Zhenhao Zhang

Shuheng Cao, Ruiqi Chen, Renjie Cao et al.

Biomedical NER is deceptively simple for modern LLMs: plausible biomedical mentions are easy to surface, but corpus-convention correctness depends on annotation conventions, span boundaries, entity granularity, and type schemas. Multi-LLM agreement is a salience signal, not corpus-convention correctness. We introduce a candidate-level panel-output benchmark for panel-surfaced candidate verification, where the unit is an aligned candidate surfaced by an explicitly defined multi-model panel rather than a standalone extractor output. The benchmark aligns eight LLMs' predictions over five public biomedical NER datasets into a candidate master table. BioConCal is an in-domain supervised scorer that instantiates this layer with inference-time gold-free agreement, mention, surface-availability, and document features for a fixed candidate stream. In domain, BioConCal improves AUROC from 0.753 for raw agreement to 0.910. At a validation-selected 0.95 precision target it selects 1,340 candidates at empirical test precision 0.939, compared with 293 for raw agreement. This corresponds to candidate-level recall 0.592 and corpus-level recall 0.523 against a within-panel row-label ceiling of 0.883. The main benefit is not recovering entities missed by every panel member, but reshaping a noisy panel stream into a higher-yield review queue. Under entity-type shift, thresholds require target-domain validation, and exact character localization remains a separate deterministic post-processing step.

Xiangzhe Yuan, Zhenhao Zhang, Haoming Tang et al.

As LLMs gain persuasive capabilities through extended dialogues, they create new opportunities for studying adversarial conversational behavior in extended interaction settings that traditional single-turn safety evaluations fail to capture. We systematically study these interactional dynamics using a controlled LLM-to-LLM simulation framework for automated red-teaming across bilingual social engineering scenarios. Evaluating eight state-of-the-art models in English and Chinese, we analyze dialogue-level outcomes, annotate attacker and defender strategy families, and model interaction dynamics between them. Results show that multi-turn adversarial dialogues follow recurrent escalation patterns, while defensive responses frequently rely on verification, delay, and channel control. We further find statistically significant cross-model and cross-lingual differences in outcome distributions, and transition analysis reveals systematic structural variation in how defender strategies respond to attacker tactics across languages. These findings highlight the importance of studying interactional structure in multi-turn adversarial dialogue settings and demonstrate how controlled LLM-to-LLM simulations can support mechanistic analysis of adversarial conversational dynamics.

Yuxi Liu, Zhenhao Zhang, Antonio Jimeno Yepes et al.

Building models for health prediction based on Electronic Health Records (EHR) has become an active research area. EHR patient journey data consists of patient time-ordered clinical events/visits from patients. Most existing studies focus on modeling long-term dependencies between visits, without explicitly taking short-term correlations between consecutive visits into account, where irregular time intervals, incorporated as auxiliary information, are fed into health prediction models to capture latent progressive patterns of patient journeys. We present a novel deep neural network with four modules to take into account the contributions of various variables for health prediction: i) the Stacked Attention module strengthens the deep semantics in clinical events within each patient journey and generates visit embeddings, ii) the Short-Term Temporal Attention module models short-term correlations between consecutive visit embeddings while capturing the impact of time intervals within those visit embeddings, iii) the Long-Term Temporal Attention module models long-term dependencies between visit embeddings while capturing the impact of time intervals within those visit embeddings, iv) and finally, the Coupled Attention module adaptively aggregates the outputs of Short-Term Temporal Attention and Long-Term Temporal Attention modules to make health predictions. Experimental results on MIMIC-III demonstrate superior predictive accuracy of our model compared to existing state-of-the-art methods, as well as the interpretability and robustness of this approach. Furthermore, we found that modeling short-term correlations contributes to local priors generation, leading to improved predictive modeling of patient journeys.

Yuxi Liu, Zhenhao Zhang, Shaowen Qin et al.

Predicting the risk of in-hospital mortality from electronic health records (EHRs) has received considerable attention. Such predictions will provide early warning of a patient's health condition to healthcare professionals so that timely interventions can be taken. This prediction task is challenging since EHR data are intrinsically irregular, with not only many missing values but also varying time intervals between medical records. Existing approaches focus on exploiting the variable correlations in patient medical records to impute missing values and establishing time-decay mechanisms to deal with such irregularity. This paper presents a novel contrastive learning-based imputation-prediction network for predicting in-hospital mortality risks using EHR data. Our approach introduces graph analysis-based patient stratification modeling in the imputation process to group similar patients. This allows information of similar patients only to be used, in addition to personal contextual information, for missing value imputation. Moreover, our approach can integrate contrastive learning into the proposed network architecture to enhance patient representation learning and predictive performance on the classification task. Experiments on two real-world EHR datasets show that our approach outperforms the state-of-the-art approaches in both imputation and prediction tasks.

