arXiv subjects

Zhijun Yin

Publications and source records attributed to Zhijun Yin.

At least 19 recordsLinked to original sources

arXiv2026

READY or Not: Reliable Enterprise Agent Deployment

An AI agent can perform well on benchmarks and still be unsuitable for deployment. Existing AI-agent benchmarks measure whether an agent can complete realistic professional work, whereas enterprise deployment asks a different question: whether an agent can meet a required reliability level, under acceptable human oversight, and at tolerable cost. We introduce Reliable Enterprise Agent Deployment (READY), a framework for qualifying AI agents for deployment on enterprise workflows. READY preserves each workflow's own definition of successful execution while applying a common qualification procedure. Given an agent, a workflow, and a class of candidate oversight policies, READY measures the reliability and operating cost of the human-AI system, selects the minimum-cost policy that satisfies a specified reliability target, and statistically qualifies it on held-out cases. The resulting deployment profile characterizes the supported operating point: reliability, human-oversight burden, and cost. READY is implemented as an open testbed that decouples workflow specification, execution, evaluation, and qualification, and runs on existing agent-evaluation infrastructure. In an end-to-end clinical-audit case study spanning 16 agent systems and 750 cases, READY reveals differences hidden by autonomous performance: two systems separated by only 0.3 points in autonomous accuracy (72.8% vs. 72.5%) require 39.2% versus 29.6% human review, respectively, to qualify at the same 76% reliability target under the evaluated oversight policy. READY thus shifts enterprise agent evaluation from how well can the agent perform the work? to under what conditions, and at what cost, can it be reliably deployed? By making those conditions explicit and statistically testable, READY provides a basis for comparing agent systems, setting oversight requirements, and making evidence-based deployment decisions.

cs.AI↗

arXiv2026

CliniCARE-Bench: Clinical Calibrated Audit of Medical Reasoning in EHR

Large language models perform strongly on medical knowledge benchmarks, but reliable clinical deployment requires agents to conduct defensible investigations over heterogeneous, longitudinal records: determining what evidence is needed, retrieving and reconciling structured and free-text data, grounding conclusions in verifiable evidence, and deferring cases that cannot be resolved reliably. We introduce CliniCARE-Bench (Clinical Calibrated Audit of Medical Reasoning in EHR), a benchmark for retrospective clinical audit: 25 clinician-validated scenarios instantiated as 750 patient-specific cases over real-patient-derived MIMIC-IV data. Systems investigate each case through a governed, logged tool environment for record retrieval, computation, and policy access, and return one of four verdicts---Yes, No, Indeterminate: Lack of Data, or Indeterminate: Medically Ambiguous---the last two separating missing evidence from residual medical ambiguity. Beyond verdict accuracy, we score patient-evidence and policy grounding, process adherence, calibrated abstention, reliability, and efficiency against case-level reference verdicts produced by independent multi-model adjudication and calibrated against Clinical Board review. Every retrieval, computation, and report is replayable, so the investigation trace is inspectable and scorable. To our knowledge, CliniCARE-Bench is the first deployment-oriented clinical-agent benchmark to jointly evaluate real longitudinal EHR investigation, claim-level evidence grounding, governing-policy use, process adherence, and calibrated abstention within a common patient-level adjudication framework. Across 16 agentic systems, four-way accuracy spans 65.3-76.1%, but raw accuracy overstates investigation quality. Defect-free accuracy, which credits a verdict only when correct and free of prohibited shortcuts, is 4.8-14.8 points lower and reorders the leaderboard.

cs.AI↗

arXiv2026

Characterizing Treatment-Context Medication Evidence Across Clinic Notes and Structured EHR Medication History

