Selected work

Representative Papers

Data Efficiency and Transfer Robustness in Biomedical Image Segmentation: A Study of Redundancy and Forgetting with Cellpose

Nov 06, 2025

This study addresses two key challenges in applying Cellpose to biomedical image segmentation: data redundancy and catastrophic forgetting during cross-domain transfer. To tackle these, we propose a Data Quantization (DQ) strategy and a selective replay mechanism. Leveraging MAE embeddings and t-SNE analysis of the latent space, we identify that only 10% representative samples suffice to achieve performance saturation—significantly improving training efficiency and feature diversity. In multi-stage cross-domain transfer, replaying just 5–10% of source-domain data effectively mitigates catastrophic forgetting and enables optimized domain transfer ordering. Extensive experiments on the Cyto dataset validate the efficacy of our approach; the code is publicly available. Our core contributions are: (i) the first systematic characterization of Cellpose’s data redundancy boundary, and (ii) a lightweight transfer learning paradigm that jointly optimizes data efficiency and knowledge retention.

0 citationsRead paper

NODA-MMH: Certified Learning-Aided Nonlinear Control for Magnetically-Actuated Swarm Experiment Toward On-Orbit Proof

Oct 23, 2025

To address four key challenges—nonholonomic constraints, underactuation, poor scalability, and high computational complexity—in large-scale magnetically actuated satellite formations, this paper proposes NODA-MMH: a decentralized, power-optimal magnetic current allocation framework integrating time-averaged dynamical modeling, neural-network-based learning modeling, and model predictive control. For the first time, the controllability enhancement of multi-satellite magnetic control systems is experimentally validated on a ground-based air-bearing platform equipped with a dual-axis coil system, with theoretical error bounds rigorously established. Results demonstrate that the proposed learning-enhanced time-averaged current control significantly improves both controllability and energy efficiency. This work establishes a novel paradigm and delivers critical enabling technologies for long-duration autonomous formation flying of on-orbit satellite swarms.

0 citationsRead paper

Recent publications

Latest Papers

Data Efficiency and Transfer Robustness in Biomedical Image Segmentation: A Study of Redundancy and Forgetting with Cellpose

Nov 06, 2025

This study addresses two key challenges in applying Cellpose to biomedical image segmentation: data redundancy and catastrophic forgetting during cross-domain transfer. To tackle these, we propose a Data Quantization (DQ) strategy and a selective replay mechanism. Leveraging MAE embeddings and t-SNE analysis of the latent space, we identify that only 10% representative samples suffice to achieve performance saturation—significantly improving training efficiency and feature diversity. In multi-stage cross-domain transfer, replaying just 5–10% of source-domain data effectively mitigates catastrophic forgetting and enables optimized domain transfer ordering. Extensive experiments on the Cyto dataset validate the efficacy of our approach; the code is publicly available. Our core contributions are: (i) the first systematic characterization of Cellpose’s data redundancy boundary, and (ii) a lightweight transfer learning paradigm that jointly optimizes data efficiency and knowledge retention.

0 citationsRead paper

NODA-MMH: Certified Learning-Aided Nonlinear Control for Magnetically-Actuated Swarm Experiment Toward On-Orbit Proof

Oct 23, 2025

To address four key challenges—nonholonomic constraints, underactuation, poor scalability, and high computational complexity—in large-scale magnetically actuated satellite formations, this paper proposes NODA-MMH: a decentralized, power-optimal magnetic current allocation framework integrating time-averaged dynamical modeling, neural-network-based learning modeling, and model predictive control. For the first time, the controllability enhancement of multi-satellite magnetic control systems is experimentally validated on a ground-based air-bearing platform equipped with a dual-axis coil system, with theoretical error bounds rigorously established. Results demonstrate that the proposed learning-enhanced time-averaged current control significantly improves both controllability and energy efficiency. This work establishes a novel paradigm and delivers critical enabling technologies for long-duration autonomous formation flying of on-orbit satellite swarms.

0 citationsRead paper