Active Learning For Experiment Automation

This project aimed at creating a server hosting machine learning algorithms for a robot to query at Argonne National Laboratory. The end goal of the entire initiative at Argonne was to automate the tedious experimentation that material scientists need to do to discover new valuable materials. These experiments often involve thousands of iterations with small deviations in test conditions. For example, testing how conductive a material is if it is tempered at 5 degrees C higher than before. The goal of my work was to create a machine learning system that can suggest what the next test conditions should be for automated experiments based on the data collected from previous experiments. My work consists of adding to the basic server infrastructure and testing machine learning methods.

Presentation

Evan Costa Final Presentation .pptx

Code Repositories

This code base is the open source Argonne repository I contributed to.

GitHub - AD-SDL/polybot-web-service: A server to control robotic synthesis of thin-film polymeric materialsA server to control robotic synthesis of thin-film polymeric materials - GitHub - AD-SDL/polybot-web-service: A server to control robotic synthesis of thin-film polymeric materials

This is my forked work where a couple teammates and I created an online interface for scientists at external institutions to suggest experimental conditions to be incorporated into the machine learning algorithms we created.

GitHub - evanpcosta/IEEEPolybotContribute to evanpcosta/IEEEPolybot development by creating an account on GitHub.

This is the forked repository where I tested the machine learning methods.

GitHub - evanpcosta/active-learning: Collection of active learning algorithmsCollection of active learning algorithms. Contribute to evanpcosta/active-learning development by creating an account on GitHub.