Salensy: Using RISE to Create Saliency Maps of Strong Gravitational Lensing Images
Advisor(s)
Dr. Brian Nord, Fermilab
Location
Room B110
Start Date
26-4-2019 10:25 AM
End Date
26-4-2019 10:40 AM
Abstract
A convolutional neural network is a system of interconnected information nodes that is used for image recognition and classification problems in computer science. Random Input Sampling for Explanation of Black-box models, otherwise known as RISE, is a program that generates a saliency map indicating how important each pixel of an image is for the classification algorithm of a convolutional neural network. Saliency maps transform the opaque “black box” of image classification networks into an explicitly defined process, allowing researchers to fine tune their model and build trust in their network. In our project, we will implement RISE with maps of astronomical images demonstrating strong gravitational lensing to allow further research into computationally constructing a model of the lensing process.
Salensy: Using RISE to Create Saliency Maps of Strong Gravitational Lensing Images
Room B110
A convolutional neural network is a system of interconnected information nodes that is used for image recognition and classification problems in computer science. Random Input Sampling for Explanation of Black-box models, otherwise known as RISE, is a program that generates a saliency map indicating how important each pixel of an image is for the classification algorithm of a convolutional neural network. Saliency maps transform the opaque “black box” of image classification networks into an explicitly defined process, allowing researchers to fine tune their model and build trust in their network. In our project, we will implement RISE with maps of astronomical images demonstrating strong gravitational lensing to allow further research into computationally constructing a model of the lensing process.