging the proposed dataset, we introduce a CLIP-based style encoder that projects natural language-based descriptions to the representations of expressions. TalkCLIP can even infer expressions for descriptions unseen during training. TalkCLIP can also use text to modulate expression intensity and edit expressions. Extensive experiments demonstrate that TalkCLIP achieves the advanced capability of generating photo-realistic talking heads with vivid facial expressions guided by text descriptions.

cs.CV↗

arXiv2024

Bayesian Design Principles for Offline-to-Online Reinforcement Learning

Offline reinforcement learning (RL) is crucial for real-world applications where exploration can be costly or unsafe. However, offline learned policies are often suboptimal, and further online fine-tuning is required. In this paper, we tackle the fundamental dilemma of offline-to-online fine-tuning: if the agent remains pessimistic, it may fail to learn a better policy, while if it becomes optimistic directly, performance may suffer from a sudden drop. We show that Bayesian design principles are crucial in solving such a dilemma. Instead of adopting optimistic or pessimistic policies, the agent should act in a way that matches its belief in optimal policies. Such a probability-matching agent can avoid a sudden performance drop while still being guaranteed to find the optimal policy. Based on our theoretical findings, we introduce a novel algorithm that outperforms existing methods on various benchmarks, demonstrating the efficacy of our approach. Overall, the proposed approach provides a new perspective on offline-to-online RL that has the potential to enable more effective learning from offline data.

cs.LG↗

arXiv2024

vMFER: Von Mises-Fisher Experience Resampling Based on Uncertainty of Gradient Directions for Policy Improvement

Reinforcement Learning (RL) is a widely employed technique in decision-making problems, encompassing two fundamental operations -- policy evaluation and policy improvement. Enhancing learning efficiency remains a key challenge in RL, with many efforts focused on using ensemble critics to boost policy evaluation efficiency. However, when using multiple critics, the actor in the policy improvement process can obtain different gradients. Previous studies have combined these gradients without considering their disagreements. Therefore, optimizing the policy improvement process is crucial to enhance learning efficiency. This study focuses on investigating the impact of gradient disagreements caused by ensemble critics on policy improvement. We introduce the concept of uncertainty of gradient directions as a means to measure the disagreement among gradients utilized in the policy improvement process. Through measuring the disagreement among gradients, we find that transitions with lower uncertainty of gradient directions are more reliable in the policy improvement process. Building on this analysis, we propose a method called von Mises-Fisher Experience Resampling (vMFER), which optimizes the policy improvement process by resampling transitions and assigning higher confidence to transitions with lower uncertainty of gradient directions. Our experiments demonstrate that vMFER significantly outperforms the benchmark and is particularly well-suited for ensemble structures in RL.

cs.LG↗

arXiv2024

Preconditioned Nonlinear Conjugate Gradient Method for Real-time Interior-point Hyperelasticity

The linear conjugate gradient method is widely used in physical simulation, particularly for solving large-scale linear systems derived from Newton's method. The nonlinear conjugate gradient method generalizes the conjugate gradient method to nonlinear optimization, which is extensively utilized in solving practical large-scale unconstrained optimization problems. However, it is rarely discussed in physical simulation due to the requirement of multiple vector-vector dot products. Fortunately, with the advancement of GPU-parallel acceleration techniques, it is no longer a bottleneck. In this paper, we propose a Jacobi preconditioned nonlinear conjugate gradient method for elastic deformation using interior-point methods. Our method is straightforward, GPU-parallelizable, and exhibits fast convergence and robustness against large time steps. The employment of the barrier function in interior-point methods necessitates continuous collision detection per iteration to obtain a penetration-free step size, which is computationally expensive and challenging to parallelize on GPUs. To address this issue, we introduce a line search strategy that deduces an appropriate step size in a single pass, eliminating the need for additional collision detection. Furthermore, we simplify and accelerate the computations of Jacobi preconditioning and Hessian-vector product for hyperelasticity and barrier function. Our method can accurately simulate objects comprising over 100,000 tetrahedra in complex self-collision scenarios at real-time speeds.

math.OC↗

arXiv2024

Towards a Simultaneous and Granular Identity-Expression Control in Personalized Face Generation

In human-centric content generation, the pre-trained text-to-image models struggle to produce user-wanted portrait images, which retain the identity of individuals while exhibiting diverse expressions. This paper introduces our efforts towards personalized face generation. To this end, we propose a novel multi-modal face generation framework, capable of simultaneous identity-expression control and more fine-grained expression synthesis. Our expression control is so sophisticated that it can be specialized by the fine-grained emotional vocabulary. We devise a novel diffusion model that can undertake the task of simultaneously face swapping and reenactment. Due to the entanglement of identity and expression, it's nontrivial to separately and precisely control them in one framework, thus has not been explored yet. To overcome this, we propose several innovative designs in the conditional diffusion model, including balancing identity and expression encoder, improved midpoint sampling, and explicitly background conditioning. Extensive experiments have demonstrated the controllability and scalability of the proposed framework, in comparison with state-of-the-art text-to-image, face swapping, and face reenactment methods.

cs.CV↗