Clinic notes and structured electronic health record (EHR) medication history often contain different medication information. Same-visit disagreement between these sources may result from note-side normalization errors, differences in terminology or timing, or actual differences in documentation. We developed a note-grounded approach that uses large language model (LLM) assisted reference construction, targeted and random human review, deterministic medication normalization, and semantic and temporal comparisons with structured medication history. We evaluated all normalization results on a patient-level held-out test set to limit adaptation to the study cohort. On 5,403 held-out mention rows, exact canonical agreement improved from 0.7226 with surface-exact matching to 0.8429 after lexical cleanup and curated alias mapping. In a random audit of previously unaudited rows, canonical-label agreement was 0.9210 among evaluable valid medication mentions, whereas treatment-action attribution was lower at 0.5326. In the full-cohort characterization analysis, only 16.44% of note-derived rows had same-visit exact overlap with structured medication history, but 55.17% had same-visit semantic overlap, 90.34% had same-visit or +/-30-day overlap, and only 3.97% remained in the strict no-structured-overlap bucket under broad project-level mapping. An ontology-backed sensitivity analysis further showed that held-out strict Observational Medical Outcomes Partnership (OMOP)-backed no-overlap fell from 43.99% to 36.68% after a development-derived alias supplement. These results show that note-to-structured-medication mismatch can arise from normalization errors, differences in terminology, and differences in documentation timing.

cs.CL↗

arXiv2026

DRIFT: Direct-Recursive Intervention-Conditioned Forecasting of ICU Physiological Trajectories

Many time-series forecasts depend not only on prior observations but also on actions specified during the forecast period. In intensive care units (ICUs), future vital signs and laboratory values are influenced by treatments such as vasopressors. However, models that predict the full future sequence all at once make little use of these treatments, whereas autoregressive models can accumulate errors. We introduce DRIFT, a hybrid framework in which a direct model produces the primary forecast and a recursive, action-conditioned model contributes constrained corrections. We evaluate DRIFT on 6,046 admissions from MIMIC-IV and 8,345 admissions from eICU-CRD. Averaged across the 8-, 24-, and 48-hour forecast endpoints, DRIFT reduces mean absolute error for mean arterial pressure (MAP) by 0.673% relative to an action-conditioned Temporal Fusion Transformer (TFT-action) on MIMIC-IV and achieves the lowest corresponding error among the compared models on eICU-CRD. Although the overall accuracy improvement is modest, a MIMIC-IV audit restricted to windows in which the supplied treatment sequence was altered showed that DRIFT achieved lower observed-target MAP error than TFT-action at 8 and 24 hours. Treatment-sequence alteration increased DRIFT's MAP error by 0.21-0.26 mmHg more than it increased TFT-action's error, with prediction changes occurring primarily after the supplied paths diverged. In a separate robustness experiment, the MAP advantage persisted under three shared checkpoint-selection rules emphasizing overall endpoint error, MAP error, or both equally.

cs.LG↗

arXiv2026

CoRA: Confidence-Rationale Alignment for Reliable Chain-of-Thought Reasoning

Chain-of-thought (CoT) reasoning can improve LLM performance, but high answer confidence may be misleading when the accompanying CoT rationale is plausible yet incomplete or poorly supported. We study confidence--rationale alignment: whether a model's confidence in its committed answer is justified by its generated rationale. We introduce a GRPO-based reinforcement learning framework that jointly rewards answer correctness, committed-answer probability, and rubric-based rationale support, where the rubric assesses grounding, coherence, task match, and connection to the selected answer without revealing the gold answer to the judge. Across MedQA, MathQA, and OpenBookQA using three open-weight LLMs, our method reduces the confidence--rationale alignment error by up to 26.51% compared with untuned checkpoints, SFT, and correctness-only GRPO, while maintaining competitive accuracy and often improving calibration. These results show that reliable CoT reasoning requires not only confident answers, but rationales that substantively support them.

cs.CL↗

arXiv2026

RadOT-Eval: Auditable Structured-Evidence Transport for Radiology Report Evaluation

