Paper: Automated Bias Reduction in Deep Learning Based Melanoma Diagnosis using a Semi-Supervised Algorithm
This repo contains code for the International Science and Engineering Fair hosted by Regeneron from May 16-21, 2021.
The main idea of the project is to be able to address racial and other
inherent forms of bias in a dataset while creating a deep learning
based diagnosis for Melanoma. Model details can be found in the Training folder of this repository. Full detail is in the paper.
Results
As you can see from the diagram, my proposed LatentNet successfully outperforms the conventional DCNN in 3 out of the 4 tested categories.
