Histology Augmentation: Scaling Federated Learning Solutions with Kubernetes
summary
Project builds an ACGAN within Federated Learning that generates class‑conditioned histology images to upsample the minority class from a colorectal dataset (≈15,000 train / 4,000 test, 75% negative / 25% positive). Implementation runs FL rounds synchronously across simulated clients using Kubernetes namespaces as hospitals on GPU nodes, aggregates model weights, and uses CNNs and Vision Transformers for evaluation. Metrics show FID decreasing during training, t‑SNE and GradCAM preserve structure, and classification accuracy on the original test set improves by about 3%.
AI-written summary · about
chapters (5)
- 0:00Federated learning in histology
- 6:05Medical image augmentation challenges
- 13:29Generative model with federated learning
- 18:18Kubernetes setup for training
- 21:33Results and accuracy gains
related talks
27mKubernetes + AI Joining Forces in the Battle Against Cancer
45mManaged federated learning on AWS: A case study for healthcare
31mFederated Machine Learning with Kubernetes- 31mFederated Machine Learning with Kubernetes
31mFrom Theory to Practice: Advances in Real-World Federated Learning
18mEnabling Medical AI through Federated Learning
26mDecentralized Federated Machine Learning: Empowering Edge Devices with Kubernetes
27mFederated Learning with Azure Machine Learning NVIDIA FLARE and MONAI