Explore a novel approach to deep learning called Filtered Transfer Learning (FTL) in this 35-minute conference talk from the Toronto Machine Learning Series. Discover how FTL addresses the challenge of training neural networks on datasets with varying levels of label confidence. Learn how this hierarchical fine-tuning process improves predictive power in noisy data systems by defining multiple tiers of data confidence as separate tasks in a transfer learning setting. Examine the application of FTL in predicting drug-protein interactions and understand its potential benefits for machine learning in fields like biology and medicine where large datasets often have uncertain labels. Gain insights into how this method can be applied to various domains, including radiological and histopathological image labeling and cancer drug resistance measurements.