Author name cluster

Che Lin

Possible papers associated with this exact author name in Arrow. This page groups case-insensitive exact name matches and is not a full identity disambiguation profile.

7 papers

2 author rows

JBHI Journal 2026 Journal Article

Beyond Curated Knowledge: Structural Protein Embeddings Enhance GNN-Based Personalized Cancer Prognosis

Sofia Ormazabal Arriagada
Tsung-Wei Lin
Marta Misztal
Che Lin

Accurate and actionable prognostic models can meaningfully influence follow-up scheduling, therapeu tic prioritization, and resource allocation in oncology. We propose GLLM, a multimodal graph learning framework that integrates RNA-seq profiles, routine clinical variables, and structural protein embeddings derived from protein language models to stratify patients by 5-year risk across multiple cancer types. Each gene is represented as a node within a protein–protein interaction graph, and we intro duce SCANE, a fusion mechanism that modulates each gene's structural embedding using patient-specific expression values. This design enables the graph neural network to propagate expression-conditioned molecular sig nals while preserving the underlying biophysical context. Across breast cancer, lung adenocarcinoma, and colorectal cancer cohorts, GLLM improves the area under the precision–recall curve relative to strong clinical and molecular baselines, while maintaining competitive concordance indices. The contributions of this work include: (1) an effective fusion strategy that enhances node representations by combining protein structural embeddings with gene expression for improved risk prediction; (2) a sys tematic evaluation demonstrating that sequence-derived structural embeddings outperform text-based biomedical embeddings; and (3) patient-level interpretability analyses showing that the model highlights established biomarkers and aligns with perturbation-based sensitivity profiles. Clinical significance: GLLM supports personalized surveillance planning by identifying high-risk patients who may benefit from earlier imaging, shorter follow-up inter vals, or prioritization for treatment discussions and clinical trial screening. Its lightweight architecture (<7 MFLOPs) enables seamless integration into existing oncology work f lows without additional computational burden. The result ing risk score is designed to complement, rather than replace, mutation profiling and clinicopathological staging, reflecting the biological and operational heterogeneity across cancer types.

Details DOI

AAAI Conference 2026 Conference Paper

HINPool: A Unified Heterogeneous Graph Pooling Framework for Accurate Molecular and Protein Property Prediction

Ming-Yi Hong
You-Chen Teng
Shao-En Lin
Chih-Yu Wang
Che Lin

Graph pooling has gained significant progress in recent years as an effective solution for graph-level property classification tasks. With the emergence of research on Heterogeneous Information Networks (HINs), this paper argues that graph-level datasets for graph classification should be treated as HINs rather than homogeneous graphs to enhance information aggregation. We propose HINPool, a novel and general graph pooling framework for graph-level property classification with HINs. First, we devise a systematic HIN construction procedure from the original data to capture complex interactions. Next, we introduce a type-aware heterogeneous graph pooling method featuring a Type-Aware Selector (TAS) to select essential nodes and a Readout Aggregator (RA) to fuse critical information into a graph-level representation. Finally, a cross-layer fusion function is applied to combine the output embeddings from each graph pooling layer, creating a final graph representation for downstream classification tasks. Our approach achieves near state-of-the-art performance on widely used graph classification benchmark datasets, demonstrating significant improvements in four out of five datasets. This work redefines the strategy for graph-level property classification with HGNNs and heterogeneous graph pooling to model intricate relationships, enhancing performance without requiring extensive domain-specific knowledge.

PDF Details DOI

AAMAS Conference 2026 Conference Paper

Learning Robust Policy for Multi-UAV Collision Avoidance via Compact Causal Feature

Zhun Fan
Gaofei Han
Che Lin
Wenji Li
Jie Xu
Jiafan Zhuang

Deepreinforcementlearning(DRL)-basedmulti-UAVcollisionavoidance methods often exhibit limited generalization when deployed in unseen environments, primarily due to the reliance on noncausal and redundant visual features. Such overfitting to spurious correlations compromises both robustness and safety during realworld deployment. To address these limitations, this study proposes a novel Compact Causal Feature Learning (CCFL) framework that enables UAVs to learn compact and generalizable causal representations. Specifically, a Causal Feature Identification module is designed to disentangle input representations into causal and noncausalcomponents, ensuringthatthelearnedfeaturespreservetrue environmental causality. Furthermore, a Redundancy Feature Compression module is introduced to remove redundant dependencies and compact the causal subspace, thereby enhancing generalization to previously unseen scenarios. Extensive experiments on a challenging UAV collision avoidance benchmark demonstrate that CCFL achieves substantial performance gains over state-of-the-art baselines, increasing individual success rates by 42. 0% and swarm success rates by 61. 6%. These results validate the effectiveness of compact causal feature learning for improving the adaptability, robustness, and safety of autonomous UAV systems operating in complex dynamic environments.

