Inductive Biases

Speaker

John Wu, STScI

Understand how different neural network architectures encode different assumptions about data structure. Compare MLPs, CNNs, RNNs, and Transformers through the lens of inductive biases using synthetic exoplanet transit detection as a case study.

Topics Covered:

• What are inductive biases in neural networks
• MLPs: independent feature assumption
• CNNs: local temporal features and translation invariance
• RNNs: sequential dependencies
• Transformers: attention and long-range dependencies
• Hands-on: transit detection in synthetic light curves

Lecture notes and slides will be up later on the github repo.

We'll be using slido for Q&A: https://app.sli.do/event/9jLSjC8GYLePQD3RPVXNv5

Meeting Connection

Join the Meeting