Why using artificial intelligence in the search for gravitational waves?
Elena Cuoco, EGO and SNS www.elenacuoco.com Twitter: @elenacuoco
What are Gravitational Waves (GWs)?
2
General Relativity (1915)
G mn Gravitational Waves (1916)
SOPHI.A.2018 AI & Space
8pG = 4 Tmn c Elena Cuoco
How we detected GWs?
SOPHI.A.2018 AI & Space
3
Elena Cuoco
Astrophysical Gravitational Wave signals
4
An artist's impression of two stars orbiting each other and progressing (from left to right) to merger with resulting gravitational waves. [Image: NASA/CXC/GSFC/T.Strohmayer]
An example signal from an inspiral gravitational wave source. [Image: A. Stuver/LIGO]
SOPHI.A.2018 AI & Space
Elena Cuoco
International Collaboration
SOPHI.A.2018 AI & Space
5
Elena Cuoco
GW150914 and GW170817
NGC 4993 GRB170817A Hubble telescope
6
First Detection of Gravitational Waves from 2 colliding Neutron Stars ~1.5-2 Solar mass each
First Detection of Gravitational Waves! 2 colliding Black Holes ~30Solar mass each SOPHI.A.2018 AI & Space
Artist’s illustration of the merger of two neutron stars, producing a short gamma-ray burst. [NSF/LIGO/Sonoma State University/A. Simonnet]
Elena Cuoco
Why Artificial Intellingence in Gravitational Wave research? SOPHI.A.2018 AI & Space
Elena Cuoco
7
8
LIGO/Virgo data are time series sequences‌ noisy time series with low amplitude GW signal buried in SOPHI.A.2018 AI & Space
Elena Cuoco
Our “signals”
9
Known GW signals
Unknown GW signals
Noise
Compact coalescing binaries has known theoretical waveforms
Core collapse supernovae
Moving lines
Optimal filter: Matched filter Too many templates to test SOPHI.A.2018 AI & Space
Broad band noise No Optimal filter
Parameters estimation
Glitch noise
“Pattern recognition” by visual inspection
Elena Cuoco
10
Example of GW signals in Time-Frequency plots
SOPHI.A.2018 AI & Space
Elena Cuoco
Glitch gallery
11
https://www.zooniverse.org/projects/zooniverse/gravity-spy
SOPHI.A.2018 AI & Space
Elena Cuoco
Example of other noise signals
12
I. Fiori courtesy SOPHI.A.2018 AI & Space
Elena Cuoco
13
Numbers about data Data Stream Flux
Data on disk
• 50MB/s
• 1-3PB
Number of events • 1/week • 1/day?
Number of glitches • 1/sec • 0.1/sec?
Should be analysed in less than 1min
SOPHI.A.2018 AI & Space
Elena Cuoco
Artificial Intelligence in our lives ▪
Virtual assistants
▪
▪
Advertising
Fraud detection
▪
▪
Facial recognition
Marketing
▪
▪
Smart cities
Domotic
▪
▪
Smart cars
Music suggestion
▪
Film suggestion
SOPHI.A.2018 AI & Space
14
Elena Cuoco
Artificial Intelligence...what we image
SOPHI.A.2018 AI & Space
15
Elena Cuoco
Artificial Intelligence in practice
SOPHI.A.2018 AI & Space
16
Elena Cuoco
Machine learning models
17 Semi-supervised
Unsupervised
No label for the data
• Few labeled data • A lot of not labeled data
Reinforcement learning Supervised
Labeled training data
SOPHI.A.2018 AI & Space
Elena Cuoco
How Machine Learning can help Data conditioning
▪ Identify Non linear noise coupling ▪ Use Deep Learning to remove noise
▪ Extract useful features to clean
18
Signal Detection/Classification/PE ▪
A lot of fake signals due to noise
▪
Fast alert system
▪
Manage parameter estimation
data
SOPHI.A.2018 AI & Space
Elena Cuoco
Glitch classification strategy for GW detectors
SOPHI.A.2018 AI & Space
Elena Cuoco
20
Selecting data format for training set ▪
