DUQ FedAvg Project
Home DUQ FedAvg Project Federated Learning CNN Backbones U-Net Models Dual-UQ

Federated Learning for Neglected Tropical Disease Classification

This site presents a fully client-side application of federated learning for parasite image classification. It supports multiple convolutional backbones and a Dual-UQ federated optimiser for uncertainty-aware aggregation.

Open Classifier Dashboard Read Background
At a Glance
  • Diseases: Intestinal Nematode Infection, Schistosomiasis, Taeniasis
  • Backbones: DenseNet121, ResNet50, VGG16
  • Training strategy: federated averaging variants with Dual-UQ
  • Purpose: PhD defence; no clinical interpretation

Core Deep Learning Architectures

Architectures used in the PhD federated learning for parasitic infection classification and localization.

ResNet50 architecture
ResNet50

Deep residual CNN with skip connections, used as a backbone for parasite classification in the federated learning setup.

CNN Backbone
DenseNet121 architecture
DenseNet121

Densely connected CNN that reuses features across layers. Serves as a strong baseline and as the backbone for the proposed Dual_UQ method.

CNN Backbone · Dual_UQ
VGG16 architecture
VGG16

Classic deep CNN with stacked 3×3 convolutions, used as a simpler baseline against more advanced backbones.

Baseline CNN
U-Net architecture
U-Net

Encoder–decoder with skip connections, used for pixel-level localization of parasitic structures in microscopy images.

Segmentation · Localization