arXiv subjects

Hieu Tran

Publications and source records attributed to Hieu Tran.

At least 19 recordsLinked to original sources

arXiv2026

EpiLoRA: Fusing Many LoRA Adapters into Low-Precision GEMMs for Serving and RL

Low-rank adapters (LoRA) let one base model serve many fine-tuned variants: multi-tenant engines keep a single copy of the base and batch the requests of many tenants, each with its own adapter. As base models move to 8-bit (FP8) and 4-bit formats, their general matrix multiplications (GEMMs) become cheaper, but the adapters do not: engines run them as separate small GPU kernels beside the GEMM, whose latency takes a growing share of every decoding step. An adapter projects its input down to a low rank and back up, and our key observation is that the GEMM needs the down-projection only in its epilogue, the final stage in which it writes its output. EpiLoRA, a plugin for the vLLM engine, therefore launches the down-projection first and the low-precision GEMM as a dependent kernel that streams its weights meanwhile and adds every tenant's up-projection in its epilogue. On NVIDIA Hopper GPUs, EpiLoRA decodes for a single adapter with up to 24% lower time per token than vLLM's fastest configuration, and faster than SGLang and TensorRT-LLM; with many tenants, its adapter kernels deliver more throughput than vLLM's on most GPU-model pairs we tested. The same low-precision base also trains: a LoRA reinforcement-learning trainer that computes on the engine's FP8 weights needs no 16-bit copy, so a 27B model trains on one 48 GB GPU, and an engine that then keeps activations in 16 bits removes most of the remaining mismatch between the model generating training data and the one trained.

cs.DC↗

arXiv2026

SoundnessBench: Can Your AI Scientist Really Tell Good Research Ideas from Bad Ones?

Autonomous AI research agents aim to speed up scientific discovery by automating the research pipeline from hypothesis generation to peer review. However, existing benchmarks rarely test a fundamental bottleneck whether LLMs can judge the methodological viability of a research idea $\textbf{before}$ investing time and computational resources. Without a reliable ``first-gate'' filter, autonomous agents risk scaling flawed methodology rather than accelerating meaningful science. We introduce $\textbf{SoundnessBench}$, a curated benchmark of 1,099 real research proposals, with source verified extraction, reviewer-derived proxy labels, and proposal-only expert judgments on a subset of 327 proposals. Across 12 frontier LLMs, our evaluation reveals a pervasive $\textbf{optimism bias}$: standard prompting yields a $74.0\%$ mean false-positive rate. Aggressive prompting reduces false approvals but substantially increases false rejections. Additional analyses show that optimism persists under controls for public corpus contamination, and surface features, as well as in literature-grounded agent evaluation. These results suggest a limitation in judging scientific soundness, indicating that current LLMs are not yet reliable as standalone gatekeepers for scientific rigor.

cs.LG↗

arXiv2026

Second Thoughts: Exploring and Planning through Comparative Imagination in World Models

World models let agents compare imagined outcomes before acting, but an agent must still decide which alternatives to explore and when a proposed plan should replace its learned policy. We introduce SDI (Sibling Disagreement in Imagination), which connects these decisions through a shared actor-based reference. SDI compares action sequences imagined from the same model state, using repeated rollouts of the actor's own sequence to measure variation under fixed behavior. This reference scales both the spread across candidate returns, which provides an exploration signal, and a proposed plan's predicted advantage, which governs action acceptance. The agent executes a planned action when its advantage exceeds the reference variability by a fixed factor; otherwise, it follows the actor. SDI reuses the base agent's world model, actor, and critic without requiring an additional disagreement ensemble. Exploration and planning guide training data collection, while evaluation uses the learned actor alone. Across Meta-World, Adroit, and DM-Control, SDI achieves a higher observed task-average score than every compared baseline in each suite and matches or exceeds the strongest baseline's performance on 15 of 18 tasks. Component ablations and robustness analyses further examine the roles of exploration, planning exposure, and action acceptance. Paired evaluations from identical simulator states show that the gate improves both recovery from actor failures and preservation of successful behavior relative to ungated planning, while substantially reducing return degradation. Together, these results support comparing imagined alternatives against a shared actor reference as a unified approach to exploring and planning with world models.

