Learning Science with Machine Learning - Opening the Pandora Box

Explore the intersection of machine learning and cosmology in this 21-minute APS Physics conference talk. Discover how Shirley Ho from the Flatiron Institute leverages advanced ML techniques to extract more information from cosmological datasets. Delve into the process of training, validating, and testing models using hundreds of simulations. Examine the challenges of interpreting machine learning models in scientific contexts, including an analysis of how models respond to specific input modes like plane waves. Investigate the UNET architecture and its applications in cosmology. Conclude by pondering the broader implications and potential of machine learning in expanding our understanding of the universe.

Syllabus

Intro
Traditional Cosmology analysis
Can we use Machine Learning to extract more information from the following datasets?
Training. Validation and Testing with O(100)s of simulations
Now as scientists, we have lots of questions..
Can we interpret what the model is learning?
Foray into understanding what the heck the Model is learning
Input mode: A Plane Wave What happen if we have power only one scale?
Interrogating the learned model What happens we change the phase of the input mode? UNET
Let's leave you with questions: Why?
My possible climb to fame?