Selected work

Representative Papers

A rule based solution to co-reference resolution in clinical text

Oct 11, 2012J. Am. Medical Informatics Assoc.

This paper addresses coreference resolution of biomedical concept mentions in clinical text. Unlike general-purpose coreference models, which suffer from poor domain generalizability, the proposed method introduces a deeply customized, rule-driven approach grounded in linguistic principles and empirical pattern analysis of training data. It constructs a multi-layered, handcrafted rule system integrating exact string matching, semantic constraints (e.g., ontological type compatibility), and contextual consistency checks. Evaluated on the 2011 i2b2 multi-center clinical dataset, the system achieves an overall F1-score of 89.6%, substantially outperforming contemporary machine learning baselines. The primary contribution is the first interpretable, high-precision rule-based framework specifically designed for clinical text—balancing domain specificity with transparent, human-verifiable inference logic. This work establishes an effective paradigm for coreference resolution in low-resource, specialized domains where labeled data is scarce and model interpretability is critical.

4 citationsRead paper

Attention when you need

Jan 13, 2025

This study investigates how mice optimize decision efficiency by balancing attentional costs against benefits during an auditory sustained-attention–value task. Method: We developed a normative reinforcement learning model integrating behavioral analysis, optimal control theory, and dynamic accumulation of sensory evidence. Contribution/Results: We propose, for the first time, that attentional resources are deployed in rhythmic, alternating high–low blocks—a “blockwise” allocation strategy. Attentional policy is jointly determined by task utility, stimulus statistics, and attentional gain modulation of sensory evidence; low-attention states fully suppress sensory input, while high-attention states activate periodically. The model successfully reproduces and explains mice’s attentional allocation patterns across varying trial durations and reward contingencies. Our work establishes a novel theoretical framework—grounded in economic principles—for understanding attentional resource optimization and provides empirical support for rhythmically gated, cost-sensitive attentional control.

2 citationsRead paper

Physics-Informed Deep Learning for False Ventricular Tachycardia Alarm Reduction in the ICU

Sep 08, 2026

该研究使用结合1D SE-ResNet、数据增强和基于Windkessel模型的物理信息重建任务的深度学习框架，有效减少ICU中假性室性心动过速警报。

0 citationsRead paper

Institution-Specific LLM Prompting Recovers PHI That De-identification Systems and Their Gold Standards Both Miss

Aug 17, 2026

研究利用大型语言模型和上下文学习解决电子健康记录中机构特定的受保护健康信息脱敏问题，通过不同提示策略提高召回率和精确度。

0 citationsRead paper

Thinking Under Uncertainty: Evidence Use and Information-Seeking in Language Models

Jul 29, 2026

This study investigates whether large language models, during reasoning, more effectively utilize existing evidence or actively seek new information to optimize decisions. Using a two-armed bandit task under uncertainty, the authors compare model behavior in “thinking” versus “non-thinking” modes, integrating cognitive modeling with decoder parameter sweeps to disentangle metacognitive monitoring and control signals in a controlled setting for the first time. Results show that thinking primarily enhances value-guided action selection, reduces choice noise unrelated to uncertainty, and increases confidence sensitivity to task difficulty and evidence strength, but does not significantly promote information seeking. The analysis further identifies both UCB-like and Thompson-like exploration strategies, revealing that thinking optimizes the exploitation of current evidence rather than driving active exploration.

0 citationsRead paper

Recent publications

Latest Papers

Physics-Informed Deep Learning for False Ventricular Tachycardia Alarm Reduction in the ICU

Sep 08, 2026

该研究使用结合1D SE-ResNet、数据增强和基于Windkessel模型的物理信息重建任务的深度学习框架，有效减少ICU中假性室性心动过速警报。

0 citationsRead paper

Institution-Specific LLM Prompting Recovers PHI That De-identification Systems and Their Gold Standards Both Miss

Aug 17, 2026

研究利用大型语言模型和上下文学习解决电子健康记录中机构特定的受保护健康信息脱敏问题，通过不同提示策略提高召回率和精确度。

0 citationsRead paper

Thinking Under Uncertainty: Evidence Use and Information-Seeking in Language Models

Jul 29, 2026

This study investigates whether large language models, during reasoning, more effectively utilize existing evidence or actively seek new information to optimize decisions. Using a two-armed bandit task under uncertainty, the authors compare model behavior in “thinking” versus “non-thinking” modes, integrating cognitive modeling with decoder parameter sweeps to disentangle metacognitive monitoring and control signals in a controlled setting for the first time. Results show that thinking primarily enhances value-guided action selection, reduces choice noise unrelated to uncertainty, and increases confidence sensitivity to task difficulty and evidence strength, but does not significantly promote information seeking. The analysis further identifies both UCB-like and Thompson-like exploration strategies, revealing that thinking optimizes the exploitation of current evidence rather than driving active exploration.

0 citationsRead paper

Interpretable EEG biomarkers with bag-of-waves: Spatial and temporal waveform dictionaries for low-data regimes

Jul 24, 2026

Traditional EEG analysis is hindered by strong prior biases from predefined spectral features or the data dependency and lack of interpretability inherent in deep models, particularly under low-data conditions. This work proposes a bag-of-waves framework that learns a small, interpretable dictionary of EEG waveform atoms via unsupervised, translation-invariant k-means, converting continuous signals into symbolic sequences. Temporal structure is captured through n-gram modeling, while spatial information is incorporated using single-channel, regional, and cross-channel atomic representations. For the first time, this approach integrates an interpretable waveform dictionary, n-gram temporal modeling, and multi-channel spatial extension, achieving performance on par with state-of-the-art deep models under extremely limited data while using significantly fewer parameters. The method explicitly recovers neurophysiologically verifiable canonical waveforms across three datasets: mouse genotype clustering, dementia resting-state classification, and six-class clinical event detection on TUEG.

0 citationsRead paper

Frequency Selection in Bayesian Spectral Modeling of Time Series Data with Applications to Wearable Device Measurements

Jul 16, 2026

This study addresses the challenge of extracting high-resolution rhythmic components from wearable-device time series by proposing a Bayesian spike-and-slab sparse modeling framework. The method jointly performs frequency selection and dimensionality reduction over a fine frequency grid, incorporating structured priors to encourage sparsity and extending via hierarchical modeling to multivariate signals to identify both shared and modality-specific physiological rhythms. Coupled with a stochastic search posterior inference algorithm, the model yields accurate and interpretable spectral estimates in both univariate and multivariate settings. Evaluated on simulated data and real-world wearable recordings—including actigraphy from epilepsy patients and synchronized activity–core temperature measurements from healthy individuals—the approach significantly outperforms existing methods, faithfully recovering circadian and ultradian rhythms and uncovering cross-modal physiological coupling mechanisms.

0 citationsRead paper