Yuxi Liu, Zhenhao Zhang, Shaowen Qin et al.

The Intensive Care Unit (ICU) is one of the most important parts of a hospital, which admits critically ill patients and provides continuous monitoring and treatment. Various patient outcome prediction methods have been attempted to assist healthcare professionals in clinical decision-making. Existing methods focus on measuring the similarity between patients using deep neural networks to capture the hidden feature structures. However, the higher-order relationships are ignored, such as patient characteristics (e.g., diagnosis codes) and their causal effects on downstream clinical predictions. In this paper, we propose a novel Hypergraph Convolutional Network that allows the representation of non-pairwise relationships among diagnosis codes in a hypergraph to capture the hidden feature structures so that fine-grained patient similarity can be calculated for personalized mortality risk prediction. Evaluation using a publicly available eICU Collaborative Research Database indicates that our method achieves superior performance over the state-of-the-art models on mortality risk prediction. Moreover, the results of several case studies demonstrated the effectiveness and robustness of the model decisions.

Yuxi Liu, Shaowen Qin, Antonio Jimeno Yepes et al.

Predicting the health risks of patients using Electronic Health Records (EHR) has attracted considerable attention in recent years, especially with the development of deep learning techniques. Health risk refers to the probability of the occurrence of a specific health outcome for a specific patient. The predicted risks can be used to support decision-making by healthcare professionals. EHRs are structured patient journey data. Each patient journey contains a chronological set of clinical events, and within each clinical event, there is a set of clinical/medical activities. Due to variations of patient conditions and treatment needs, EHR patient journey data has an inherently high degree of missingness that contains important information affecting relationships among variables, including time. Existing deep learning-based models generate imputed values for missing values when learning the relationships. However, imputed data in EHR patient journey data may distort the clinical meaning of the original EHR patient journey data, resulting in classification bias. This paper proposes a novel end-to-end approach to modeling EHR patient journey data with Integrated Convolutional and Recurrent Neural Networks. Our model can capture both long- and short-term temporal patterns within each patient journey and effectively handle the high degree of missingness in EHR data without any imputation data generation. Extensive experimental results using the proposed model on two real-world datasets demonstrate robust performance as well as superior prediction accuracy compared to existing state-of-the-art imputation-based prediction methods.

Yuxi Liu, Shaowen Qin, Zhenhao Zhang et al.

Electronic Health Records (EHRs) exhibit a high amount of missing data due to variations of patient conditions and treatment needs. Imputation of missing values has been considered an effective approach to deal with this challenge. Existing work separates imputation method and prediction model as two independent parts of an EHR-based machine learning system. We propose an integrated end-to-end approach by utilizing a Compound Density Network (CDNet) that allows the imputation method and prediction model to be tuned together within a single framework. CDNet consists of a Gated recurrent unit (GRU), a Mixture Density Network (MDN), and a Regularized Attention Network (RAN). The GRU is used as a latent variable model to model EHR data. The MDN is designed to sample latent variables generated by GRU. The RAN serves as a regularizer for less reliable imputed values. The architecture of CDNet enables GRU and MDN to iteratively leverage the output of each other to impute missing values, leading to a more accurate and robust prediction. We validate CDNet on the mortality prediction task on the MIMIC-III dataset. Our model outperforms state-of-the-art models by significant margins. We also empirically show that regularizing imputed values is a key factor for superior prediction performance. Analysis of prediction uncertainty shows that our model can capture both aleatoric and epistemic uncertainties, which offers model users a better understanding of the model results.

Hanqing Wang, Yuan Tian, Mingyu Liu et al.

In the rapidly evolving landscape of Multimodal Large Language Models (MLLMs), the safety concerns of their outputs have earned significant attention. Although numerous datasets have been proposed, they may become outdated with MLLM advancements and are susceptible to data contamination issues. To address these problems, we propose \textbf{SDEval}, the \textit{first} safety dynamic evaluation framework to controllably adjust the distribution and complexity of safety benchmarks. Specifically, SDEval mainly adopts three dynamic strategies: text, image, and text-image dynamics to generate new samples from original benchmarks. We first explore the individual effects of text and image dynamics on model safety. Then, we find that injecting text dynamics into images can further impact safety, and conversely, injecting image dynamics into text also leads to safety risks. SDEval is general enough to be applied to various existing safety and even capability benchmarks. Experiments across safety benchmarks, MLLMGuard and VLSBench, and capability benchmarks, MMBench and MMVet, show that SDEval significantly influences safety evaluation, mitigates data contamination, and exposes safety limitations of MLLMs. Code is available at https://github.com/hq-King/SDEval

Zijian Gu, Yuxi Liu, Zhenhao Zhang et al.

