Convolutional Neural Networks (CNNs)

Speaker

John Wu, STScI

Build a convolutional neural network (CNN) to estimate physical properties of galaxies directly from images. Train a model to predict gas-phase metallicity from SDSS galaxy images, replicating the approach from Wu & Boada (2019).

Topics Covered:

• Loading galaxy images as tensors (g, r, i bands)
• Building CNNs from scratch in PyTorch
• Convolutional layers, pooling, and feature extraction
• Training and optimization with gradient descent
• Model evaluation and performance analysis
• Hands-on: predicting metallicity from SDSS images

Meeting Connection

Join the Meeting