on Chest XRays Matthew Beaubien advised by Dr Lubomir Hadjiiski and Dr Heang Ping Chan Motivation How do we train on a lot of data without sharing it ID: 1042909
Download Presentation The PPT/PDF document "Federated Learning for Classification of..." is the property of its rightful owner. Permission is granted to download and print the materials on this web site for personal, non-commercial use only, and to display it on your personal computer provided you do not modify the materials and that you retain all copyright notices contained in the materials. By downloading content from our website, you accept the terms of this agreement.
1. Federated Learning for Classification of COVID-19 Severity based on Chest X-RaysMatthew Beaubien, advised by Dr. Lubomir Hadjiiski and Dr. Heang-Ping Chan
2. MotivationHow do we train on a lot of data without sharing it?What kind of model should we use?Why do we want to classify severity?
3. Chest X-Rays of Patients with Different Levels of COVID SeverityModerate and Severe look very similar
4. Dataset Sourcing and Splits
5. Image Preprocessing Before Training
6. Cyclic Sequential Training
7. Cyclic Ensemble Training
8. Results
9. Results
10. Results
11. Results
12. Results
13. Results
14. Results
15. ConclusionUsing Federated Learning, we can train models that perform as well or better than the baseline.The presence of multiple sites results in a more accurate model.Ensemble learning does not seem to “forget” in the same way that sequential learning can.Artificially splitting up the bigger dataset resulted in a much more accurate model.