Automatic evaluation is critical for high-stakes text generation, where errors often involve omitted findings, hallucinated content, polarity reversals, location changes, uncertainty mismatches, and temporal-comparison errors rather than low surface similarity alone. Radiology report generation provides a challenging test case because generated reports must preserve structured clinical evidence across sources. We present RadOT-Eval, an interpretable structured-evidence optimal transport framework for offline auditing of radiology report generation. RadOT-Eval decomposes reference and candidate reports into attribute-structured clinical evidence units, aligns corresponding evidence using entropy-regularized optimal transport, and uses clinically meaningful side-channel discrepancies in a monotone risk model to predict error burden. All transport, feature, and readout choices are selected using the ReXVal dataset, and the frozen system is evaluated on the independent RadEvalX dataset. RadOT-Eval achieves Spearman correlations of 0.715, 0.548, and 0.399 with total, clinically significant, and clinically insignificant annotated error burden, respectively, yielding higher point estimates than standard evaluation metrics and the open-source large language model (LLM)-based evaluator GREEN-radllama2-7B. In a frozen auxiliary corruption-sensitivity stress test on ReXErr-v1, RadOT-Eval achieves 0.768 AUROC and a 0.990 corrupted-greater-than-clean paired win rate. These results show that structured evidence transport provides an auditable, rank-oriented evaluation tool for high-stakes generated clinical text under ReXVal-only model selection and frozen RadEvalX testing.

cs.CL↗

arXiv2026

PSEBench: A Controllable and Verifiable Benchmark for Evaluating LLMs in Patient Safety Event Triage

Patient safety event triage, determining whether a clinical event is reportable under jurisdiction-specific policy, is a high-stakes task typically performed manually by patient safety experts. Although LLMs may support this workflow, reliable evaluation is limited by the lack of benchmarks to capture evidence-grounded policy reasoning, proactive information seeking for incomplete reports, and principled abstention in irreducibly ambiguous cases. We address this gap with a policy-grounded construction methodology centered on the clause card, a structured representation that factorizes regulatory text into auditable decision specifications. Combining clause cards with anchor-driven instantiation and closed-loop verification, our scalable pipeline produces narratives with by-construction ground truth and naturally supports generating missing information and uncertain variants. We instantiate this method on Minnesota's 29 Reportable Adverse Health Events, producing PSEBench, a 5,074-case benchmark with an agentic evaluation environment. Evaluation on 15 representative LLMs reveals consistent capability trends, demonstrates the benchmark's utility, and identifies actionable gaps toward reliable LLM-based patient safety event triage.

cs.AI↗

arXiv2026

Vectors Are Not Neutral: Sensitive-Information Inference from Exported LLM Representations in Summarization

Large language model (LLM) summarization systems may pass compact vector representations of private inputs to downstream retrieval, monitoring, audit, or analytic workflows. Even when source documents remain access-restricted, derived vectors may be handled under different access controls and still support sensitive-information inference, creating a residual information-disclosure risk. We study this issue in clinical discharge-summary generation as a high-stakes case study, using electronic health record (EHR)-recorded race as a controlled sensitive-label audit. We audit two artifacts that a system might retain or expose to downstream components: the final prompt-token hidden state and the mean-pooled prompt representation. Our results show that reducing recoverability of the case-study sensitive label from one exported artifact does not necessarily reduce recoverability from another. As a mitigation case study, we introduce SurfaceLoRA, an exported-vector-targeted parameter-efficient fine-tuning method that uses a gradient-reversal discriminator attached to a designated exported vector. Under a balanced five-way probing protocol, SurfaceLoRA reduces EHR-recorded race recoverability from the targeted final-token artifact toward chance while preserving summarization utility, yet recoverability remains substantially higher from untargeted pooled artifacts. These findings show that privacy auditing and mitigation should be performed on the exact vector artifact retained or exposed to downstream components.

cs.CL↗

arXiv2026

It's Not Always Sycophancy: Measuring LLM Conformity as a Function of Epistemic Uncertainty

Large language models (LLMs) are known to abandon their initial stance to conform to user pushback. While prior research largely attributes this behavior to sycophancy learned during reinforcement learning from human feedback, we hypothesize that conformity is also driven by a model's epistemic uncertainty at inference time. In this paper, we introduce MUSE, a two-stage evaluation framework to disentangle the mechanisms driving LLM conformity. Specifically, MUSE maps a model's epistemic uncertainty in responding to a query against its likelihood to yield to user pushback in a subsequent turn. We demonstrate that the mechanisms driving conformity extend beyond sycophancy alone. Specifically, we characterize two distinct factors that jointly drive conformity: sycophantic conformity, where a model aligns with user pushback even with absolute certainty in its initial response, and uncertainty-driven conformity, where a model's likelihood for conformity increases alongside its uncertainty. Furthermore, we conduct ablation studies to demonstrate that both sycophantic conformity and uncertainty-driven conformity grow with 1) the LLM's perceived expertise of the user and 2) the plausibility of the user's suggestions. More broadly, MUSE informs more targeted intervention strategies by distinguishing alignment-induced sycophancy and training-corpora-driven uncertainty.

