arXiv subjects

Yexin Li

Publications and source records attributed to Yexin Li.

At least 19 recordsLinked to original sources

arXiv2026

SAGA: Scene-Aware, Goal-Evolving Agents for Long-Horizon Strategy Game Planning

Grand-strategy games such as Civilization pose a distinctive long-horizon planning problem: an agent must divide one shared resource pool among six competing domains -- technology, government, diplomacy, city development, expansion, and military -- under partial observability, with no feedback except a delayed final score. Current LLM agents fall short in three ways: 1) they cannot infer spatial relations from raw coordinates; 2) they allocate resources poorly, because feeding the entire growing state into one prompt and planning all domains in a single output diffuses attention and biases decisions toward urgent events; and 3) they cannot improve, as the delayed score gives no signal within or across games. We present SAGA, an LLM multi-agent framework pairing one mechanism with each weakness: (i) a Map-Semantic Scene Graph turning coordinates into per-entity statements of distance, direction, and threat; (ii) a Tool-Augmented Planner that retrieves only the state a decision needs, cutting the order of magnitude of its input, and issues a separate plan per domain to six specialist controllers, so urgent events do not derail long-term plans; and (iii) a Dual-Horizon Feedback Loop setting short-term goals during play and distilling each game into lessons for the next. On CivRealm, a Civilization-style benchmark, SAGA leads five LLM baselines on mean final score and is the only method significantly ahead of all of them on city development, the first investment baselines sacrifice, with 27% fewer output tokens; with cross-game learning it scores highest after five games, and its fifth game consistently surpasses its first across four maps. Our code is available at https://github.com/Kazecloudk/SAGA-Scene-Aware-Goal-Evolving-Agents-for-Long-Horizon-Strategy-Game-Planning.

cs.AI↗

arXiv2026

Black-Box Continual Learning for Vision-Language Models

The rapid deployment of Vision-Language Models (VLMs) in dynamic environments necessitates the ability to learn continuously without forgetting. However, traditional continual learning (CL) settings often rely on white-box paradigms, which is increasingly invalidated by the shift toward cloud-hosted models. In this paper, we introduce Black-CL, a more realistic benchmark for VLMs that enforces three primary real-world challenges: weight and architecture inaccessibility, constrained computation, and task-agnostic inference. The learner can query only output embeddings or logits, with no gradient flow through or structural modification of the backbone. Current CL methodologies, which rely on backbone backpropagation or complex parameter expansion, are fundamentally incompatible with these constraints. Under this setting, we propose BETA, a simple yet effective baseline built on the key insight that solely optimizing textual prototypes can navigate the complexities of CL. BETA integrates three core components: Semantic Projection Accumulation (SPA) for incremental knowledge acquisition, Latent Distribution Replay (LDR) for anchoring the embedding space against catastrophic forgetting, and Test-Time Prototype Adaptation (TTPA) for dynamic, instance-aware boundary refinement. Extensive experiments across ten diverse datasets and various backbones demonstrate that BETA significantly outperforms existing black-box tuners. Remarkably, with only 0.05 M trainable parameters, a 180--3000$\times$ reduction compared to competitive methods, BETA achieves performance on par with or even exceeding white-box CL methods. We believe Black-CL and BETA provide a foundational framework for future advancements in continual learning and accelerates the transition of continual learning from academia to real-world systems.

cs.CV↗

arXiv2026

Hierarchical Multiagent Reinforcement Learning for Multi-Group Tax Game

Reinforcement learning has increasingly been applied to economic decision-making, including taxation, public spending, and labor supply. However, existing RL-based economic models typically consider only a single government-household group, overlooking strategic interactions among competing governments. To address this limitation, we formulate taxation as a hierarchical multi-group game. Within each group, the government and households form a leader--follower game, while governments compete across groups through strategic fiscal policies. This coupled structure is difficult to solve using standard multi-agent reinforcement learning (MARL) methods. We therefore propose a bilevel MARL framework with \textit{Curriculum Learning} and a \textit{Closed-Loop Sequential Update} mechanism to improve training stability and convergence. We instantiate the framework in a taxation simulation environment grounded in classical economic models, supporting the evaluation of taxation policies under inter-group competition. Experiments show that the proposed method learns stable and sustainable tax policies. Compared with a two-group baseline without the proposed mechanisms, our approach avoids premature game collapse, extends the effective game duration by 60.92\%, and reduces GDP disparities among governments by 44.12\%.

