Wh Why using ing artifici icial al intelligenc igence e in the search rch for gr gravita vitational tional wa waves? s? Elena Cuoco, EGO and SNS www.elenacuoco.com Twitter: @elenacuoco
What are Gravitational Waves (GWs)? 2 General Relativity (1915) G m n = 8 p G c 4 T m n Gravitational Waves (1916) Elena Cuoco
How we detected GWs? 3 Elena Cuoco
Astrophysical Gravitational Wave signals 4 An artist's impre ression of two stars rs orbiti ting each other r and progre ressing (from left t to right) t) to merg rger r with th result lting gravitati vitational waves ves. [Image: NASA/CXC/ C/GSFC/T. FC/T.Stro Strohmaye yer] r] An example le signal from an inspir iral l gravit itational wave source. [Image: e: A. Stuver/LIGO GO] Elena Cuoco
International Collaboration 5 Elena Cuoco
GW150914 and GW170817 6 NGC 4993 GRB170817A Hubble telescope First Detection n of Gravitational nal Waves from 2 colliding ng Neutron n Stars ~1.5-2 Solar mass each First Detection n of Gravitational nal Waves! 2 colliding ding Black ck Holes ~30Sola olar r mass each 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
7 Why Machine Learning in Gravitational Wave research Elena Cuoco
Outline 8 Glitches classification • Image-based • Wavelet-based Machine learning for New ideas and possible Gravitational collaborations Noise in COST Removal action Wave Data framework analysis Real time analysis (on going work) Elena Cuoco
9 LIGO/Virgo data are time series sequences … noisy sy time series ies with low amplitude GW signal buried in Elena Cuoco
Our “signals ” 10 Known GW signals ls Unknown GW signals ls Noise Compact coalescing Core collapse binaries has known supernovae Moving lines theoretical waveforms Broad band noise Glitch noise Optimal filter: Matched No Optimal filter filter “Pattern recognition ” Too many templates to Parameters estimation by visual inspection test Elena Cuoco
11 Example of GW signals in Time-Frequency plots Elena Cuoco
12 https://www.zooniverse.org/projects/zooniverse/gravity-spy Example of Glitch signals Elena Cuoco
Example of other noise signals 13 I. Fiori courtesy Elena Cuoco
14 Numbers about data Data Stream Number of Number of Data on disk Flux events glitches • 50MB/s • 1-3PB • 1/week • 1/sec • 1/day? • 0.1/sec? Should be analysed in less than 1min Elena Cuoco
How Machine Learning can help 15 Data a conditio ionin ing g SignalDetecti tion/ on/Clas Classificat sification/ on/PE ▪ Non linear noise coupling A lot of fake signals due to noise ▪ ▪ Use Neural Network to learn ▪ Fast alert system noise ▪ Manage parameter estimation ▪ Use Neural Network to remove noise Elena Cuoco
16 What is going in the ML LIGO/Virgo group 136 LIGO/Virgo members 30 active projects Elena Cuoco
Example of interesting works 17 Noise Removal ▪ Non-linear and Labelling glitches: Gravity Spy ▪ non-stationary noise subtraction with Deep Learning G. Vajente courtesy S. Coughlin courtesy Elena Cuoco
18 Sign gnal al detection ion Hunter r Gabbard rd, Michael ael William ams, s, Fergus s Hayes, s, and Chris s Messe senge nger Phys. s. Rev. Lett. . 120, 0, 141103 03 Deep learning procedure requiring only the raw data time series as input with minimal signal pre-processing. Performance similar to Optimal Wiener Filter Elena Cuoco
Glitches Classification Strategy Elena Cuoco
Glitches classification efforts in 20 LIGO/Virgo Community Gravity Spy (M. Zevin,S. Coughlin,J. R. Deep Transfer Learning (Daniel George, ● ● Smith, A. Lundgren, D. Macleod, V. Hongyu Shen, E.A. Huerta) Kalogera) Gstlal-iDQ (P. Godwin, R. Essick, D. WDF-ML (E. Cuoco, A. Torres) ● Meacher, S. Chamberlain, C. Hanna, E. ● Katsavounidis, L. Wade, M. Wade, D. WDFX (E. Cuoco, M. Razzano, A. Utina) ● Moffa, K. Rose) PCAT (M.Cavagli à , D. Trifir ò ) ● New ranking statistic for gstlal (K. Kim, Karoo GP (K. Staats, M. Cavagli à ) ● T.G.F. Li, R.K.-L. Lo, S. Sachdev, R.S.H. ● Wavelet-DBNN (N. Mukund S. Abraham Yuen) ● S. Mitra et al) RGB image SN CNN (P. Astone, S. Frasca, ● ImageGlitch CNN (M. Razzano, E. Cuoco) C. Palomba, F. Ricci, M. Drago, I. Di ● Palma, F. Muciaccia, Pablo Cerda-Duran) Low latency transient detection and ● classification (I. Pinto, V. Pierro, L. Troiano, E. Mejuto-Villa, V. Matta, P. Addesso) Elena Cuoco