cs.CL↗

arXiv2026

MHGraphBench: Knowledge Graph-Grounded Benchmarking of Mental Health Knowledge in Large Language Models

Large language models (LLMs) are increasingly used in the mental health domain, yet it remains unclear how well they capture related biomedical knowledge and how reliably they apply it to clinically salient structured judgments. Here, we present a knowledge-graph (KG)-grounded benchmark for assessing LLMs on mental-health entity recognition, relation judgment, and two-hop reasoning. The benchmark is derived from PrimeKG and comprises nine task families with KG-supported answers and controlled negative options. Experiments across 15 closed- and open-source LLMs reveal a persistent recognition-to-judgment gap: leading models achieve near-ceiling performance on entity typing and on the small relation-typing subset, yet they still struggle with relation prediction and two-hop reasoning. Additionally, short KG-derived snippets benefit some models but degrade performance for others. Moreover, output-format reliability can substantially influence measured performance under constrained multiple-choice settings, highlighting the critical role of response validity in benchmark-based evaluation. MHGraphBench should therefore be interpreted as evaluating agreement with a curated mental-health slice of PrimeKG under a constrained multiple-choice interface, rather than as a direct assessment of real-world clinical safety.

cs.CL↗

arXiv2026

CLEAR: Revealing How Noise and Ambiguity Degrade Reliability in LLMs for Medicine

Medical large language model (LLM) evaluations rely on simplified, exam-style benchmarks that rarely reflect the ambiguity of real-world medical inquiries. We introduce the CLinical Evaluation of Ambiguity and Reliability (CLEAR) framework, which assesses how decision-space presentation, ambiguity, and uncertainty affect LLMs' reasoning on medical benchmarks. CLEAR systematically perturbs (1) the number of plausible answer options, (2) the presence of a ground truth or abstention option, and (3) the semantic framing of answer options. Applying CLEAR on three benchmarks evaluated across 17 LLMs reveals three notable limitations of existing evaluation methods. First, increasing the number of plausible answers degrades a model's ability to identify the correct answer and abstain against incorrect ones. Second, this lack of caution intensifies as the framing of abstention shifts from assertive rejection like "None of the Above" to uncertainty admission like "I don't know" (IDK). Notably, just including IDK in the answer space increases incorrect answer selections. Lastly, we formalize the performance gap between identifying the correct answer and abstaining from incorrect ones as the humility deficit, which worsens with model scale. Our findings reveal limitations in standard medical benchmarks and underscore that scaling alone does not resolve LLM reliability issues.

cs.CL↗

arXiv2026

Robust Nasality Representation Learning for Cleft Palate-Related Velopharyngeal Dysfunction Screening in Real-World Settings

Velopharyngeal dysfunction (VPD) is characterized by inadequate velopharyngeal closure during speech and often causes hypernasality and reduced intelligibility. Although speech-based machine learning models can perform well under standardized clinical recording conditions, their performance often drops in real-world settings because of domain shift caused by differences in devices, channels, noise, and room acoustics. To improve robustness, we propose a two-stage framework for VPD screening. First, a nasality-focused speech representation is learned by supervised contrastive pre-training on an auxiliary corpus with phoneme alignments, using oral-context versus nasal-context supervision. Second, the encoder is frozen and used with lightweight classifiers on 0.5-second speech chunks, whose probabilities are aggregated to produce recording-level decisions with a fixed threshold. On an in-domain clinical cohort of 82 subjects, the proposed method achieved perfect recording-level screening performance (macro-F1 = 1.000, accuracy = 1.000). On a separate out-of-domain set of 131 heterogeneous public Internet recordings, large pretrained speech representations degraded substantially, while MFCC was the strongest baseline (macro-F1 = 0.612, accuracy = 0.641). The proposed method achieved the best out-of-domain performance (macro-F1 = 0.679, accuracy = 0.695), improving on the strongest baseline under the same evaluation protocol. These results suggest that learning a nasality-focused representation before clinical classification can reduce sensitivity to recording artifacts and improve robustness for deployable speech-based VPD screening.

