Speakers
Description
The Dogons of Mali have a long cultural history in interpreting and classifying celestial phenomena such as the stars Sirius A and Sirius B, reflecting human’s everlasting curiosity with the cosmos. Historically, surveys depended massively on human effort for classification which led to citizen-science projects like Zooniverse when early automated methods were unable to capture the subtle features needed for accurate identification. Modern neural networks, however, offer a way to automate this process more effectively but their accuracy still varies depending on architecture and training data.
This project aims to develop and compare the performance of two types of neural networks for a binary classification to identify whether an image contains a black hole, galaxy or star. With the implementation of a Convolutional Neural Network (CNN) known for capturing local spatial patterns and edges and a Dense Network known for treating each pixel independently without spatial awareness, this project investigates how the architectural differences such as the presence of convolutional layers impact accuracy and efficiency using supervised learning and a dataset of pre-tagged images of celestial bodies. The performance of these two architectural structures will be evaluated and compared to provide a more efficient solution in classifying astronomical images.
| Stream | Science or Engineering |
|---|