PDF

AAAI Conference 2026 Conference Paper

MM4Rec: Multi-Source and Multi-Scenario Recommender for Unified User Preference

Chu-Chun Yu
Ming-Yi Hong
Miao-Chen Chiang
Min Chen Hsieh
Che Lin

As online ecosystems grow increasingly complex, personalized recommendation systems must integrate user preferences across heterogeneous content sources and interaction scenarios. However, conventional methods typically model each source and scenario in isolation, hindering their ability to capture shared and complementary signals across contexts. In this work, we propose MM4Rec, a unified framework for multi-source and multi-scenario recommendation. MM4Rec introduces a Source-Aware Transformer Encoder to jointly model heterogeneous inputs, a Multi-Scenario Behavior Extraction Layer based on a multi-mixture-of-experts architecture to capture scenario-specific dynamics, and a Trend-Aware Learner to enhance temporal representation learning. Extensive experiments on three real-world datasets demonstrate that MM4Rec consistently outperforms strong baselines across standard recommendation metrics. To facilitate future research, we also release two large-scale datasets encompassing diverse sources and scenarios.

PDF Details DOI

ICML Conference 2025 Conference Paper

MTSTRec: Multimodal Time-Aligned Shared Token Recommender

Ming-Yi Hong 0002
Yen-Jung Hsu
Miao-Chen Chiang
Che Lin

Sequential recommendation in e-commerce utilizes users’ anonymous browsing histories to personalize product suggestions without relying on private information. Existing item ID-based methods and multimodal models often overlook the temporal alignment of modalities like textual descriptions, visual content, and prices in user browsing sequences. To address this limitation, this paper proposes the Multimodal Time-aligned Shared Token Recommender (MTSTRec), a transformer-based framework with a single time-aligned shared token per product for efficient cross-modality fusion. MTSTRec preserves the distinct contributions of each modality while aligning them temporally to better capture user preferences. Extensive experiments demonstrate that MTSTRec achieves state-of-the-art performance across multiple sequential recommendation benchmarks, significantly improving upon existing multimodal fusion. Our code is available at https: //github. com/idssplab/MTSTRec.

Details

AAMAS Conference 2025 Conference Paper

Robust Policy Learning for Multi-UAV Collision Avoidance with Causal Feature Selection

Jiafan Zhuang
Gaofei Han
Zihao Xia
Che Lin
Boxi Wang
Dongliang Wang
Wenji Li
Zhifeng Hao

Collision avoidance navigation for unmanned aerial vehicle (UAV) swarms in complex and unseen outdoor environments presents a significant challenge, as UAVs are required navigate through various obstacles and intricate backgrounds. While existing deep reinforcement learning (DRL)-based collision avoidance methods have shown promising performance, they often suffer from poor generalization, leading to degraded performance in unseen environments. To address this limitation, we investigate the root causes of weak generalization in DRL models and propose a novel causal feature selection module. This module can be integrated into the policy network to effectively filter out non-causal factors in representations, thereby minimizing the impact of spurious correlations between non-causal elements and action predictions. Experimental results demonstrate that the proposed method achieves robust navigation performance and effective collision avoidance, particularly in scenarios with unseen backgrounds and obstacles, which significantly outperforms state-of-the-art (SOTA) algorithms.

PDF

JBHI Journal 2024 Journal Article

Multimodal Attention Network for Dementia Prediction

Hsinhan Tsai
Ta-Wei Yang
Kai-Hao Ou
Tung-Hung Su
Che Lin
Cheng-Fu Chou

The early identification of an individual's dementia risk is crucial for disease prevention and the design of insurance products in an aging society. This study aims to accurately predict the future incidence risk of dementia in individuals by leveraging the advantages of neural networks. This is, however, complicated by the high dimensionality and sparsity of the International Classification of Diseases (ICD) codes when utilizing data from Taiwan's National Health Insurance, which includes individual profiles and medical records. Inspired by the click-through rate (CTR) problem in recommendation systems, where future user behavior is predicted based on their past consumption records, we address these challenges with a multimodal attention network for dementia (MAND), which incorporates an ICD code embedding layer and multihead self-attention to encode ICD codes and capture interactions among diseases. Additionally, we investigate the applicability of several CTR methods to the dementia prediction problem. MAND achieves an AUC of 0. 9010, surpassing traditional CTR models and demonstrating its effectiveness. The highly flexible pipelined design allows for module replacement to meet specific requirements. Furthermore, the analysis of attention scores reveals diseases highly correlated with dementia, aligning with prior research and emphasizing the interpretability of the model. This research deepens our understanding of the diseases associated with dementia, and the accurate prediction provided can serve as an early warning for dementia occurrence, aiding in its prevention.

Details DOI