Author name cluster

Bing Song

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.

4 papers

2 author rows

AAAI Conference 2026 Conference Paper

EfficientFSL: Enhancing Few-Shot Classification via Query-Only Tuning In Vision Transformers

Wenwen Liao
Hang Ruan
Jianbo Yu
Bing Song
Yuansong Wang
Xiaofeng Yang

Large models such as Vision Transformers (ViTs) have demonstrated remarkable superiority over smaller architectures like ResNet in few-shot classification, owing to their powerful representational capacity. However, fine-tuning such large models demands extensive GPU memory and prolonged training time, making them impractical for many real-world low-resource scenarios. To bridge this gap, we propose EfficientFSL, a query-only fine-tuning framework tailored specifically for few-shot classification with ViT, which achieves competitive performance while significantly reducing computational overhead. EfficientFSL fully leverages the knowledge embedded in the pre-trained model and its strong comprehension ability, achieving high classification accuracy with an extremely small number of tunable parameters. Specifically, we introduce a lightweight trainable Forward Block to synthesize task-specific queries that extract informative features from the intermediate representations of the pre-trained model in a query-only manner. We further propose a Combine Block to fuse multi-layer outputs, enhancing the depth and robustness of feature representations. Finally, a Support-Query Attention Block mitigates distribution shift by adjusting prototypes to align with the query set distribution. With minimal trainable parameters, EfficientFSL achieves state-of-the-art performance on four in-domain few-shot datasets and six cross-domain datasets, demonstrating its effectiveness in real-world applications.

PDF Details DOI

JBHI Journal 2026 Journal Article

Graph-Enhanced Multi-Task Learning for Type 2 Diabetes Comorbidity Risk Prediction

Liyun Tang
Jiaxin Lu
Daohua Pan
Zhongfu Zuo
Xiqiao He
Haoqiang Zhang
Bing Song

Early prediction of Type 2 diabetes mellitus (T2DM) complications holds significant clinical importance for improving patient outcomes and reducing healthcare burden, yet existing prediction methods exhibit notable limitations. This paper proposes a Graph-Enhanced Multi-Task Learning (GEMTL) framework for simultaneously predicting the occurrence risk of multiple diabetic complications. The framework constructs disease relation graphs through a hybrid strategy that linearly combines a data-driven statistical graph derived from disease co-occurrence patterns with a knowledge-driven prior graph encoding clinically established association strengths, employs graph neural networks to capture higher-order dependencies among diseases, designs cross-attention mechanisms to achieve heterogeneous information fusion between patient features and disease graph embeddings, and utilizes multi-gating expert network architecture for task-specific modeling. Large-scale experimental validation was conducted on a dataset constructed from MIMIC-IV. Results demonstrate that the GEMTL framework achieves macro-averaged F1 score of 0. 723, micro-averaged F1 score of 0. 856, and mean Average Precision of 0. 759, significantly outperforming baseline methods across all evaluation metrics, including traditional machine learning methods, deep multi-task learning methods, graph neural network methods, and multi-expert architectures. This study provides an effective technical framework for complex medical multi-task prediction problems, with broad application prospects in diabetes precision management and clinical decision support.

Details DOI

JBHI Journal 2026 Journal Article

Patient-Centric Digital Twin Framework with Hybrid Knowledge Distillation for Federated Class-Incremental Learning in Precision Medicine

Hongwei Zheng
Yanbo Xu
Xiqiao He
Hongxiao Wang
Bing Song

Precision medicine requires patient-centric digital twins that continuously adapt to evolving disease taxonomies while collaborating across distributed health care institutions. However, federated learning in medical environments faces dual heterogeneity challenges: spatial divergence from non-identical patient populations across hospitals and temporal dynamics from continuously emerging disease categories. This paper proposes the digital twin-enabled precision medicine with a hybrid knowledge distillation framework that integrates adaptive patient classification loss, clinical semantic distillation loss, and biomarker attention distillation loss. The adaptive weighting mechanism dynamically adjusts preservation strength for different disease categories by computing gradient magnitudes during training, automatically providing stronger retention for categories exhibiting higher forget ting susceptibility while allowing stable categories to receive proportionally less emphasis. The hybrid approach simultaneously preserves soft-label disease relationships, intermediate convolutional feature patterns, and gradient based adaptive weighting across disease categories. Experimental results demonstrate consistent superiority over baseline methods. On the CIFAR100 benchmark, the frame work achieves 71. 05% accuracy, representing a 1. 87 per centage point improvement. Medical imaging evaluation on OrganAMNIST, OrganCMNIST, and OrganSMNIST datasets shows 2. 18%, 1. 92%, and 1. 95% accuracy gains, respectively. Validation using authentic clinical laboratory data encounters confirms 2. 54% accuracy improvement, establishing practical viability for real-world precision medicine deployment where continuous diagnostic knowledge expansion must occur without compromising patient privacy or historical disease recognition capabilities.

Details DOI

ICRA Conference 2020 Conference Paper

Automatic Snake Gait Generation Using Model Predictive Control

Emily Hannigan
Bing Song
Gagan Khandate
Maximilian Haas-Heger
Ji Yin
Matei Ciocarlie

In this paper, we propose a method for generating undulatory gaits for snake robots. Instead of starting from a pre-defined movement pattern such as a serpenoid curve, we use a Model Predictive Control (MPC) approach to automatically generate effective locomotion gaits via trajectory optimization. An important advantage of this approach is that the resulting gaits are automatically adapted to the environment that is being modeled as part of the snake dynamics. To illustrate this, we use a novel model for anisotropic dry friction, along with existing models for viscous friction and fluid dynamic effects such as drag and added mass. For each of these models, gaits generated without any change in the method or its parameters are as efficient as Pareto-optimal serpenoid gaits tuned individually for each environment. Furthermore, the proposed method can also produce more complex or irregular gaits, e. g. for obstacle avoidance or executing sharp turns.

Details