Learning with less data
Investigating how dataset size, augmentation, and fine-tuning depth shape model behavior.
The question
How does the amount of available training data affect the depth of fine-tuning that is useful? And how do augmentation choices influence stability and the balance of errors?
The approach
This research studies transfer-learning configurations for chest X-ray classification. The work compares adaptation choices and examines the results beyond a single overall score. The two linked papers describe the experiments, data partitions, methods, and limits in detail.
Why it matters
Training decisions interact. A strategy that works with one dataset size may behave differently with another. Comparing controlled configurations helps make those dependencies visible.
Read the research
- Dataset size effects on fine-tuning depth (2026)
- Augmentation strength, training stability, and clinical trade-offs (2026)
This page introduces the research direction. It does not imply a clinically deployed system.
A useful project starts with a clear question.
Explore related research, practical learning, or an idea we could work on together.