Selected work

Representative Papers

FedSSMCoOp: SSM Encoders for light-weight Federated Prompt Learning for Few-shot Classification

Oct 07, 2026

This study addresses the challenge of multimodal alignment in federated biomedical settings, where data heterogeneity, label scarcity, and privacy constraints pose significant obstacles. To this end, we propose a lightweight federated few-shot classification framework that leverages Vision Mamba and Cross Mamba encoders to jointly optimize soft prompts and communication prompts. This approach achieves cross-modal feature alignment without relying on external large language models. By fine-tuning only the prompt parameters, the method substantially reduces computational overhead while effectively capturing cross-modal interaction structures. Extensive experiments demonstrate that the proposed framework consistently outperforms baseline methods across multiple biomedical datasets, achieving an average model size reduction of 1.96×. Overall, this work enables efficient and privacy-preserving multimodal learning for federated biomedical applications.

0 citationsRead paper

Optimal compression with quantum retrieval

Oct 05, 2026

This study addresses the problem of achieving optimal lossless compression and efficient bit retrieval for high Hamming weight strings within the quantum query model. To balance storage space and retrieval efficiency, this work proposes adaptive and non-adaptive quantum query mechanisms built upon Quantum Random Access Codes (QRACs) and standard oracle encodings. The adaptive scheme attains an optimal compression rate up to a logarithmic factor, while the non-adaptive scheme achieves near-optimality with respect to parameter m, degrading by at most a quadratic factor in n. By realizing compression rates that approach theoretical limits, this research significantly optimizes both the space complexity and query performance of quantum data retrieval.

0 citationsRead paper

Atoms to Processes: The Role of Artificial Intelligence and Machine Learning in Chemical Engineering

Oct 01, 2026

This study addresses critical bottlenecks in artificial intelligence models for chemical engineering, including prediction inaccuracies, data scarcity, and limited interpretability, by exploring the deep integration of AI with first-principles modeling. Methodologically, this work proposes a paradigm shift from purely black-box approaches to physics-informed hybrid frameworks, synergizing atomistic simulations, process systems engineering, and physics-informed neural networks to ensure strict adherence to thermodynamic consistency and physical conservation laws. Consequently, the project significantly enhances predictive robustness and reliability while enabling efficient human–machine collaboration. By safeguarding engineering safety, this research amplifies the scientific value of core chemical engineering principles.

0 citationsRead paper

A 3GPP-Compliant Benchmark Dataset for RIS-Aided Beyond 5G Networks

Sep 30, 2026

This study addresses the lack of standardized, high-fidelity open-source datasets for reconfigurable intelligent surface (RIS) research in beyond 5G (B5G) networks. Based on the 3GPP TR 38.901 channel model, we construct a large-scale millimeter-wave RIS dataset encompassing diverse propagation scenarios and globally optimal phase configurations. To our knowledge, this is the only publicly available dataset providing global optimal RIS phase labels. Furthermore, we propose a scalable channel state information (CSI)-to-channel quality indicator (CQI) scalar classification benchmark task alongside a deep learning evaluation framework. Experimental results validate the effectiveness of the proposed dataset under both in-distribution and out-of-distribution settings, as well as through hardware measurements. This work establishes a solid foundation for data-driven research in RIS-assisted wireless networks.

0 citationsRead paper

TO-mdiSPAs: Topology Optimization of multi-directional Soft Pneumatic Actuators

Sep 30, 2026

This study addresses the design challenge of balancing manufacturing robustness with arbitrary directional bending in multi-directional soft pneumatic actuators by proposing a systematic topology optimization method. The approach introduces a three-field formulation to ensure manufacturing robustness and incorporates Darcy’s law with a drainage term to model design-dependent loads. Furthermore, a minimax optimization model based on target deformations is constructed and efficiently solved using the method of moving asymptotes. The resulting high-performance, unconventional geometries are validated through numerical simulations, which confirm that the optimized actuators successfully achieve flexible multi-directional motion capabilities in three-dimensional space. This work provides a new paradigm for the innovative structural design of soft robots.

