Searcharxiv⌕ Search

arXiv subjects

Thuy-Trang Vu

Publications and source records attributed to Thuy-Trang Vu.

At least 19 recordsLinked to original sources

MATO: Multi-objective Personalized Alignment with Test-time Optimization for Large Language Models

Aligning large language models (LLMs) with diverse and multifaceted user preferences is a fundamental challenge in personalized AI systems. Existing multi-objective alignment methods either rely on costly training or require pre-trained reward models for each preference, making it difficult for them to adapt to evolving preferences. Prompt-based personalization offers a training-free alternative, but prompting alone often provides limited steerability, as LLMs may overemphasize or overlook certain preferences and fail to give users reliable control over the relative importance of different objectives when conflicts arise, leading to suboptimal alignment. In this paper, we introduce MATO, a training-free framework for Multi-objective personalized Alignment with Test-time Optimization. MATO formulates personalization as a test-time optimization problem that steers the relative importance of multiple objectives through controllable weights during decoding, without modifying model parameters or requiring external reward models. Specifically, a reward discovery module constructs prompt-derived surrogate rewards from the backbone LLM for diverse objectives specified in natural language, while a weight optimization module dynamically adjusts objective weights based on the user's initial preferences and the partially generated response to balance competing objectives during generation. The resulting surrogate rewards and weights jointly guide an online optimization procedure over the token distribution, enabling better alignment with the target objectives. Extensive experiments across multiple datasets and backbone LLMs show that MATO consistently outperforms strong baselines, achieving Pareto-improving multi-objective alignment and stronger steerability. These results highlight test-time optimization as a promising direction for scalable, controllable, and model-agnostic personalized alignment.

cs.CL↗

Anticipating Social Consequences with Distributional World Models

Social world models predict how interactions develop after an action, helping agents negotiate, cooperate and mediate. In these situations, one action can lead to several plausible futures, hence predicting only one future can overlook consequential alternatives, while exploring them through repeated simulations increases inference cost. We propose DSWM, a Distributional Social World Model that predicts alternative responses, their probabilities and immediate and final outcome values in one output per candidate action. We train a 4B model on repeated simulations using supervised fine-tuning and GRPO. At inference time, the actor uses these possible consequences to choose an action. Across three seen and two unseen environments, with interaction backbones absent from training collection, DSWM achieves the best average performance on 13 of 15 metrics. Ablations show that retaining multiple futures and their predicted probabilities improves outcomes in all five environments.

cs.AI↗

UniGuard: Multi-View Guardrail Learning for Cross-Interface Safety Generalization

Guardrail models are increasingly used as safety layers for LLMs and agentic systems, but most are trained on fixed prompt or response interfaces and become unreliable when the same safety case appears as a conversation, agent trajectory, multi-agent trace, or policy-conditioned input. We propose UniGuard, a training framework that learns interface-invariant safety representations from multiple semantically aligned views of the same safety case. We also construct the UniGuard-FT dataset by transforming 1,000 public safety cases into prompt, response, multi-turn conversation, agent-trace, and multi-agent-trace views, and train UniGuard-0.6B with classification, supervised contrastive learning, and asymmetric confidence calibration losses. Across unseen-interface benchmarks ATBench, PolicyGuardBench, and DynaBench, UniGuard-0.6B improves over its pretrained guardrail by 45.12 F1 points on average and reduces expected calibration error by 35.41 points, while remaining competitive with much larger guardrails such as Claude-Sonnet-4.5 and GPT-5.2 - despite being a 0.6B model and trained only with synthetic data.

cs.CL↗

HERD: Harness-Evolving Recursive Distillation for Continual Learning

Continual learning requires a model to keep acquiring new skills and knowledge without compromising its existing capabilities. Parametric methods suffer from catastrophic forgetting and loss of plasticity. Non-parametric methods, such as prompts, skills, and harnesses, leverage in-context learning but are limited by context length and lossy plain-text compression. To effectively combine their complementary strengths, we propose HERD (Harness-Evolving Recursive Distillation), a hybrid continual learning framework that learns efficiently in the harness and internalizes experience into cumulative capabilities via distillation. On the harness side, HERD continually refines its skills to capture reusable patterns, while retaining high-residual examples that the skills fail to capture in a bounded memory. On the parameter side, HERD performs drift-constrained distillation with dual anchors, substantially reducing forgetting over long recursive runs. On the 25-stage task stream of TRACE+17, under fixed memory and compute budgets, HERD matches the strongest parametric baseline in new-task acquisition while exhibiting the least forgetting and the strongest retention of general capabilities.

