I just doing my master thesis and my theme is AI of medical image. I have a total of 1000 training images, but I should to clarify how many images I need to reduce the number of training images to. So I investigate how the number of training images affects the segmentation result by reducing the number of images by 100. However, if I reduce the number of images by 100, the training images will become more and more biased. How should I select the training images?

im a first year ug student and im currently persuing b tech in an AI related field. so recently my college announced that they’ll be holding a little competition type of thing for research papers. The topic can be anything related to ML and AI. i have a basic knowledge of these, and now im wondering if i should go with something more broad like “Ethics in AI” or something more specific like “Using ML algos to detect Mental health issues”? You can reccomend more topics too.

Building a ML model that will detect anomalies in the SQL database, what approach should I take