[Paper Review] Deep Learning-based Search for Microlensing Signature from Binary Black Hole Events in GWTC-1 and -2

Kyungmin Kim, J. Lee+2 more|arXiv (Cornell University)|Jun 16, 2022

Pulsars and Gravitational Waves Research4 citations

TL;DR

This study presents the first deep learning-based search for microlensing signatures in gravitational waves from binary black hole (BBH) events in GWTC-1 and GWTC-2. Using a VGG-19-based convolutional neural network trained on spectrograms of simulated signals, the method classifies events as lensed or unlensed; although GW190707_093326 shows a high median probability of 0.984 for being lensed, uncertainty analysis suggests it is unlikely to be a true lensed signal, leading to the conclusion that no significant microlensing signature was detected.

ABSTRACT

We present the result of the first deep learning-based search for the signature of microlensing in gravitational waves. This search seeks the signature induced by lenses with masses between $10^3M_\odot$--$10^5M_\odot$ from spectrograms of the binary black hole events in the first and second gravitational-wave transient catalogs. We use a deep learning model trained with spectrograms of simulated noisy gravitational-wave signals to classify the events into two classes, lensed or unlensed. We introduce ensemble learning and a majority voting-based consistency test for the predictions of ensemble learners. The classification scheme of this search primarily classifies one event, GW190707_093326, into the lensed class. To verify the primary classification of this event, we also examine the median probability to the lensed class and observe the resulting value, $0.984^{+0.012}_{-0.342}$, agrees with an empirical criterion $>\!0.6$ for claiming the detection of a lensed signal. However, the uncertainty of the estimated $p$-value for the median probability and error, ranging from 0 to 0.1, convinces us GW190707_093326 is less likely a lensed event because it includes $p\!\geq\!0.05$ where the unlensed hypothesis is true. Therefore, we conclude our search finds no significant evidence of microlensing signature from the evaluated binary black hole events.

Motivation & Objective

• To detect gravitational wave microlensing signatures from binary black hole events in GWTC-1 and GWTC-2 using deep learning.
• To address the challenge of identifying weak, beat-pattern signatures from microlensing in noisy gravitational wave data.
• To develop and validate an ensemble deep learning framework capable of distinguishing lensed from unlensed signals with high reliability.
• To test whether existing BBH events show statistically significant evidence of microlensing via probabilistic classification.

Proposed method

• A deep learning model based on VGG-19 is trained on spectrograms of simulated gravitational wave signals with and without microlensing effects.
• The model classifies events into two classes: lensed or unlensed, using spectrogram images as input.
• Ensemble learning is employed by training 10 independent models to improve robustness and reduce overfitting.
• Majority voting and consistency testing across ensemble members are applied to enhance prediction reliability.
• The median probability to the lensed class is computed across ensemble predictions to assess confidence in classification.
• Statistical uncertainty analysis of the median probability is performed to evaluate the validity of lensed event claims.

Experimental results

Research questions

• RQ1 Can deep learning detect microlensing signatures in gravitational wave signals from binary black hole mergers in GWTC-1 and GWTC-2?
• RQ2 Is GW190707_093326 a statistically significant candidate for microlensing, based on deep learning classification?
• RQ3 How reliable are ensemble deep learning predictions in identifying weak microlensing-induced beat patterns in noisy data?
• RQ4 Does the uncertainty in the median probability support or reject the lensed hypothesis for candidate events?
• RQ5 Can the proposed method distinguish microlensing effects from intrinsic signal features in BBH events?

Key findings

• The deep learning model achieved high performance in distinguishing lensed from unlensed signals, with all ensemble learners showing >92% efficiency and area under the ROC curve >0.99.
• GW190707_093326 received the highest median probability of 0.984 to be lensed, exceeding the empirical threshold of 0.6.
• Despite the high probability, the uncertainty range of the median probability (0.0 to 0.1) includes values where the unlensed hypothesis is valid (p ≥ 0.05), indicating low confidence in the lensed classification.
• The ensemble model showed no signs of overfitting or underfitting, as training and validation loss curves converged smoothly.
• The search found no statistically significant evidence of microlensing in any of the 46 BBH events from GWTC-1 and GWTC-2.
• The study concludes that current data from GWTC-1 and GWTC-2 do not support the presence of detectable microlensing signatures from lenses with masses between 10³ and 10⁵ M☉.