cs.CL↗

A Responsibility-Centered Coordination Mechanism for Decentralized Multi-Agent Systems

Centralized orchestration provides comprehensive multi-agent coordination but creates bottlenecks in scalability, adaptability, and collective performance. Decentralized approaches alleviate these bottlenecks by distributing coordination across agents, yet often leave responsibility for delegated dependencies and failure recovery unclear. To address this gap, we propose RCM, a responsibility-centered coordination mechanism that leverages commitments to couple task delegation with explicit responsibilities of the involved parties. Through recursive delegation, agents retain their upstream responsibilities while locally orchestrating downstream commitments, thereby preserving accountability across delegation chains, creating a network of local orchestration. We operationalize this mechanism locally through responsibility-constrained decision-making and a responsibility-oriented communication protocol (RCP), enabling agents to autonomously establish, manage, and terminate commitments. As agents act in parallel, their local actions continuously evolve the commitment graph, allowing global coordination to emerge and adapt toward a shared mission without centralized control. Experiments across reasoning, coding, and executable agent environments demonstrate that RCM outperforms strong centralized and decentralized coordination baselines by up to 18.7 points in task performance and 10.2 points in Responsibility, while improving failure and harm detection without task-specific training.

cs.MA↗

Stochastic Teacher Intervention for Agentic On-Policy Distillation

On-policy distillation (OPD) efficiently transfers capabilities from a stronger teacher to a student language model through dense token-level supervision on student-generated rollouts and has shown promise on complex tasks such as mathematical reasoning. However, in multi-turn agentic tasks, student decisions shape subsequent observations, causing early errors to accumulate across turns. The resulting trajectories can drift away from the teacher's rollout distribution, making the teacher's token-level supervision less reliable or even counterproductive for OPD training. To address this issue, we introduce STI-OPD, a stochastic teacher intervention framework for multi-turn agentic OPD. During multi-turn interaction, STI-OPD uses teacher intervention guided by teacher-student policy discrepancy to replace the student's proposed action with a teacher-generated one to maximize the acquisition of reliable supervision. We further develop a stochastic intervention strategy, addressing the limitations of previous threshold-based or fixed-schedule approaches, that estimates policy discrepancy using KL divergence and maps it to an intervention probability. By sampling whether to intervene from this probability, STI-OPD adaptively balances teacher control with student exploration. To learn from the resulting mixed-policy trajectories, we introduce an Importance-Weighted Reverse KL objective that corrects the token sampling mismatch between teacher-generated responses and the student policy to preserve the original OPD objective. Across tool-integrated reasoning and long-horizon interaction, STI-OPD outperforms the strongest prior OPD baseline on every evaluated benchmark and student size. Ablations further show that both discrepancy-guided intervention and importance weighting contribute to these gains.

cs.CL↗

AIPO: Learning to Reason from Active Interaction

Recent advances in LLMs have demonstrated strong reasoning capabilities, largely stimulated by RLVR. However, the exploration of existing RLVR algorithms remains largely constrained by the knowledge and reasoning strategies already accessible to the policy model. Although recent methods introduce external expert demonstrations to broaden exploration, they typically rely on complete trajectory-level guidance, which can be sample-inefficient, information-sparse, and insufficiently adaptive to intermediate reasoning bottlenecks. Inspired by collaborative multi-agent systems, we propose AIPO, an enhanced reinforcement learning framework that improves LLM reasoning through active multi-agent interaction during exploration. Specifically, when encountering reasoning bottlenecks, AIPO enables the policy model to proactively consult three functional collaborative agents, namely the verify agent, knowledge agent, and reasoning agent, thereby obtaining fine-grained and state-dependent guidance during rollout. The resulting mixed-policy trajectories expose the policy to reasoning directions that may be difficult to discover through isolated on-policy exploration. To learn effectively from collaborator-provided tokens, we further introduce a corrected importance sampling coefficient together with a lower-bound clipping strategy to mitigate off-policy discrepancy and vanishing gradients. After training, the policy model reasons independently without relying on collaborative agents. Extensive experiments across mathematical, scientific, coding, and puzzle reasoning benchmarks show that AIPO consistently improves reasoning performance and generalizes across different policy models, collaborator backbones, and RLVR algorithms.