cs.MA↗

arXiv2026

Large Language Models Explore by Latent Distilling

Generating diverse responses is crucial for test-time scaling of large language models (LLMs), yet standard stochastic sampling mostly yields surface-level lexical variation, limiting semantic exploration. In this paper, we propose Exploratory Sampling (ESamp), a decoding approach that explicitly encourages semantic diversity during generation. ESamp is motivated by the well-known observation that neural networks tend to make lower-error predictions on inputs similar to those encountered before, and incur higher prediction error on novel ones. Building on this property, we train a lightweight Distiller at test time to predict deep-layer hidden representations of the LLM from its shallow-layer representations to model the LLM's depth-wise representation transitions. During decoding, the Distiller continuously adapts to the mappings induced by the current generation context. ESamp uses the prediction error as a novelty signal to reweight candidate token extensions conditioned on the current prefix, thereby biasing decoding toward less-explored semantic patterns. ESamp is implemented with an asynchronous training--inference pipeline, with less than 5% worst case overhead (1.2% in the optimized release). Empirical results show that ESamp significantly boosts the Pass@k efficiency of reasoning models, showing superior or comparable performance to strong stochastic and heuristic baselines. Notably, ESamp achieves robust generalization across mathematics, science, and code generation benchmarks and breaks the trade-off between diversity and coherence in creative writing. Our code has released at: https://github.com/LinesHogan/tLLM.

cs.CL↗

arXiv2026

IDER: IDempotent Experience Replay for Reliable Continual Learning

Catastrophic forgetting, the tendency of neural networks to forget previously learned knowledge when learning new tasks, has been a major challenge in continual learning (CL). To tackle this challenge, CL methods have been proposed and shown to reduce forgetting. Furthermore, CL models deployed in mission-critical settings can benefit from uncertainty awareness by calibrating their predictions to reliably assess their confidences. However, existing uncertainty-aware continual learning methods suffer from high computational overhead and incompatibility with mainstream replay methods. To address this, we propose idempotent experience replay (IDER), a novel approach based on the idempotent property where repeated function applications yield the same output. Specifically, we first adapt the training loss to make model idempotent on current data streams. In addition, we introduce an idempotence distillation loss. We feed the output of the current model back into the old checkpoint and then minimize the distance between this reprocessed output and the original output of the current model. This yields a simple and effective new baseline for building reliable continual learners, which can be seamlessly integrated with other CL approaches. Extensive experiments on different CL benchmarks demonstrate that IDER consistently improves prediction reliability while simultaneously boosting accuracy and reducing forgetting. Our results suggest the potential of idempotence as a promising principle for deploying efficient and trustworthy continual learning systems in real-world applications.Our code is available at https://github.com/YutingLi0606/Idempotent-Continual-Learning.

cs.LG↗

arXiv2026

Hierarchical Reinforcement Learning for Cooperative Air-Ground Delivery in Urban System

Cooperative air-ground delivery has emerged as a promising logistics paradigm by leveraging the complementary strengths of UAVs and ground carriers. However, effective dispatching in such heterogeneous systems faces two critical challenges: i) the heterogeneity between flight and road dynamics, ii) the scalability bottleneck raised by the exponential decision variables in large-scale fleets. To address these challenges, we propose HRL4AG, a Hierarchical Reinforcement Learning framework for cooperative Air-Ground delivery. Specifically, HRL4AG employs a high-level manager to tackle the scalability bottleneck by decomposing the joint action space, and mode-specific workers that encode distinct flight and road dynamics to address the heterogeneity. Furthermore, a novel internal reward mechanism is designed to guide the hierarchical policy learning, addressing the credit assignment problem in sparse-reward settings. Extensive experiments on two real-world datasets and an evaluation platform demonstrate that HRL4AG significantly outperforms state-of-the-art baselines, improving the delivery success rate by up to 26% while achieving an 80-fold increase in computational efficiency.

cs.CY↗

arXiv2026

Adaptive Value Decomposition: Coordinating a Varying Number of Agents in Urban Systems