21 Image ges-based ased gl glitch h classi ssification ication Deep learning with CNN Massimiliano Razzano and Elena Cuoco 2018 Class. Quantum Grav. 35 095016 Elena Cuoco Elena Cuoco
Deep learning for Glitch Classification 22 Many approaches to data: we choose image classification of time frequency images The architecture is based on Convolutional deep Neural Networks (CNNs). CNNs are more complex than simple NNs but are optimized to catch features in images, so they are the best choice for image classification Elena Cuoco
Pipeline structure 23 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 Elena Cuoco
Test on simulation 24 24 To test the pipeline, Add 6 different we prepared ad-hoc classes of glitch simulations shapes Simulate colored noise using public H1 sensitivity curve Elena Cuoco
Simulated signal families 25 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 Elena Cuoco
Signal distribution 26 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 Elena Cuoco
Building the images 27 Spectrogram for each image 2-seconds time window to highlight fatures in long glitches Data is whitened Optional contrast stretch Elena Cuoco
Training the CNN 28 Datasets s of 14000 00 images Training/ g/va validati dation/ n/test est → 70/15/1 /15 Image size 241px x x 513px Reduced the images s by a factor r 0.55 5 due to memory ry constrain raints Use validati ation on set to tune hyperpa para ramet meters ers On our hardware, are, training time ~8 8 hrs hrs for r ~100 epochs s When training is done, classi sifi ficati ation on require res s ~1 1 ms ms/image ge (on our configura uration) n) Elena Cuoco
Classification Results 29 We compared classification performances with simpler architectures Linear Support Vector Machine CNN with 1 hidden layer CNN with one block (2 CNNs+Pooling&Dropout) Deep 4-blocks CNNs Elena Cuoco
Classification accuracy 30 Normalized Confusion Matrix SVM SVM Deep Deep CNN Deep CNN better at distinguishing similar morphologies Elena Cuoco
Example of classification results 31 Some cases of more glitches in the time window, always identify the right class 100% Sine-Gaussian Elena Cuoco
32 Wavelet Detection Filter (WDF) workflow Whitening Wavelet Parameter Trigger in time De-noising Data transform estimation list domain Elena Cuoco
Wavelet Detection Filter 33 Wavel elet et transform orm in the sele lecte cted window ow size Retain ain only ly coefficien cients above ve a fixed xed thresho eshod (Donoh oho-Joh Johnston on denois oise method od) Creat ate e a metric rics for the energ ergy y usin ing g the sele lecte cted coef efficien icients and give e back the trigg gger er with all the wavel velet et coeffic icien ients. In the e wavel elet et plane, e, sele lect ct the e highes est valu lues es and closest coeffic icien ients to build ld the e event Put to zero o all the e other er coefficien cients Invers erse e wavel elet et tran ansfor orm Estimat imate e mean and max frequen ency cy and snr r max of the e clean aned ed event Gps, duration, snr, snr@max, freq_mean, freq@max, wavelet type triggered + corresponding wavelets coefficients. Elena Cuoco
eXtreme Gradient Boosting 34 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, Tree Ensemble see also the Project Page at UW. 𝐿 𝑧 𝑜 = 𝑔 𝑙 𝑦 𝑜 𝑙=1 Elena Cuoco
35 Wavelet Detection Filter and XGBoost (WDFX) Supervised classification Elena Cuoco
Recommend
More recommend