Vision-language models achieve expert-level performance on medical imaging tasks but exhibit significant diagnostic accuracy disparities across demographic groups. We introduce fairness-aware Low-Rank Adaptation for medical VLMs, combining parameter efficiency with explicit fairness optimization. Our key algorithmic contribution is a differentiable MaxAccGap loss that enables end-to-end optimization of accuracy parity across demographic groups. We propose three methods: FR-LoRA integrates MaxAccGap regularization into the training objective, GR-LoRA applies inverse frequency weighting to balance gradient contributions, and Hybrid-LoRA combines both mechanisms.Evaluated on 10,000 glaucoma fundus images, GR-LoRA reduces diagnostic accuracy disparities by 69% while maintaining 53.15% overall accuracy. Ablation studies reveal that strong regularization strength achieves optimal fairness with minimal accuracy trade-off, and race-specific optimization yields 60% disparity reduction. Our approach requires only 0.24% trainable parameters, enabling practical deployment of fair medical AI in resource-constrained healthcare settings.

Yang Wan, Zheng Cao, Zhenhao Zhang et al.

Large language models excel as few-shot learners when provided with appropriate demonstrations, yet this strength becomes problematic in multiturn agent scenarios, where LLMs erroneously mimic their own previous responses as few-shot examples. Through attention analysis, we identify conversational inertia, a phenomenon where models exhibit strong diagonal attention to previous responses, which is associated with imitation bias that constrains exploration. This reveals a tension when transforming few-shot LLMs into agents: longer context enriches environmental feedback for exploitation, yet also amplifies conversational inertia that undermines exploration. Our key insight is that for identical states, actions generated with longer contexts exhibit stronger inertia than those with shorter contexts, enabling construction of preference pairs without environment rewards. Based on this, we propose Context Preference Learning to calibrate model preferences to favor low-inertia responses over highinertia ones. We further provide context management strategies at inference time to balance exploration and exploitation. Experimental results across eight agentic environments and one deep research scenario validate that our framework reduces conversational inertia and achieves performance improvements.

Siying Hu, Zhenhao Zhang

Sedentary behavior is a critical health risk for older adults. Although digital interventions are widely available, they primarily rely on screen-based notifications that can feel clinical or cognitively demanding, and are thus often ignored over time. This paper presents a three-phase Research through Design methodology to explore data physicalization approaches that ambiently represent sedentary data patterns using decor artifacts in older adults' homes. These artifacts transformed abstract data into aesthetic, evolving forms that became part of the domestic landscape. Our research revealed how these physicalizations fostered self-reflection, family conversations, and encouraged active lifestyles. We demonstrate how qualities like aesthetic ambiguity and slow revelation can empower older adults, fostering a reflective relationship with their well-being. Ultimately, we argue that creating data physicalizations for older adults necessitates a shift from merely informing users to enabling them to live with and through their data.

Zhenhao Zhang, Ye Shi, Lingxiao Yang et al.

Understanding and synthesizing realistic 3D hand-object interactions (HOI) is critical for applications ranging from immersive AR/VR to dexterous robotics. Existing methods struggle with generalization, performing well on closed-set objects and predefined tasks but failing to handle unseen objects or open-vocabulary instructions. We introduce OpenHOI, the first framework for open-world HOI synthesis, capable of generating long-horizon manipulation sequences for novel objects guided by free-form language commands. Our approach integrates a 3D Multimodal Large Language Model (MLLM) fine-tuned for joint affordance grounding and semantic task decomposition, enabling precise localization of interaction regions (e.g., handles, buttons) and breakdown of complex instructions (e.g., "Find a water bottle and take a sip") into executable sub-tasks. To synthesize physically plausible interactions, we propose an affordance-driven diffusion model paired with a training-free physics refinement stage that minimizes penetration and optimizes affordance alignment. Evaluations across diverse scenarios demonstrate OpenHOI's superiority over state-of-the-art methods in generalizing to novel object categories, multi-stage tasks, and complex language instructions. Our project page at \href{https://openhoi.github.io}

Hanqing Wang, Zhenhao Zhang, Kaiyang Ji et al.