Multi-agent reinforcement learning (MARL) provides a promising paradigm for coordinating multi-agent systems (MAS). However, most existing methods rely on restrictive assumptions, such as a fixed number of agents and fully synchronous action execution. These assumptions are often violated in urban systems, where the number of active agents varies over time, and actions may have heterogeneous durations, resulting in a semi-MARL setting. Moreover, while sharing policy parameters among agents is commonly adopted to improve learning efficiency, it can lead to highly homogeneous actions when a subset of agents make decisions concurrently under similar observations, potentially degrading coordination quality. To address these challenges, we propose Adaptive Value Decomposition (AVD), a cooperative MARL framework that adapts to a dynamically changing agent population. AVD further incorporates a lightweight mechanism to mitigate action homogenization induced by shared policies, thereby encouraging behavioral diversity and maintaining effective cooperation among agents. In addition, we design a training-execution strategy tailored to the semi-MARL setting that accommodates asynchronous decision-making when some agents act at different times. Experiments on real-world bike-sharing redistribution tasks in two major cities, London and Washington, D.C., demonstrate that AVD outperforms state-of-the-art baselines, confirming its effectiveness and generalizability.

cs.MA↗

arXiv2026

Spectral Disentanglement and Enhancement: A Dual-domain Contrastive Framework for Representation Learning

Large-scale multimodal contrastive learning has recently achieved impressive success in learning rich and transferable representations, yet it remains fundamentally limited by the uniform treatment of feature dimensions and the neglect of the intrinsic spectral structure of the learned features. Empirical evidence indicates that high-dimensional embeddings tend to collapse into narrow cones, concentrating task-relevant semantics in a small subspace, while the majority of dimensions remain occupied by noise and spurious correlations. Such spectral imbalance and entanglement undermine model generalization. We propose Spectral Disentanglement and Enhancement (SDE), a novel framework that bridges the gap between the geometry of the embedded spaces and their spectral properties. Our approach leverages singular value decomposition to adaptively partition feature dimensions into strong signals that capture task-critical semantics, weak signals that reflect ancillary correlations, and noise representing irrelevant perturbations. A curriculum-based spectral enhancement strategy is then applied, selectively amplifying informative components with theoretical guarantees on training stability. Building upon the enhanced features, we further introduce a dual-domain contrastive loss that jointly optimizes alignment in both the feature and spectral spaces, effectively integrating spectral regularization into the training process and encouraging richer, more robust representations. Extensive experiments on large-scale multimodal benchmarks demonstrate that SDE consistently improves representation robustness and generalization, outperforming state-of-the-art methods. SDE integrates seamlessly with existing contrastive pipelines, offering an effective solution for multimodal representation learning.

cs.LG↗

arXiv2026

Grad2Reward: From Sparse Judgment to Dense Rewards for Improving Open-Ended LLM Reasoning

Reinforcement Learning with Verifiable Rewards (RLVR) has catalyzed significant breakthroughs in complex LLM reasoning within verifiable domains, such as mathematics and programming. Recent efforts have sought to extend this paradigm to open-ended tasks by employing LLMs-as-a-Judge to provide sequence-level rewards for policy optimization. However, these rewards are inherently sparse, failing to provide the fine-grained supervision necessary for generating complex, long-form trajectories. Furthermore, current work treats the Judge as a black-box oracle, discarding the rich intermediate feedback signals encoded in it. To address these limitations, we introduce Grad2Reward, a novel framework that extracts dense process rewards directly from the Judge's model inference process via a single backward pass. By leveraging gradient-based attribution, Grad2Reward enables precise token-level credit assignment, substantially enhancing training efficiency and reasoning quality. Additionally, Grad2Reward introduces a self-judging mechanism, allowing the policy to improve through its own evaluative signals without training specialized reward models or reliance on superior external Judges. The experiments demonstrate that policies optimized with Grad2Reward achieve outstanding performance across diverse open-ended tasks, affirming its effectiveness and broad generalizability.

cs.LG↗

arXiv2025

Linking Process to Outcome: Conditional Reward Modeling for LLM Reasoning

Process Reward Models (PRMs) have emerged as a promising approach to enhance the reasoning capabilities of large language models (LLMs) by guiding their step-by-step reasoning toward a final answer. However, existing PRMs either treat each reasoning step in isolation, failing to capture inter-step dependencies, or struggle to align process rewards with the final outcome. Consequently, the reward signal fails to respect temporal causality in sequential reasoning and faces ambiguous credit assignment. These limitations make downstream models vulnerable to reward hacking and lead to suboptimal performance. In this work, we propose Conditional Reward Modeling (CRM) that frames LLM reasoning as a temporal process leading to a correct answer. The reward of each reasoning step is not only conditioned on the preceding steps but also explicitly linked to the final outcome of the reasoning trajectory. By enforcing conditional probability rules, our design captures the causal relationships among reasoning steps, with the link to the outcome allowing precise attribution of each intermediate step, thereby resolving credit assignment ambiguity. Further, through this consistent probabilistic modeling, the rewards produced by CRM enable more reliable cross-sample comparison. Experiments across Best-of-N sampling, beam search and reinforcement learning demonstrate that CRM consistently outperforms existing reward models, offering a principled framework for enhancing LLM reasoning. In particular, CRM is more robust to reward hacking and delivers stable downstream improvements without relying on verifiable rewards derived from ground truth.