eess.AS↗

arXiv2026

Blending Human and LLM Expertise to Detect Hallucinations and Omissions in Mental Health Chatbot Responses

As LLM-powered chatbots are increasingly deployed in mental health services, detecting hallucinations and omissions has become critical for user safety. However, state-of-the-art LLM-as-a-judge methods often fail in high-risk healthcare contexts, where subtle errors can have serious consequences. We show that leading LLM judges achieve only 52% accuracy on mental health counseling data, with some hallucination detection approaches exhibiting near-zero recall. We identify the root cause as LLMs' inability to capture nuanced linguistic and therapeutic patterns recognized by domain experts. To address this, we propose a framework that integrates human expertise with LLMs to extract interpretable, domain-informed features across five analytical dimensions: logical consistency, entity verification, factual accuracy, linguistic uncertainty, and professional appropriateness. Experiments on a public mental health dataset and a new human-annotated dataset show that traditional machine learning models trained on these features achieve 0.717 F1 on our custom dataset and 0.849 F1 on a public benchmark for hallucination detection, with 0.59-0.64 F1 for omission detection across both datasets. Our results demonstrate that combining domain expertise with automated methods yields more reliable and transparent evaluation than black-box LLM judging in high-stakes mental health applications.

cs.CL↗

arXiv2026

Stop Listening to Me! How Multi-turn Conversations Can Degrade LLM Reliability

Large language models (LLMs) excel on static benchmarks, but their performance across multi-turn conversations, which better reflect real-world usage, remains understudied. Addressing this gap is critical in high-stakes settings like healthcare, where patients and clinicians are turning to LLM chatbots to address their medical inquiries. Here, we introduce the "stick-or-switch" (SoS) framework, which partitions a question-answer space into multiple sequential presentations to model two safety-centric behaviors: conviction (i.e., sticking to a correct answer selection or abstention against incorrect suggestions) and flexibility (i.e., switching to a correct suggestion when it is introduced). Evaluating 17 LLMs across three clinical benchmarks, we observe a pervasive conversation tax, where partitioning an answer-space into sequential presentations reduces end-to-end accuracy and abstention against incorrect suggestions by an average of up to 30%, reaching 65% in certain models. We also observe blind switching, where models transition an initial abstention to incorrect and correct suggestions at near-identical rates reaching 50%. Finally, we show that increasing model scale mitigates some of these conversational inefficacies while exacerbating others, such as a higher propensity to adopt an incorrect suggestion from an initial abstention. Together our findings demonstrate that the general proficiency captured by static benchmarks do not translate over multi-turn dialogues.

cs.CL↗

arXiv2026

Coverage-Controlled Preference Mining from Noisy Claim Verification for Evidence-Grounded Generation

Evidence-grounded generation produces summaries whose claims should be supported by supplied evidence, but claim-level verifiers provide noisy feedback and can reward models that simply say less. We study this problem in clinical Brief Hospital Course summarization, where outputs must remain grounded in patient-specific EHR evidence. We introduce VERI-DPO, a preference-mining framework that converts noisy claim verification into coverage-controlled summary-level preferences. For each evidence-window prompt, VERI-DPO samples multiple candidate summaries, decomposes them into claims, verifies each claim against patient evidence, and forms a preference pair only when the chosen summary has better aggregate verifier-estimated support while retaining comparable verifiable content. Standard Direct Preference Optimization then distills these pairs into a single-sample policy, avoiding inference-time reranking. On patient-disjoint MIMIC-III-Ext-VeriFact-BHC test data, VERI-DPO reduces Not Supported rates from 10.7% to 1.9% under the mining verifier and from 11.6% to 6.4% under a separately prompted GPT-4o judge. In 100 blinded pairwise assessments by two domain researchers, VERI-DPO is preferred over the base model 56 times versus 18 times for factual faithfulness. In a locked zero-shot MIMIC-IV-Ext-BHC transfer test with 1,000 patients and no model adaptation, VERI-DPO lowers Not Supported rates with nearly unchanged scored-claim counts. Multi-seed ablations show that verifier-guided pair construction drives the gains, while coverage and anti-degeneration controls prevent apparent factuality improvements from coming from shorter or less checkable outputs.