cs.LG↗

arXiv2026

Measure First, Then Allocate: Calibrated Rollout Budgets for MCP Agents in Reinforcement Learning

Training tool-using language models requires costly interaction, yet identical rewards within a rollout group provide no group-relative learning signal. We use reward disagreement to guide harness evaluation and the allocation of rollouts for policy training in Model Context Protocol (MCP) environments. We first introduce Discord, which tests harness edits using exact paired inference and decision-preserving early stopping: only disagreeing paired outcomes favor one harness over another. This insight motivates prospective allocation in BRACE (Bank-calibrated Rollout Allocation under Correlated Episodes). BRACE combines historical task evidence, uncertainty in task success rates, and a correlated reward model to forecast mixed-reward groups before allocating generation budget. Controlled harness experiments favor Discord, while live MCP experiments expose a retention-cost trade-off. At three Qwen3-VL model scales, BRACE exceeds the strongest plotted baseline peaks with fewer new training rollouts; historical-bank collection is counted separately. It also leads the compared methods on BFCL and NESTFUL without further training. Ablations probe the bank, estimator, and allocation score; prospective evaluation checks forecast calibration. These results support forecasting reward contrast before generation.

cs.LG↗

arXiv2026

Is Anyone Enforcing the Cartel? What Reinforcement Learning Teaches LLM Agents in Markets

Evaluations of large language model (LLM) agents report that pricing agents collude at others' expense. Yet collusion is defined by enforcement: a cartel, in which competitors share monopoly profits, holds only if a member who cheats is punished. We introduce an audit that tests enforcement by forcing a member to cheat in counterfactual copies of a market, and use it to study what reinforcement learning (RL) teaches agents in pricing and trading. Small LLMs trained by RL learn what their training rivals reward, but only what they readily try. In self-play, they learn fixed rules ignoring their rivals; against a punishing rival, to be deterred; against a cheating rival, to match its cuts, never the costly price war that would deter it, which they rarely try. A war forced into training is learned, but mostly only where it was forced. Against their own copies, deterred and matching agents can earn the same cartel profits, but only matching ones can enforce them, and then thinly: as a by-product and only against large cuts. Q-learners enforce collusion only in pricing. Prompted LLMs reach cartel-like outcomes mostly through rules that ignore the market. Outcome measures cannot tell these cases apart; the audit can.

cs.CL↗

arXiv2026

ArcFlow: Training Flow Policies for Multiple Solver Budgets

Flow-matching policies adjust inference cost through the number of solver steps, but reinforcement learning at a single fixed solver budget can hide degradation at cheaper deployment budgets and even erase useful one-step behavior present in a pretrained policy. We introduce Action-Ratio-Clip Flow (ArcFlow), a deployment-aware algorithm for training one flow policy across a specified range of solver budgets. ArcFlow combines action-aligned supervision, budget exposure, and worst-budget checkpoint selection. For reconstruction objectives, its bridge aggregates temporal probes into one surrogate ratio and one PPO-style clip per action, with a boundary residual exactly equal to deployed one-step reconstruction error. Cycling exposes the shared field to the deployment range, and selection optimizes its performance floor. For native velocity objectives, multi-budget endpoint supervision applies the same principle without replacing the native update. Both realizations preserve the architecture and deployment-time cost. Across DM Control, manipulation, and locomotion, ArcFlow attains the highest mean one-step return among five flow-policy baselines on 25 of 30 tasks under each method's reported training settings. Component studies identify complementary improvements from boundary supervision, the fixed-grid bridge, budget exposure, and deployment selection. These results establish deployment-budget-aware training as an effective route to strong low-step control with one policy serving multiple inference costs.