cs.LG↗

arXiv2025

A Unified Multi-Agent Framework for Universal Multimodal Understanding and Generation

Real-world multimodal applications often require any-to-any capabilities, enabling both understanding and generation across modalities including text, image, audio, and video. However, integrating the strengths of autoregressive language models (LLMs) for reasoning and diffusion models for high-fidelity generation remains challenging. Existing approaches rely on rigid pipelines or tightly coupled architectures, limiting flexibility and scalability. We propose MAGUS (Multi-Agent Guided Unified Multimodal System), a modular framework that unifies multimodal understanding and generation via two decoupled phases: Cognition and Deliberation. MAGUS enables symbolic multi-agent collaboration within a shared textual workspace. In the Cognition phase, three role-conditioned multimodal LLM agents - Perceiver, Planner, and Reflector - engage in collaborative dialogue to perform structured understanding and planning. The Deliberation phase incorporates a Growth-Aware Search mechanism that orchestrates LLM-based reasoning and diffusion-based generation in a mutually reinforcing manner. MAGUS supports plug-and-play extensibility, scalable any-to-any modality conversion, and semantic alignment - all without the need for joint training. Experiments across multiple benchmarks, including image, video, and audio generation, as well as cross-modal instruction following, demonstrate that MAGUS outperforms strong baselines and state-of-the-art systems. Notably, on the MME benchmark, MAGUS surpasses the powerful closed-source model GPT-4o.

cs.LG↗

arXiv2025

ECMF: Enhanced Cross-Modal Fusion for Multimodal Emotion Recognition in MER-SEMI Challenge

Emotion recognition plays a vital role in enhancing human-computer interaction. In this study, we tackle the MER-SEMI challenge of the MER2025 competition by proposing a novel multimodal emotion recognition framework. To address the issue of data scarcity, we leverage large-scale pre-trained models to extract informative features from visual, audio, and textual modalities. Specifically, for the visual modality, we design a dual-branch visual encoder that captures both global frame-level features and localized facial representations. For the textual modality, we introduce a context-enriched method that employs large language models to enrich emotional cues within the input text. To effectively integrate these multimodal features, we propose a fusion strategy comprising two key components, i.e., self-attention mechanisms for dynamic modality weighting, and residual connections to preserve original representations. Beyond architectural design, we further refine noisy labels in the training set by a multi-source labeling strategy. Our approach achieves a substantial performance improvement over the official baseline on the MER2025-SEMI dataset, attaining a weighted F-score of 87.49% compared to 78.63%, thereby validating the effectiveness of the proposed framework.

cs.CV↗

arXiv2025

Learning to Select In-Context Demonstration Preferred by Large Language Model

In-context learning (ICL) enables large language models (LLMs) to adapt to new tasks during inference using only a few demonstrations. However, ICL performance is highly dependent on the selection of these demonstrations. Recent work explores retrieval-based methods for selecting query-specific demonstrations, but these approaches often rely on surrogate objectives such as metric learning, failing to directly optimize ICL performance. Consequently, they struggle to identify truly beneficial demonstrations. Moreover, their discriminative retrieval paradigm is ineffective when the candidate pool lacks sufficient high-quality demonstrations. To address these challenges, we propose GenICL, a novel generative preference learning framework that leverages LLM feedback to directly optimize demonstration selection for ICL. Experiments on 19 datasets across 11 task categories demonstrate that GenICL achieves superior performance than existing methods in selecting the most effective demonstrations, leading to better ICL performance.