cs.CL↗

arXiv2026

Disentangling Prompt Element Level Risk Factors for Hallucinations and Omissions in Mental Health LLM Responses

Mental health concerns are often expressed outside clinical settings, including in high-distress help seeking, where safety-critical guidance may be needed. Consumer health informatics systems increasingly incorporate large language models (LLMs) for mental health question answering, yet many evaluations underrepresent narrative, high-distress inquiries. We introduce UTCO (User, Topic, Context, Tone), a prompt construction framework that represents an inquiry as four controllable elements for systematic stress testing. Using 2,075 UTCO-generated prompts, we evaluated Llama 3.3 and annotated hallucinations (fabricated or incorrect clinical content) and omissions (missing clinically necessary or safety-critical guidance). Hallucinations occurred in 6.5% of responses and omissions in 13.2%, with omissions concentrated in crisis and suicidal ideation prompts. Across regression, element-specific matching, and similarity-matched comparisons, failures were most consistently associated with context and tone, while user-background indicators showed no systematic differences after balancing. These findings support evaluating omissions as a primary safety outcome and moving beyond static benchmark question sets.

cs.CL↗

arXiv2026

Learning When to Sample: Confidence-Aware Selective Sampling for Efficient Chain-of-Thought Reasoning

Large language models (LLMs) can achieve strong reasoning performance through chain-of-thought (CoT) reasoning, yet they often generate unnecessarily long reasoning paths that incur high inference cost. Self-consistency-based approaches push accuracy higher still, but they require sampling and aggregating multiple reasoning trajectories, leading to substantial computational overhead. In this paper, we introduce a confidence-aware selective sampling framework that, at inference time, analyzes a single reasoning trajectory to adaptively determine whether to rely on that trajectory alone or trigger multi-path sampling. The framework uses trajectory-level numeric features and sentence-level linguistic features extracted from reasoning states to guide selective multi-path reasoning. We train it on MedQA and evaluate it in-domain on MedQA and under calibration-only transfer on MathQA, MedMCQA, and MMLU, without further fine-tuning. Experimental results show that the proposed framework maintains comparable performance to full and efficient multi-path reasoning baselines, with accuracy changes of $-0.41 \pm 0.58$ and $-0.31 \pm 0.58$ percentage points, respectively, while reducing token usage by $71.7 \pm 5.0%$ and $36.6 \pm 9.1%$. These findings demonstrate that reasoning trajectories contain rich signals for uncertainty estimation, enabling a simple, transferable mechanism to balance accuracy and efficiency in LLM reasoning.

cs.CL↗

arXiv2026

Optimizing Domain-Adaptive Self-Supervised Learning for Clinical Voice-Based Disease Classification

The human voice is a promising non-invasive digital biomarker, yet deep learning for voice-based health analysis is hindered by data scarcity and domain mismatch, where models pre-trained on general audio fail to capture the subtle pathological features characteristic of clinical voice data. To address these challenges, we investigate domain-adaptive self-supervised learning (SSL) with Masked Autoencoders (MAE) and demonstrate that standard configurations are suboptimal for health-related audio. Using the Bridge2AI-Voice dataset, a multi-institutional collection of pathological voices, we systematically examine three performance-critical factors: reconstruction loss (Mean Absolute Error vs. Mean Squared Error), normalization (patch-wise vs. global), and masking (random vs. content-aware). Our optimized design, which combines Mean Absolute Error (MA-Error) loss, patch-wise normalization, and content-aware masking, achieves a Macro F1 of $0.688 \pm 0.009$ (over 10 fine-tuning runs), outperforming a strong out-of-domain SSL baseline pre-trained on large-scale general audio, which has a Macro F1 of $0.663 \pm 0.011$. The results show that MA-Error loss improves robustness and content-aware masking boosts performance by emphasizing information-rich regions. These findings highlight the importance of component-level optimization in data-constrained medical applications that rely on audio data.

eess.AS↗