cs.LG↗

arXiv2026

When AI Reviews Train AI Reviewers: Scientific-Judgment Collapse and Mitigation

LLMs increasingly assist scientific peer review. This creates a feedback loop when generated reviews enter the training data of future LLMs. While recursive training has been studied broadly, its effects on scientific judgments remain poorly understood. We empirically study this effect in a controlled setting by varying synthetic-review exposure while fixing other training conditions. After one recursive step, increasing exposure from $0\%$ to $100\%$ makes recommendation ratings more concentrated, independently generated reviews of the same paper about $11\%$ more similar, and reviews across different papers about $5\%$ less diverse. A further step of recursive training amplifies homogenization when synthetic supervision is reintroduced. We characterize this task-specific phenomenon as $\textbf{scientific-judgment collapse}$. We also explore a training-free method using contrastive activation steering with a limited reference reviews. Controlled experiments show a partial recovery in recommendation agreement and judgment diversity across different papers, but not among independently generated reviews of the same paper. To explore this approach beyond controlled experiments, we develop $\textbf{TrustReviewer}$. We train the model on a large-scale curated data and apply activation steering at inference time. On held-out papers, steering increases exact recommendation match from $73.85\%$ to $75.40\%$, compared with $73.10\%$ for OpenReviewer on the same LLM backbone, while maintaining relatively diverse scientific judgments. Our findings quantify judgment homogenization under recursive training and motivate further research on preserving judgment diversity.

cs.LG↗

arXiv2026

Identity-Based Signing Shifts, Not Solves, Usability: Evidence from Five Identity-Based Signing Ecosystems

Identity-based software signing tools aim to make software artifact provenance verifiable while reducing the burden of long-lived key management. These tools are becoming critical infrastructure for secure software supply chains and are increasingly embedded in build, deployment, and release workflows. However, there is limited cross-tool, longitudinal evidence about which usability problems arise in practice and how reported concerns change over time. This gap matters because unusable signing and verification workflows can reduce adoption or lead to misconfiguration, undermining intended integrity guarantees. We conducted the first mining-software-repositories study of five identity-based signing ecosystems with public development repositories: Sigstore, OpenPubKey, HashiCorp Vault, Keyfactor, and Notary v2. We analyzed approximately 3,900 GitHub issue discussions from November 2021 to November 2025. We coded each issue for reported usability concerns and affected surfaces, comparing patterns across tools and over time. Across ecosystems, reported concerns are especially prominent around CLI tooling, policy/configuration, and verification workflows. Longitudinal Poisson trend analysis shows that reported issue counts decline substantially for several repositories, increase for OpenPubKey, and show no statistically detectable overall trend for the Keyfactor repositories, while trajectories vary across concern types. These results indicate that identity-based signing reduces direct key-management burdens while relocating complexity to policy configuration, verification, cross-component coordination, and workflow integration. Our findings motivate treating verification semantics, policy validation, cross-component observability, and deployment integration as first-class usability targets.

cs.SE↗

arXiv2026

MultiViewDx: Evidence-Linked Multi-View Clinical Diagnosis

Medical multimodal large language models (MLLMs) can perform well on existing medical visual question answering (MedVQA) benchmarks, but their training data often does not match clinical diagnosis. Most supervision is organized around isolated images or short QA pairs, leaving two structures weakly specified: how evidence leads to a decision, and how views, series, modalities, and patient context from the same case are linked. We introduce MultiViewDx, a partly physician-validated multimodal instruction dataset for evidence-linked multi-view medical imaging diagnosis. MultiViewDx uses the clinical case as the supervision unit. It links imaging studies with patient context, normalizes heterogeneous reports into an evidence-linked workflow (evidence -> findings -> differential discussion -> diagnosis), and uses a unified image-text retriever to constrain instruction synthesis to source-supported evidence. It covers X-ray, CT, MRI, ultrasound, histopathology, and other clinical visual sources. We fine-tune MultiViewDx-8B-AN and evaluate it on both existing MedVQA benchmarks and real-world case-based diagnostic reasoning. Across four MedVQA benchmarks, it achieves the best average accuracy among compared systems (79.0%), outperforming HuatuoGPT-Vision-34B (66.7%) and Claude3-Opus (55.7%). Beyond MedVQA, on JAMA Clinical Challenge cases, it receives the strongest overall rating under a physician-designed rubric for key clinical points, diagnostic inference, and evidence grounding. Controlled ablations and clinician evaluation show that both case-level multi-view organization and evidence-linked reasoning targets contribute to the gain.

cs.CL↗

arXiv2026

Can Activation Steering Capture Multidimensional Authorship Style?

