Cross-lingual Representation Learning via Centroid Intervention Fusion
Fuses multilingual intervention projections into a shared operator to improve cross-lingual transfer without updating model parameters.
Publications
Publications and research, including work in medical NLP, multilingual language models, and explainable AI.
See Google Scholar for citation information and the latest indexed records. Google Scholar
2026
Event Interaction in Low-Rank Bottlenecks for Temporal Relation Extraction
Models event-pair interactions inside low-rank bottlenecks for temporal relation extraction with Adapter and LoRA fine-tuning.
Weakly-supervised Argument Mining with Boundary Refinement and Relation Denoising
Refines component boundaries and denoises relations in LLM-generated labels, matching fully supervised argument mining with less than 20% of the human annotations.
2025
Mitigating Negative Interference in Multilingual Sequential Knowledge Editing through Null-Space Constraints
LangEdit constrains sequential knowledge updates in null spaces to reduce interference across languages and previously edited knowledge.
Generating Explanations in Medical Question-Answering by Expectation Maximization Inference over Evidence
Jointly learns evidence selection, answer prediction, and explanation generation using expectation maximization over medical evidence.
2024
DMON: A Simple Yet Effective Approach for Argument Structure Learning
A dual-tower, multi-scale convolutional network captures argument relations and document-level argumentative structure.
A Unified Review of Deep Learning for Automated Medical Coding
Organizes medical coding models into a unified framework of encoders, deep architectures, decoders, and auxiliary information.
Alleviating Exposure Bias in Diffusion Models through Sampling with Shifted Time Steps
Introduces a time-shift sampling method that reduces exposure bias in diffusion models with little additional computation.
ANTIDOTE: ArgumeNtaTIon-Driven explainable artificial intelligence fOr digiTal mEdicine
Using Unsupervised Topic Modeling to Uncover Document Hierarchy and Latent Topics in Prostate Cancer Clinical Texts
2023
Weak Supervision and Clustering-Based Sample Selection for Clinical Named Entity Recognition
Combines medical ontologies, domain adaptation, and representative sample selection for clinical NER with limited annotations.
Content Reduction, Surprisal and Information Density Estimation for Long Documents
Towards Explainability in Automated Medical Code Prediction from Clinical Records
2022
Multitask Balanced and Recalibrated Network for Medical Code Prediction
Extends multitask medical coding with feature recalibration and focal loss to address noisy clinical notes and imbalanced codes.
Extracting Medical Entities from Radiology Reports with Ontology-Based Distant Supervision
2021
Multitask Recalibrated Aggregation Network for Medical Code Prediction
MT-RAM jointly predicts ICD and CCS codes, sharing information across coding systems and aggregating features from lengthy clinical notes.