Selected work

Representative Papers

Real-Time State-of-Health Estimation and Online Degradation Prognosis from Partial Battery Discharge Using Physics-Informed Neural Networks

Aug 14, 2026

This study addresses the limitations of fixed discharge protocols and real-time prediction challenges in lithium-ion battery state-of-health (SOH) estimation by proposing a novel framework integrating physical mechanisms with deep learning. By constructing a physics-informed neural network that utilizes partial discharge curves from arbitrary voltage intervals, the method enables adaptive monitoring without requiring prior knowledge or extensive historical data. This approach overcomes traditional protocol constraints, achieving a mean absolute percentage error below 4% under complex operating conditions while accurately capturing critical aging transitions in real time. Consequently, the proposed framework significantly enhances the flexibility and reliability of battery lifecycle management, offering a robust solution for practical deployment where standardized testing is infeasible.

0 citationsRead paper

Grow, Assess, Compress: Adaptive Backbone Scaling for Memory-Efficient Class Incremental Learning

Mar 09, 2026

This work addresses the challenges of parameter explosion and catastrophic forgetting in class-incremental learning by proposing GRACE, a dynamic scaling framework. GRACE employs a closed-loop “grow–evaluate–compress” mechanism to adaptively regulate model capacity: after expanding the backbone network, it assesses capacity saturation and intelligently decides whether to compress redundant structures, thereby balancing new task acquisition with retention of prior knowledge. The framework innovatively integrates dynamic network expansion, a saturation-aware evaluation algorithm, and model compression techniques to achieve highly efficient memory utilization. Experimental results demonstrate that GRACE achieves state-of-the-art performance across multiple class-incremental benchmarks while reducing memory consumption by up to 73% compared to pure expansion-based approaches.

0 citationsRead paper

Privacy-Preserving Feature Valuation in Vertical Federated Learning Using Shapley-CMI and PSI Permutation

Dec 16, 2025

In vertical federated learning (VFL), securely and fairly evaluating feature contributions remains challenging under model-agnostic assumptions, especially without access to raw data or shared model training. Method: This paper proposes the first privacy-preserving Shapley-CMI framework, integrating Shapley values with conditional mutual information (CMI) to enable model-agnostic, data- and training-free feature valuation. We design an encrypted permutation and intersection-size computation protocol based on private set intersection (PSI) servers, achieving fully decentralized, trustless implementation. To enhance efficiency, we introduce ID-based grouping discretization and lightweight homomorphic operations. Contributions/Results: Extensive experiments validate the framework’s correctness, end-to-end privacy guarantees, and scalability across multi-party VFL settings. Our approach establishes an auditable, interpretable foundation for fair feature valuation in early-stage VFL data collaboration—without requiring raw data sharing, model exchange, or a trusted third party.

0 citationsRead paper

On-site estimation of battery electrochemical parameters via transfer learning based physics-informed neural network approach

Mar 28, 2025

To address the urgent need for online, lightweight estimation of electrochemical parameters in battery management systems (BMS), this paper proposes a two-stage physics-informed neural network (PINN) framework. In Stage I, a lightweight neural network embedding the single-particle model (SPM) equations is pre-trained on synthetic simulation data via transfer learning. In Stage II, the network’s backbone is frozen, and only key electrochemical parameters—such as active material volume fraction—are optimized, enabling zero-shot field deployment and real-time edge inference on resource-constrained hardware (e.g., Raspberry Pi). The method eliminates reliance on labeled experimental data while preserving physical interpretability and computational efficiency. Evaluated on aging NMC batteries, it achieves a mere 3.89% relative error in volume fraction estimation and high accuracy in capacity retention prediction, demonstrating superior precision, low computational overhead, and strong generalization across degradation conditions.

0 citationsRead paper

Recent publications

Latest Papers

Real-Time State-of-Health Estimation and Online Degradation Prognosis from Partial Battery Discharge Using Physics-Informed Neural Networks

Aug 14, 2026

This study addresses the limitations of fixed discharge protocols and real-time prediction challenges in lithium-ion battery state-of-health (SOH) estimation by proposing a novel framework integrating physical mechanisms with deep learning. By constructing a physics-informed neural network that utilizes partial discharge curves from arbitrary voltage intervals, the method enables adaptive monitoring without requiring prior knowledge or extensive historical data. This approach overcomes traditional protocol constraints, achieving a mean absolute percentage error below 4% under complex operating conditions while accurately capturing critical aging transitions in real time. Consequently, the proposed framework significantly enhances the flexibility and reliability of battery lifecycle management, offering a robust solution for practical deployment where standardized testing is infeasible.

0 citationsRead paper

Grow, Assess, Compress: Adaptive Backbone Scaling for Memory-Efficient Class Incremental Learning

Mar 09, 2026

This work addresses the challenges of parameter explosion and catastrophic forgetting in class-incremental learning by proposing GRACE, a dynamic scaling framework. GRACE employs a closed-loop “grow–evaluate–compress” mechanism to adaptively regulate model capacity: after expanding the backbone network, it assesses capacity saturation and intelligently decides whether to compress redundant structures, thereby balancing new task acquisition with retention of prior knowledge. The framework innovatively integrates dynamic network expansion, a saturation-aware evaluation algorithm, and model compression techniques to achieve highly efficient memory utilization. Experimental results demonstrate that GRACE achieves state-of-the-art performance across multiple class-incremental benchmarks while reducing memory consumption by up to 73% compared to pure expansion-based approaches.

0 citationsRead paper

Privacy-Preserving Feature Valuation in Vertical Federated Learning Using Shapley-CMI and PSI Permutation

Dec 16, 2025

In vertical federated learning (VFL), securely and fairly evaluating feature contributions remains challenging under model-agnostic assumptions, especially without access to raw data or shared model training. Method: This paper proposes the first privacy-preserving Shapley-CMI framework, integrating Shapley values with conditional mutual information (CMI) to enable model-agnostic, data- and training-free feature valuation. We design an encrypted permutation and intersection-size computation protocol based on private set intersection (PSI) servers, achieving fully decentralized, trustless implementation. To enhance efficiency, we introduce ID-based grouping discretization and lightweight homomorphic operations. Contributions/Results: Extensive experiments validate the framework’s correctness, end-to-end privacy guarantees, and scalability across multi-party VFL settings. Our approach establishes an auditable, interpretable foundation for fair feature valuation in early-stage VFL data collaboration—without requiring raw data sharing, model exchange, or a trusted third party.

0 citationsRead paper

On-site estimation of battery electrochemical parameters via transfer learning based physics-informed neural network approach

Mar 28, 2025

To address the urgent need for online, lightweight estimation of electrochemical parameters in battery management systems (BMS), this paper proposes a two-stage physics-informed neural network (PINN) framework. In Stage I, a lightweight neural network embedding the single-particle model (SPM) equations is pre-trained on synthetic simulation data via transfer learning. In Stage II, the network’s backbone is frozen, and only key electrochemical parameters—such as active material volume fraction—are optimized, enabling zero-shot field deployment and real-time edge inference on resource-constrained hardware (e.g., Raspberry Pi). The method eliminates reliance on labeled experimental data while preserving physical interpretability and computational efficiency. Evaluated on aging NMC batteries, it achieves a mere 3.89% relative error in volume fraction estimation and high accuracy in capacity retention prediction, demonstrating superior precision, low computational overhead, and strong generalization across degradation conditions.

0 citationsRead paper