Images
SOPHI.A.2018 AI & Space
▪
Time series
Elena Cuoco
Deep learning for Classification
21
Dog versus Cat
Copyright https://selmandesign.com/qa-on-machine-learning/
SOPHI.A.2018 AI & Space
Elena Cuoco
22
Images-based glitch classification Deep learning with CNN
Massimiliano Razzano and Elena Cuoco 2018 Class. Quantum Grav. 35 095016 SOPHI.A.2018 AI & Space
Elena Cuoco
Test on simulated data sets To test the pipeline, we prepared ad-hoc simulations
23
Add 6 different classes of glitch shapes
Simulate colored noise using public H1 sensitivity curve
SOPHI.A.2018 AI & Space
Elena Cuoco
Simulated signal families
24 Waveform Gaussian Sine-Gaussian Ring-Down Chirp-like Scattered-like Whistle-like NOISE (random) To show the glitch time-series here we don’t show the noise contribution Razzano M., Cuoco E. CQG-104381.R3
SOPHI.A.2018 AI & Space
Elena Cuoco
Signal distribution
25
Simulated time series with 8kHz sampling rate
Glitches distributed with Poisson statistics m=0.5 Hz
2000 glitches per each family Glitch parameters are varied randomly to achieve various shapes and Signal-To-Noise ratio
SOPHI.A.2018 AI & Space
Elena Cuoco
Building the images
26
Spectrogram for each image 2-seconds time window to highlight fatures in long glitches Data is whitened Optional contrast stretch
SOPHI.A.2018 AI & Space
Elena Cuoco
Pipeline structure
27
Input GW data • • • •
Image processing Time series whitening Image creation from time series (FFT spectrograms) Image equalization & contrast enhancement
Classification • A probability for each class, take the max • Add a NOISE class to crosscheck glitch detection
Network layout • Tested various networks, including a 4-block layers
Run on GPU Nvidia GeForce GTX 780 • 2.8k cores, 3 Gb RAM) • Developed in Python + CUDA-optimized libraries
SOPHI.A.2018 AI & Space
Elena Cuoco
Example of classification results
28
Some cases of more glitches in the time window, always identify the right class
100% Sine-Gaussian
SOPHI.A.2018 AI & Space
Elena Cuoco
Classification accuracy : confusion matrix
Deep CNN able at distinguishing similar morphologies
SOPHI.A.2018 AI & Space
Elena Cuoco
29
30
Working with time series and wavelet decomposition Whitening in time domain
Data
Wavelet transform
De-noising
Parameter estimation
Trigger list
Gps, duration, snr, snr@max, freq_mean, freq@max, wavelet type triggered + corresponding wavelets coefficients. SOPHI.A.2018 AI & Space
Elena Cuoco
eXtreme Gradient Boosting â—? â—?
â—?
https://github.com/dmlc/xgboost Tianqi Chen and Carlos Guestrin. XGBoost: A Scalable Tree Boosting System. In 22nd SIGKDD Conference on Knowledge Discovery and Data Mining, 2016 XGBoost originates from research project at University of Washington, see also the Project Page at UW.
31
Tree Ensemble đ??ž
đ?‘Śđ?‘› = ŕˇ? đ?‘“đ?‘˜ đ?‘Ľđ?‘› đ?‘˜=1
SOPHI.A.2018 AI & Space
Elena Cuoco
WDFX: Binary Classification Results
32
Overall accuracy >90%
Wavelet-based Classification of Transient Signals for Gravitational Wave Detectors, E. Cuoco, M. Razzano, A. Utina, EUSIPCO 2018
SOPHI.A.2018 AI & Space
Elena Cuoco
33
WDFX Results: Multi-Label Classification
Overall accuracy >80%
SOPHI.A.2018 AI & Space
Elena Cuoco
What’s next?
SOPHI.A.2018 AI & Space
34
Elena Cuoco
35
Thanks! If you have any questions, I’ll be around SOPHI.A.2018 AI & Space
Elena Cuoco