# DIRECT: Deep Discriminative Embedding for Clustering of LIGO Data

**Authors:** Sara Bahaadini, Vahid Noroozi, Neda Rohani, Scott Coughlin, Michael Zevin, Aggelos K. Katsaggelos  
**arXiv:** [1805.02296](https://arxiv.org/abs/1805.02296)  
**Date:** 2018-05-07  
**Summary page:** https://www.summarizepaper.com/en/paper/1805.02296v1/

## Key points

- Authors Sara Bahaadini, Vahid Noroozi, Neda Rohani, Scott Coughlin, Michael Zevin, and Aggelos K. Katsaggelos introduce a novel approach to clustering tasks using deep neural networks.
- The proposed method utilizes a discriminative embedding function as a feature extractor trained on labeled morphologically-distinct images or classes.
- The application targeted is the Gravity Spy Project focusing on characterizing transient, non-Gaussian noise in data from LIGO.
- By identifying new classes of noise through the embedding function, researchers aim to enhance understanding of noise disturbances in LIGO data and detectors.
- The work has been accepted for presentation at the 25th IEEE International Conference on Image Processing (ICIP), highlighting its significance in machine learning and signal processing fields.

## Summary

In their paper titled "DIRECT: Deep Discriminative Embedding for Clustering of LIGO Data," authors Sara Bahaadini, Vahid Noroozi, Neda Rohani, Scott Coughlin, Michael Zevin, and Aggelos K. Katsaggelos introduce a novel approach to clustering tasks using deep neural networks. The proposed method leverages the powerful capabilities of deep learning in estimating non-linear functions by utilizing a discriminative embedding function as a feature extractor. This function is trained on a labeled set of morphologically-distinct images or classes and can transfer knowledge to a new domain where new classes can be identified and isolated. The specific application targeted in this study is the Gravity Spy Project which aims to characterize transient, non-Gaussian noise present in data from the Advanced Laser Interferometer Gravitational-wave Observatory (LIGO). By accumulating large sets of labeled noise features and identifying new classes of noise through the proposed embedding function, the researchers seek to enhance understanding of the origin of these disturbances. This improved understanding enables more effective removal of unwanted noise from LIGO data and detectors. The work by Bahaadini et al. has been accepted for presentation at the 25th IEEE International Conference on Image Processing (ICIP), highlighting its significance in the field of machine learning and signal processing. Through their innovative approach to clustering using deep discriminative embedding, the researchers contribute valuable insights towards improving data analysis techniques in gravitational wave detection and characterization.