Search research articles

ABOUT JoVE

Overview Leadership Blog JoVE Help Center

AUTHORS

Publishing Process Editorial Board Scope & Policies Peer Review FAQ Submit

LIBRARIANS

Testimonials Subscriptions Access Resources Library Advisory Board FAQ

RESEARCH

JoVE Journal Methods Collections JoVE Encyclopedia of Experiments Archive

EDUCATION

JoVE Core JoVE Business JoVE Science Education JoVE Lab Manual Faculty Resource Center Faculty Site

Terms & Conditions of Use

Search research articles

Related Experiment Video

Updated: May 5, 2026

Inverse Probability of Treatment Weighting Propensity Score using the Military Health System Data Repository and National Death Index

Inverse Probability of Treatment Weighting Propensity Score using the Military Health System Data Repository and National Death Index

Published on: January 8, 2020

Step-Wise Dual Dynamic DPSGD: Enhancing Performance on Imbalanced Medical Datasets with Differential Privacy.

Xiaobo Huang¹, Fang Xie¹

¹Guangdong Provincial Key Laboratory of IRADS, Beijing Normal-Hong Kong Baptist University, Zhuhai 519087, China.

Entropy (Basel, Switzerland)

|May 4, 2026

Summary

Related Concept Videos

You might also read

Related Articles

Articles linked to this work by shared authors, journal, and citation graph.

Sort by

Same author

Bovine Lactoferrin Inhibits Pseudorabies Virus Attachment Through Blockade of Glycoprotein C-Heparan Sulfate Interaction.

Transboundary and emerging diseases·2026

Same author

Establishment and Application of an Indirect ELISA for Detecting Getah Virus IgG Antibodies in Swine Based on the E2EP3 Peptide.

Veterinary sciences·2026

Same author

Expert Consensus on the Combined Application of Radiotherapy and Novel Systemic Agents in Breast Cancer Treatment.

Journal of evidence-based medicine·2026

Same author

CXCL8 is associated with aflatoxin B1-triggered injury and caspase-3 activation in porcine kidney epithelial PK15 cells: integrated transcriptomics and CRISPR/Cas9 knockout.

Food and chemical toxicology : an international journal published for the British Industrial Biological Research Association·2026

Same author

PHP (phosphoric acid-hydrogen peroxide) pretreatment for efficiently deconstructing lignocellulose with simultaneously organic pollutants degradation by the derived-gas.

Bioresource technology·2026

Same author

Design, Synthesis, Anti-Oomycete, Antifungal, and Nematicidal Activities of 2'-Acyloxyfenjuntong Derivatives.

Chemistry & biodiversity·2026

Differential privacy degrades deep learning on imbalanced medical data. A new method, SDD-DPSGD, uses dynamic scheduling to preserve gradient information, improving model performance on datasets like HAM10000.

Area of Science:

Medical Informatics
Computer Science
Machine Learning

Background:

Differential privacy methods often degrade deep learning model performance on imbalanced medical datasets.
Standard techniques like adding noise to gradients are ineffective on small, imbalanced datasets (e.g., HAM10000, ISIC2019) due to gradient clipping and majority class dominance.

Purpose of the Study:

To address the performance degradation of differential privacy in deep learning for imbalanced medical datasets.
To propose a novel method that preserves crucial gradient information from few-shot classes.

Main Methods:

Introduced SDD-DPSGD (Step-wise Dynamic Differential Privacy Stochastic Gradient Descent).
Employs a step-wise dynamic exponential scheduling for noise and clipping thresholds.

Keywords:

convolutional neural network deep learning differential privacy dynamic differentially private mechanism imbalanced dataset

Related Experiment Videos

Last Updated: May 5, 2026

Inverse Probability of Treatment Weighting Propensity Score using the Military Health System Data Repository and National Death Index

Inverse Probability of Treatment Weighting Propensity Score using the Military Health System Data Repository and National Death Index

Published on: January 8, 2020

Allocates a larger privacy budget and higher clipping thresholds during initial training phases.

Main Results:

SDD-DPSGD effectively preserves gradient information, preventing models from converging to suboptimal solutions early.
Experimental results demonstrate superior performance of SDD-DPSGD compared to existing algorithms.
Validation on HAM10000 and ISIC2019 datasets confirmed the method's efficacy.

Conclusions:

SDD-DPSGD offers a significant improvement for applying differential privacy to imbalanced medical deep learning tasks.
The proposed dynamic scheduling mechanism enhances model robustness and performance in privacy-preserving machine learning.
This approach mitigates information loss in few-shot classes, leading to better overall model accuracy.