Activation steering has shown promise for controlling LLM generation along well-defined attributes, but it remains unclear whether it can handle the multidimensional and hard-to-define nature of authorship style. We ask whether structured contrastive prompting along rhetorically-motivated dimensions can construct rich style representations directly in activation space, bypassing the need for natural language style descriptors or dedicated training. We find that the resulting directions share a common authorship backbone while conflicting on aspect-specific residuals that carry genuine stylistic signal, explaining why naive aggregation fails. We operationalize this in Aspect-Aware Activation Steering (A3S), a training-free framework that merges per-aspect contrastive directions with interference-aware aggregation and tunes steering strength per instance. A3S improves authorship style transfer where it is genuinely multi-aspect, outperforms a trained baseline in preference evaluations on out-of-domain benchmarks, and keeps target-exemplar overlap consistently low.

cs.CL↗

arXiv2026

Enhancing LLMs for Identifying and Prioritizing Important Medical Jargons from Electronic Health Record Notes Utilizing Data Augmentation: A Comparative Study

OpenNotes gives patients access to their EHR notes, but dense medical jargon limits comprehension. We evaluate closed-source and open-source LLMs for extracting and prioritizing the jargon terms most relevant to individual patients, using 90 expert-annotated EHR notes. We test combinations of general vs. structured prompts, zero-shot vs. few-shot prompting, fine-tuning, and GPT-4o-based data augmentation, the last paired with a ranking technique to refine training in low-resource settings. To assess the effect of dataset size, we fine-tune on augmented datasets scaled from 10 to 9,995 examples. All settings are evaluated with 10-fold cross-validation, reporting F1 and Mean Reciprocal Rank (MRR) under two string-matching criteria (relaxed matching and Jaccard Index), followed by an error analysis of model outputs. Open-source models performed best when fine-tuned on the gold-standard dataset: under Jaccard-based matching, DeepSeek 8B achieved the top F1 (0.431, SD 0.046) and BioMistral 7B the top MRR (0.577, SD 0.109). Under relaxed matching, however, open-source models did not match closed-source performance even with augmentation or fine-tuning. Few-shot prompting offered no advantage over zero-shot in vanilla models; prompting style substantially affected results; fine-tuning on a small gold-standard set improved performance; and data augmentation matched or exceeded fine-tuning, though its benefit depended heavily on augmented-data quality. These findings show that prompting strategy, fine-tuning, and data augmentation each meaningfully improve LLM performance on patient-centered jargon extraction in low-resource clinical settings.

cs.CL↗

arXiv2026

VietMix: A Naturally-Occurring Parallel Corpus and Augmentation Framework for Vietnamese-English Code-Mixed Machine Translation

Machine translation (MT) systems universally degrade when faced with code-mixed text. This problem is more acute for low-resource languages that lack dedicated parallel corpora. This work directly addresses this gap for Vietnamese-English, a language context characterized by challenges including orthographic ambiguity and the frequent omission of diacritics in informal text. We introduce VietMix, the first expert-translated, naturally occurring parallel corpus of Vietnamese-English code-mixed text. We establish VietMix's utility by developing a data augmentation pipeline that leverages iterative fine-tuning and targeted filtering. Experiments show that models augmented with our data outperform strong back-translation baselines by up to +3.5 xCOMET points and improve zero-shot models by up to +11.9 points. Our work delivers a foundational resource for a challenging language pair and provides a validated, transferable framework for building and augmenting corpora in other low-resource settings.