0 citationsRead paper

Recent publications

Latest Papers

FedSSMCoOp: SSM Encoders for light-weight Federated Prompt Learning for Few-shot Classification

Oct 07, 2026

This study addresses the challenge of multimodal alignment in federated biomedical settings, where data heterogeneity, label scarcity, and privacy constraints pose significant obstacles. To this end, we propose a lightweight federated few-shot classification framework that leverages Vision Mamba and Cross Mamba encoders to jointly optimize soft prompts and communication prompts. This approach achieves cross-modal feature alignment without relying on external large language models. By fine-tuning only the prompt parameters, the method substantially reduces computational overhead while effectively capturing cross-modal interaction structures. Extensive experiments demonstrate that the proposed framework consistently outperforms baseline methods across multiple biomedical datasets, achieving an average model size reduction of 1.96×. Overall, this work enables efficient and privacy-preserving multimodal learning for federated biomedical applications.

0 citationsRead paper

Optimal compression with quantum retrieval

Oct 05, 2026

This study addresses the problem of achieving optimal lossless compression and efficient bit retrieval for high Hamming weight strings within the quantum query model. To balance storage space and retrieval efficiency, this work proposes adaptive and non-adaptive quantum query mechanisms built upon Quantum Random Access Codes (QRACs) and standard oracle encodings. The adaptive scheme attains an optimal compression rate up to a logarithmic factor, while the non-adaptive scheme achieves near-optimality with respect to parameter m, degrading by at most a quadratic factor in n. By realizing compression rates that approach theoretical limits, this research significantly optimizes both the space complexity and query performance of quantum data retrieval.

0 citationsRead paper

Atoms to Processes: The Role of Artificial Intelligence and Machine Learning in Chemical Engineering

Oct 01, 2026

This study addresses critical bottlenecks in artificial intelligence models for chemical engineering, including prediction inaccuracies, data scarcity, and limited interpretability, by exploring the deep integration of AI with first-principles modeling. Methodologically, this work proposes a paradigm shift from purely black-box approaches to physics-informed hybrid frameworks, synergizing atomistic simulations, process systems engineering, and physics-informed neural networks to ensure strict adherence to thermodynamic consistency and physical conservation laws. Consequently, the project significantly enhances predictive robustness and reliability while enabling efficient human–machine collaboration. By safeguarding engineering safety, this research amplifies the scientific value of core chemical engineering principles.

0 citationsRead paper

A 3GPP-Compliant Benchmark Dataset for RIS-Aided Beyond 5G Networks

Sep 30, 2026

This study addresses the lack of standardized, high-fidelity open-source datasets for reconfigurable intelligent surface (RIS) research in beyond 5G (B5G) networks. Based on the 3GPP TR 38.901 channel model, we construct a large-scale millimeter-wave RIS dataset encompassing diverse propagation scenarios and globally optimal phase configurations. To our knowledge, this is the only publicly available dataset providing global optimal RIS phase labels. Furthermore, we propose a scalable channel state information (CSI)-to-channel quality indicator (CQI) scalar classification benchmark task alongside a deep learning evaluation framework. Experimental results validate the effectiveness of the proposed dataset under both in-distribution and out-of-distribution settings, as well as through hardware measurements. This work establishes a solid foundation for data-driven research in RIS-assisted wireless networks.

0 citationsRead paper

TO-mdiSPAs: Topology Optimization of multi-directional Soft Pneumatic Actuators

Sep 30, 2026

This study addresses the design challenge of balancing manufacturing robustness with arbitrary directional bending in multi-directional soft pneumatic actuators by proposing a systematic topology optimization method. The approach introduces a three-field formulation to ensure manufacturing robustness and incorporates Darcy’s law with a drainage term to model design-dependent loads. Furthermore, a minimax optimization model based on target deformations is constructed and efficiently solved using the method of moving asymptotes. The resulting high-performance, unconventional geometries are validated through numerical simulations, which confirm that the optimized actuators successfully achieve flexible multi-directional motion capabilities in three-dimensional space. This work provides a new paradigm for the innovative structural design of soft robots.

0 citationsRead paper