cs.CL↗

E$^3$-Orch: Towards Effective, Efficient, and Extensible Agentic Orchestration with Reinforcement Learning

Agentic orchestration enables multiple autonomous agents to solve complex tasks through adaptive decomposition, delegation, and execution. However, existing orchestrators often rely on hand-crafted logic and prompting strategies, limiting adaptation and generalization across tasks and executor configurations. We propose E$^3$-Orch, a reinforcement learning framework for effective, efficient, and extensible agentic orchestration based on a milestone-plan-act workflow. Instead of planning all subtasks upfront, E$^3$-Orch organizes execution around milestones, a mid-level abstraction that scopes planning around meaningful intermediate objectives and allows orchestration decisions to adapt as execution progresses. For each milestone, the orchestrator builds a dependency-aware plan, assigns subtasks to suitable executors, and executes independent subtasks in parallel. We train the orchestration policy from execution feedback, using milestone and plan decisions as units for fine-grained credit assignment. Tree-structured rollouts compare alternative decisions under shared execution histories, while complementary rewards optimize task performance, execution cost, and planning completeness, including an uncertainty-aware performance reward for stochastic downstream outcomes. Across seven benchmarks, E$^3$-Orch achieves the best task performance under multiple executor configurations, improving over the strongest baselines by $0.7$--$3.8$ points and delivering $1.16$--$1.59\times$ higher intelligence efficiency. The learned policy also transfers to unseen executor configurations introduced only at evaluation time, supporting extensible agentic orchestration.

cs.CL↗

Efficient Temporal-aware Matryoshka Adaptation for Temporal Information Retrieval

Retrievers are a key bottleneck in temporal Retrieval-Augmented Generation (RAG) systems: failing to retrieve temporally relevant context can degrade downstream generation, regardless of LLM reasoning. We propose Temporal-aware Matryoshka Representation Learning (TMRL), an efficient method that equips retrievers with temporal-aware Matryoshka embeddings. TMRL leverages the nested structure of Matryoshka embeddings to introduce a temporal subspace, enhancing temporal encoding while preserving general semantic representations. Experiments show that TMRL efficiently adapts diverse standard encoder-only text embedding models, achieving competitive temporal retrieval and temporal RAG performance compared to existing Matryoshka-based adaptation and temporal retrieval methods, while enabling flexible accuracy-efficiency trade-offs.

cs.IR↗

MixLoRA-DSI: Dynamically Expandable Mixture-of-LoRA Experts for Rehearsal-Free Generative Retrieval over Dynamic Corpora

Continually updating model-based indexes in generative retrieval with new documents remains challenging, as full retraining is computationally expensive and impractical under resource constraints. We propose MixLoRA-DSI, a novel framework that combines an expandable mixture of Low-Rank Adaptation experts with a layer-wise out-of-distribution (OOD)-driven expansion strategy. Instead of allocating new experts for each new corpus, our proposed expansion strategy enables sublinear parameter growth by selectively introducing new experts only when significant number of OOD documents are detected. Experiments on NQ320k and MS MARCO Passage demonstrate that MixLoRA-DSI outperforms full-model update baselines, with minimal parameter overhead and substantially lower training costs.

cs.IR↗

The Social Cost of Intelligence: Emergence, Propagation, and Amplification of Stereotypical Bias in Multi-Agent Systems