cs.CL↗

arXiv2025

PRIME: Planning and Retrieval-Integrated Memory for Enhanced Reasoning

Inspired by the dual-process theory of human cognition from \textit{Thinking, Fast and Slow}, we introduce \textbf{PRIME} (Planning and Retrieval-Integrated Memory for Enhanced Reasoning), a multi-agent reasoning framework that dynamically integrates \textbf{System 1} (fast, intuitive thinking) and \textbf{System 2} (slow, deliberate thinking). PRIME first employs a Quick Thinking Agent (System 1) to generate a rapid answer; if uncertainty is detected, it then triggers a structured System 2 reasoning pipeline composed of specialized agents for \textit{planning}, \textit{hypothesis generation}, \textit{retrieval}, \textit{information integration}, and \textit{decision-making}. This multi-agent design faithfully mimics human cognitive processes and enhances both efficiency and accuracy. Experimental results with LLaMA 3 models demonstrate that PRIME enables open-source LLMs to perform competitively with state-of-the-art closed-source models like GPT-4 and GPT-4o on benchmarks requiring multi-hop and knowledge-grounded reasoning. This research establishes PRIME as a scalable solution for improving LLMs in domains requiring complex, knowledge-intensive reasoning.

cs.AI↗

arXiv2025

Chatbot To Help Patients Understand Their Health

Patients must possess the knowledge necessary to actively participate in their care. We present NoteAid-Chatbot, a conversational AI that promotes patient understanding via a novel 'learning as conversation' framework, built on a multi-agent large language model (LLM) and reinforcement learning (RL) setup without human-labeled data. NoteAid-Chatbot was built on a lightweight LLaMA 3.2 3B model trained in two stages: initial supervised fine-tuning on conversational data synthetically generated using medical conversation strategies, followed by RL with rewards derived from patient understanding assessments in simulated hospital discharge scenarios. Our evaluation, which includes comprehensive human-aligned assessments and case studies, demonstrates that NoteAid-Chatbot exhibits key emergent behaviors critical for patient education, such as clarity, relevance, and structured dialogue, even though it received no explicit supervision for these attributes. Our results show that even simple Proximal Policy Optimization (PPO)-based reward modeling can successfully train lightweight, domain-specific chatbots to handle multi-turn interactions, incorporate diverse educational strategies, and meet nuanced communication objectives. Our Turing test demonstrates that NoteAid-Chatbot surpasses non-expert human. Although our current focus is on healthcare, the framework we present illustrates the feasibility and promise of applying low-cost, PPO-based RL to realistic, open-ended conversational domains, broadening the applicability of RL-based alignment methods.

cs.AI↗

arXiv2025

Exploiting Tree Structure for Credit Assignment in RL Training of LLMs

Reinforcement learning improves LLM reasoning, yet sparse delayed reward over long sequences makes token-level credit assignment the key bottleneck. We study the verifiable-reward setting, where the final answer is checkable and multiple responses can be drawn per prompt. Reasoning tasks in math and medical QA align with this setup, where only a few decision tokens significantly impact the outcome. PPO offers token-level advantages with a learned value model, but it is complex to train both the actor and critic models simultaneously, and it is not easily generalizable, as the token-level values from the critic model can make training prone to overfitting. GRPO is critic-free and supports verifiable rewards, but spreads a single sequence-level return across tokens and ignores branching. We introduce \textbf{Prefix-to-Tree (P2T)}, a simple procedure that converts a group of responses into a prefix tree and computes \emph{nonparametric} prefix values $V(s)$ by aggregating descendant outcomes. Built on P2T, we propose \textbf{TEMPO} (\emph{\textbf{T}ree-\textbf{E}stimated \textbf{M}ean Prefix Value for \textbf{P}olicy \textbf{O}ptimization}), a critic-free algorithm that augments the group-relative outcome signal of GRPO with \emph{branch-gated} temporal-difference corrections derived from the tree. At non-branch tokens, the temporal-difference (TD) term is zero, so TEMPO reduces to GRPO; at branching tokens, it supplies precise token-level credit without a learned value network or extra judges/teachers. On Qwen3-1.7B/4B, TEMPO outperforms PPO and GRPO on in-distribution (MATH, MedQA) and out-of-distribution (GSM-HARD, AMC23, MedMCQA, MMLU-Medical) benchmarks, and reaches higher validation accuracy with roughly the same wall-clock time.

