LIGO Projects
Noise due to light scattering shows up as arches in the time-frequency plane.
Reducing noise due to Scattered Light
Noise in the detector due to scattered light diminishes its sensitivity for the detection of gravitational waves. After finding the exact mechanism through which this noise couples to the detector, we were able to implement a technique that reduced it considerably. Find out more about this project in the links below.
​
Fast scattering triggers recognized by the new model of GravitySpy algorithm during Feb 2020
Noise classification with GravitySpy
​
GravitySpy is a deep learning algorithm that uses a convolutional neural network (CNN) to classify noise based on its appearance in the time-frequency plane.
I trained a new model to recognize two new classes of noise, namely "Fast Scattering" and "Low-frequency Blips". The identification of Fast Scattering led to an improved understanding of its possible origin in the gravitational wave detector.
Loud triggers at LIGO
​
Loud triggers are these monstrous noisy events that result in a huge drop in the sensitivity of the LIGO detector, and there are quite a few of them every day. In this project, we answer the question "Do we have more loud triggers/noise during Observing run 3, or are we just more sensitive to noise compared to Observing run 2?" The tougher question to answer "Where do these loud triggers originate?" is an ongoing investigation.
3D plot showing the relationship between the amplitude, Signal-to-Noise ratio of noise, and the astrophysical range (Mpc) which is a measure of how far in space we can detect gravitational waves from.
Personal Projects
Binary classification using CNN
​
In this project, I am using a deep layer convolutional neural network for binary classification. The network is trained over about 12000 images of cats and dogs, downloaded from the internet. More details about the model architecture, training, and validation loss and accuracy can be found at this Github repository.
Multiclass classification using CNN
​
Here I am developed my own mini noise classifier using a deep layer convolutional neural network which classifies noise into three different categories. The algorithm was trained on time-frequency spectrogram images of these three glitch classes as shown on the left. More details on this can be found here.