3D object affordance grounding aims to predict the touchable regions on a 3d object, which is crucial for human-object interaction, human-robot interaction, embodied perception, and robot learning. Recent advances tackle this problem via learning from demonstration images. However, these methods fail to capture the general affordance knowledge within the image, leading to poor generalization. To address this issue, we propose to use text-to-image diffusion models to extract the general affordance knowledge because we find that such models can generate semantically valid HOI images, which demonstrate that their internal representation space is highly correlated with real-world affordance concepts. Specifically, we introduce the DAG, a diffusion-based 3d affordance grounding framework, which leverages the frozen internal representations of the text-to-image diffusion model and unlocks affordance knowledge within the diffusion model to perform 3D affordance grounding. We further introduce an affordance block and a multi-source affordance decoder to endow 3D dense affordance prediction. Extensive experimental evaluations show that our model excels over well-established methods and exhibits open-world generalization.

Peiqi He, Zhenhao Zhang, Yixiang Zhang et al.

Precise spatial modeling in the operating room (OR) is foundational to many clinical tasks, supporting intraoperative awareness, hazard avoidance, and surgical decision-making. While existing approaches leverage large-scale multimodal datasets for latent-space alignment to implicitly learn spatial relationships, they overlook the 3D capabilities of MLLMs. However, this approach raises two issues: (1) Operating rooms typically lack multiple video and audio sensors, making multimodal 3D data difficult to obtain; (2) Training solely on readily available 2D data fails to capture fine-grained details in complex scenes. To address this gap, we introduce Spatial-ORMLLM, the first large vision-language model for 3D spatial reasoning in operating rooms using only RGB modality to infer volumetric and semantic cues, enabling downstream medical tasks with detailed and holistic spatial context. Spatial-ORMLLM incorporates a Spatial-Enhanced Feature Fusion Block, which integrates 2D modality inputs with rich 3D spatial knowledge extracted by the estimation algorithm and then feeds the combined features into the visual tower. By employing a unified end-to-end MLLM framework, it combines powerful spatial features with textual features to deliver robust 3D scene reasoning without any additional expert annotations or sensor inputs. Experiments on multiple benchmark clinical datasets demonstrate that Spatial-ORMLLM achieves state-of-the-art performance and generalizes robustly to previously unseen surgical scenarios and downstream tasks.

Siying Hu, Zhenhao Zhang

Workplace stress is often addressed through visual or auditory interventions, yet these modalities can compete with attention and contribute to sensory overload. We explore olfaction as an alternative ambient medium for representing stress-related physiological signals in office settings. We present AuraDesk, an olfactory data physicalization system that translates wearable-derived physiological cues into situated scent expressions at the workstation. The system combines local physiological state inference with a constrained actuation strategy to produce temporally regulated and spatially localized scent output suitable for everyday work environments. To examine the feasibility and experiential qualities of this approach, we conducted a one-day in-situ field deployment with 25 knowledge workers at their actual workstations. Our findings show that participants often interpreted the scent output not as an explicit alert, but as a subtle atmospheric cue that supported momentary awareness, micro-break taking, and perceived environmental attunement. At the same time, participants raised important concerns regarding scent preference, habituation, and contextual appropriateness in shared offices. This work contributes (1) an olfactory interface for physiologically driven ambient feedback in the workplace, (2) a hybrid mapping approach for coupling continuous biosignal interpretation with constrained scent actuation, and (3) empirical insights into how workers perceive, negotiate, and appropriate ambient olfactory feedback in real office contexts. Rather than claiming therapeutic efficacy, we position AuraDesk as a probe into the design space of olfactory data physicalization for workplace wellbeing and attention-sensitive interaction.

Zhenhao Zhang, Hanqing Wang, Xiangyu Zeng et al.

Understanding and recognizing human-object interaction (HOI) is a pivotal application in AR/VR and robotics. Recent open-vocabulary HOI detection approaches depend exclusively on large language models for richer textual prompts, neglecting their inherent 3D spatial understanding capabilities. To address this shortcoming, we introduce HOID-R1, the first HOI detection framework that integrates chain-of-thought (CoT) guided supervised fine-tuning (SFT) with group relative policy optimization (GRPO) within a reinforcement learning (RL) paradigm. Specifically, we initially apply SFT to imbue the model with essential reasoning capabilities, forcing the model to articulate its thought process in the output. Subsequently, we integrate GRPO to leverage multi-reward signals for policy optimization, thereby enhancing alignment across diverse modalities. To mitigate hallucinations in the CoT reasoning, we introduce an "MLLM-as-a-judge" mechanism that supervises the CoT outputs, further improving generalization. Extensive experiments show that HOID-R1 achieves state-of-the-art performance on HOI detection benchmarks and outperforms existing methods in open-world generalization to novel scenarios.