cs.CL↗

arXiv2025

A Robot That Listens: Enhancing Self-Disclosure and Engagement Through Sentiment-based Backchannels and Active Listening

As social robots get more deeply integrated intoour everyday lives, they will be expected to engage in meaningful conversations and exhibit socio-emotionally intelligent listening behaviors when interacting with people. Active listening and backchanneling could be one way to enhance robots' communicative capabilities and enhance their effectiveness in eliciting deeper self-disclosure, providing a sense of empathy,and forming positive rapport and relationships with people.Thus, we developed an LLM-powered social robot that can exhibit contextually appropriate sentiment-based backchannelingand active listening behaviors (active listening+backchanneling) and compared its efficacy in eliciting people's self-disclosurein comparison to robots that do not exhibit any of these listening behaviors (control) and a robot that only exhibitsbackchanneling behavior (backchanneling-only). Through ourexperimental study with sixty-five participants, we found theparticipants who conversed with the active listening robot per-ceived the interactions more positively, in which they exhibited the highest self-disclosures, and reported the strongest senseof being listened to. The results of our study suggest that the implementation of active listening behaviors in social robotshas the potential to improve human-robot communication andcould further contribute to the building of deeper human-robot relationships and rapport.

cs.HC↗

arXiv2025

MedReadCtrl: Personalizing medical text generation with readability-controlled instruction learning

Generative AI has demonstrated strong potential in healthcare, from clinical decision support to patient-facing chatbots that improve outcomes. A critical challenge for deployment is effective human-AI communication, where content must be both personalized and understandable. We introduce MedReadCtrl, a readability-controlled instruction tuning framework that enables LLMs to adjust output complexity without compromising meaning. Evaluations of nine datasets and three tasks across medical and general domains show that MedReadCtrl achieves significantly lower readability instruction-following errors than GPT-4 (e.g., 1.39 vs. 1.59 on ReadMe, p<0.001) and delivers substantial gains on unseen clinical tasks (e.g., +14.7 ROUGE-L, +6.18 SARI on MTSamples). Experts consistently preferred MedReadCtrl (71.7% vs. 23.3%), especially at low literacy levels. These gains reflect MedReadCtrl's ability to restructure clinical content into accessible, readability-aligned language while preserving medical intent, offering a scalable solution to support patient education and expand equitable access to AI-enabled care.

cs.CL↗

arXiv2025

RARE: Retrieval-Augmented Reasoning Enhancement for Large Language Models

This work introduces RARE (Retrieval-Augmented Reasoning Enhancement), a versatile extension to the mutual reasoning framework (rStar), aimed at enhancing reasoning accuracy and factual integrity across large language models (LLMs) for complex, knowledge-intensive tasks such as commonsense and medical reasoning. RARE incorporates two innovative actions within the Monte Carlo Tree Search (MCTS) framework: A6, which generates search queries based on the initial problem statement, performs information retrieval using those queries, and augments reasoning with the retrieved data to formulate the final answer; and A7, which leverages information retrieval specifically for generated sub-questions and re-answers these sub-questions with the relevant contextual information. Additionally, a Retrieval-Augmented Factuality Scorer is proposed to replace the original discriminator, prioritizing reasoning paths that meet high standards of factuality. Experimental results with LLaMA 3.1 show that RARE enables open-source LLMs to achieve competitive performance with top open-source models like GPT-4 and GPT-4o. This research establishes RARE as a scalable solution for improving LLMs in domains where logical coherence and factual integrity are critical.

cs.CL↗