Bias in large language models (LLMs) remains a persistent challenge, often leading to stereotyping and unfair treatment across social groups. While prior work has mainly focused on individual LLMs, the emergence of multi-agent systems (MAS), where multiple LLMs collaborate and communicate, introduces new and underexplored dynamics in how bias emerges, propagates, and amplifies. To systematically investigate these dynamics, we propose a simple evaluation framework with three agent-level metrics that quantify bias emergence, propagation, and amplification throughout multi-agent interaction. We evaluate MAS across three bias benchmarks under varying LLM backbones, social-group configurations, communication behaviors, and adversarial settings. Our results show that communication can trigger up to 70\% new bias emergence, propagate bias across over 80\% of agents, and amplify stereotypes by more than 3$\times$. We further find that denser and competitive communication generally increases bias. Finally, we demonstrate that MAS are highly vulnerable to simple bias injection attacks, and existing defense strategies provide only limited protection. Our findings provide important insights into the fairness and robustness of multi-agent LLM systems.

cs.MA↗

TriAlign: Towards Universal Truth Consistency in Personalized LLM Alignment

Personalized large language models adapt responses to users' preferences and social attributes, but can introduce substantial universal truth inconsistencies across social groups, where some groups systematically receive less accurate responses on objective tasks. Existing alignment methods either ignore personalization or mainly focus on subjective preference alignment, largely overlooking fairness and consistency in universal truths. To address this gap, we study Truth-Invariant Alignment (TIA), an alignment problem for personalized LLMs that aims to ensure universal truths remain consistent across social groups while preserving personalization. We propose TriAlign, the first offline multi-agent reinforcement learning (MARL) framework for TIA, where each social group is modeled as an agent interacting. TriAlign jointly optimizes universal truth accuracy, cross-group truth consistency, and personalization through a fairness-aware objective and an explicit inconsistency penalty. Experiments across diverse benchmarks demonstrate that TriAlign achieves a stronger balance among these three objectives than strong baselines, reducing universal truth disparities across social groups while improving both objective task performance and personalization quality.

cs.AI↗

Adaptive Subspace Projection for Generative Personalization

Generative personalization often suffers from the semantic collapsing problem (SCP), where a learned personalized concept overpowers the rest of the text prompt, causing the model to ignore important contextual details. To address this, we first analyze the underlying cause, revealing that the semantic drift responsible for SCP is not random but is concentrated within a specific low-dimensional subspace. We also discover that the personalization process perturbs the embedding of the original base concept, making it an unstable reference point. Based on these insights, we introduce Test-time Embedding Adjustment with Adaptive Subspace Projection (AdaptSP), a training-free method that uses the stable, pre-trained embedding as an anchor. AdaptSP isolates the semantic drift and projects it onto the identified subspace, performing a precise adjustment that mitigates SCP while maintaining the subject identity. Our experiments show that this targeted approach significantly improves prompt fidelity and contextual alignment.

cs.CV↗

G-reasoner: Foundation Models for Unified Reasoning over Graph-structured Knowledge

Large language models (LLMs) excel at complex reasoning but remain limited by static and incomplete parametric knowledge. Retrieval-augmented generation (RAG) mitigates this by incorporating external knowledge, yet existing RAGs struggle with knowledge-intensive tasks due to fragmented information and weak modeling of knowledge structure. Graphs offer a natural way to model relationships within knowledge, but LLMs are inherently unstructured and cannot effectively reason over graph-structured data. Recent graph-enhanced RAG (GraphRAG) attempts to bridge this gap by constructing tailored graphs and enabling LLMs to reason on them. However, these methods often depend on ad-hoc graph designs, heuristic search, or costly agent pipelines, which hinder scalability and generalization. To address these challenges, we present G-reasoner, a unified framework that integrates graph and language foundation models for scalable reasoning over diverse graph-structured knowledge. Central to our approach is QuadGraph, a standardized four-layer abstraction that unifies heterogeneous knowledge sources into a common graph representation. Building on this, we introduce a 34M-parameter graph foundation model (GFM) that jointly captures graph topology and textual semantics, and is integrated with LLMs to enhance reasoning in downstream applications. To ensure scalability and efficiency, mixed-precision training and distributed message-passing are implemented to scale GFM with more GPUs. Extensive experiments on six benchmarks show that G-reasoner consistently outperforms state-of-the-art baselines, significantly enhances LLM reasoning, and achieves strong efficiency and cross-graph generalization.