cs.LG↗

arXiv2025

CAE: Repurposing the Critic as an Explorer in Deep Reinforcement Learning

Exploration remains a fundamental challenge in reinforcement learning, as many existing methods either lack theoretical guarantees or fall short in practical effectiveness. In this paper, we propose CAE, i.e., the Critic as an Explorer, a lightweight approach that repurposes the value networks in standard deep RL algorithms to drive exploration, without introducing additional parameters. CAE leverages multi-armed bandit techniques combined with a tailored scaling strategy, enabling efficient exploration with provable sub-linear regret bounds and strong empirical stability. Remarkably, it is simple to implement, requiring only about 10 lines of code. For complex tasks where learning reliable value networks is difficult, we introduce CAE+, an extension of CAE that incorporates an auxiliary network. CAE+ increases the parameter count by less than 1% while preserving implementation simplicity, adding roughly 10 additional lines of code. Extensive experiments on MuJoCo, MiniHack, and Habitat validate the effectiveness of CAE and CAE+, highlighting their ability to unify theoretical rigor with practical efficiency.

stat.ML↗

arXiv2025

Enhanced Continual Learning of Vision-Language Models with Model Fusion

Vision-Language Models (VLMs) represent a significant breakthrough in artificial intelligence by integrating visual and textual modalities to achieve impressive zero-shot capabilities. However, VLMs are susceptible to catastrophic forgetting when sequentially fine-tuned on multiple downstream tasks. Existing continual learning methods for VLMs face various limitations, often relying on additional reference datasets, compromising zero-shot performance, or being restricted to parameter-efficient fine-tuning scenarios. In this paper, we propose a novel Continual Decoupling-Unifying (ConDU) approach that pioneers the use of model fusion for continual learning in VLMs. Specifically, ConDU maintains a unified model along with task triggers and prototype sets, employing an iterative process of decoupling task experts for previous tasks and unifying them with the task expert for the newly learned task. Additionally, we introduce an inference strategy for zero-shot scenarios by aggregating predictions from multiple decoupled task experts. Extensive experiments on the MTIL benchmark show that ConDU achieves up to a 2\% improvement in average performance across all seen tasks compared to state-of-the-art baselines, while also enhancing zero-shot capabilities relative to the original VLM. Our code is available at https://github.com/zhangzicong518/ConDU.

cs.CV↗

arXiv2024

Are the Values of LLMs Structurally Aligned with Humans? A Causal Perspective

As large language models (LLMs) become increasingly integrated into critical applications, aligning their behavior with human values presents significant challenges. Current methods, such as Reinforcement Learning from Human Feedback (RLHF), typically focus on a limited set of coarse-grained values and are resource-intensive. Moreover, the correlations between these values remain implicit, leading to unclear explanations for value-steering outcomes. Our work argues that a latent causal value graph underlies the value dimensions of LLMs and that, despite alignment training, this structure remains significantly different from human value systems. We leverage these causal value graphs to guide two lightweight value-steering methods: role-based prompting and sparse autoencoder (SAE) steering, effectively mitigating unexpected side effects. Furthermore, SAE provides a more fine-grained approach to value steering. Experiments on Gemma-2B-IT and Llama3-8B-IT demonstrate the effectiveness and controllability of our methods.

cs.CL↗

arXiv2024

A Contextual Combinatorial Bandit Approach to Negotiation

Learning effective negotiation strategies poses two key challenges: the exploration-exploitation dilemma and dealing with large action spaces. However, there is an absence of learning-based approaches that effectively address these challenges in negotiation. This paper introduces a comprehensive formulation to tackle various negotiation problems. Our approach leverages contextual combinatorial multi-armed bandits, with the bandits resolving the exploration-exploitation dilemma, and the combinatorial nature handles large action spaces. Building upon this formulation, we introduce NegUCB, a novel method that also handles common issues such as partial observations and complex reward functions in negotiation. NegUCB is contextual and tailored for full-bandit feedback without constraints on the reward functions. Under mild assumptions, it ensures a sub-linear regret upper bound. Experiments conducted on three negotiation tasks demonstrate the superiority of our approach.

cs.AI↗

arXiv2024

A novel directly energy-preserving method for charged particle dynamics

In this paper, we apply the coordinate increment discrete gradient (CIDG) method to solve the Lorentz force system which can be written as a non-canonical Hamiltonian system. Then we can obtain a new energy-preserving CIDG-I method for the system. The CIDG-I method can combine with its adjoint method CIDG-II which is also a energy-preserving method to form a new method, namely CIDG-C method. The CIDG-C method is symmetrical and can conserve the Hamiltonian energy directly and exactly. With comparison to the well-used Boris method, numerical experiments indicate that the CIDG-C method holds advantage over the Boris method in terms of energy-conserving.

math.NA↗