AMLD 2021 Workshop
Practical part for AMLD 2021 Workshop based on the original work of Pawel Rosikiewicz available at Github.
SkinDiagnosticAI
Comparison of 5000 AI methods for cancer detection and classyfication on dermatoscopic images from Harvard HAM10000 dataset using FastClassAI workbench
Author: Pawel Rosikiewicz, Founder, and Team Leader at SwissAI
License: MIT
The goal is to quickly test and optimize large number of ml and deep learning models and dataset preprocessing procedures, that are integrated in one python enviroment with FastClassAI workebench.
Main Goals are:
- to identify main challenges with the dataset used for model training,
- to explore different strategies for data preparation, treatment and feature extraction,
- to test, of the shelf AI solutions, with extensive grid search,
- to develope baseline for further analyses,
- to evaluate what statistics and error fucntions shodul be used for developing final and ensemble models,
Presentation on SkinDiagnosticAI Project
- all images were created wiht FastClassAI pipeline,
- the slides shows full analyis done on over 5000 compared models and data treatment procedures,
- Jupyter notebooks in notebook/ folder shows light vervion of that analyis that can be reapeated by the user and build up to any number of compared models,
![skindiagnosticai presentation slide](images/Slide19.png