cs.AI↗

GTS: Inference-Time Scaling of Latent Reasoning with a Learnable Gaussian Thought Sampler

Inference-time scaling (ITS) in latent reasoning models typically relies on heuristic perturbations, such as dropout or fixed Gaussian noise, to generate diverse candidate trajectories. However, we show that stronger perturbations do not necessarily yield better sampling quality: they often induce larger distribution shifts without producing more useful reasoning paths or better final decisions. A key limitation is that these perturbations inject stochasticity without defining an explicit conditional sampling distribution, making latent exploration difficult to control or optimize. To address this, we propose the Gaussian Thought Sampler (GTS), a lightweight module that reformulates latent exploration as sampling from a learned conditional distribution over continuous reasoning states. GTS predicts context-dependent perturbation distributions and is trained with GRPO-style policy optimization while keeping the backbone frozen, turning heuristic perturbation into an explicit probabilistic sampling policy. Experiments across multiple benchmarks and two latent reasoning architectures show that GTS yields more reliable inference-time scaling than heuristic baselines, suggesting that effective latent ITS requires better-controlled and optimizable sampling rather than simply amplifying stochasticity.

cs.CL↗

LiveCultureBench: a Multi-Agent, Multi-Cultural Benchmark for Large Language Models in Dynamic Social Simulations

Large language models (LLMs) are increasingly deployed as autonomous agents, yet evaluations focus primarily on task success rather than cultural appropriateness or evaluator reliability. We introduce LiveCultureBench, a multi-cultural, dynamic benchmark that embeds LLMs as agents in a simulated town and evaluates them on both task completion and adherence to socio-cultural norms. The simulation models a small city as a location graph with synthetic residents having diverse demographic and cultural profiles. Each episode assigns one resident a daily goal while others provide social context. An LLM-based verifier generates structured judgments on norm violations and task progress, which we aggregate into metrics capturing task-norm trade-offs and verifier uncertainty. Using LiveCultureBench across models and cultural profiles, we study (i) cross-cultural robustness of LLM agents, (ii) how they balance effectiveness against norm sensitivity, and (iii) when LLM-as-a-judge evaluation is reliable for automated benchmarking versus when human oversight is needed.

cs.AI↗

Simultaneous Machine Translation with Large Language Models

Real-world simultaneous machine translation (SimulMT) systems face more challenges than just the quality-latency trade-off. They also need to address issues related to robustness with noisy input, processing long contexts, and flexibility for knowledge injection. These challenges demand models with strong language understanding and generation capabilities which may not often equipped by dedicated MT models. In this paper, we investigate the possibility of applying Large Language Models (LLM) to SimulMT tasks by using existing incremental-decoding methods with a newly proposed RALCP algorithm for latency reduction. We conducted experiments using the \texttt{Llama2-7b-chat} model on nine different languages from the MUST-C dataset. The results show that LLM outperforms dedicated MT models in terms of BLEU and LAAL metrics. Further analysis indicates that LLM has advantages in terms of tuning efficiency and robustness. However, it is important to note that the computational cost of LLM remains a significant obstacle to its application in SimulMT.

cs.CL↗

Conversational SimulMT: Efficient Simultaneous Translation with Large Language Models

Simultaneous machine translation (SimulMT) presents a challenging trade-off between translation quality and latency. Recent studies have shown that LLMs can achieve good performance in SimulMT tasks. However, this often comes at the expense of high inference cost and latency. In this paper, we propose a conversational SimulMT framework to enhance the inference efficiency of LLM-based SimulMT through multi-turn-dialogue-based decoding. Our experiments with Llama2-7b-chat on two SimulMT benchmarks demonstrate the superiority of LLM in translation quality while achieving comparable computational latency to specialized SimulMT models.

cs.CL↗