Hardware-accelerated Inference for Real-Time Gravitational-Wave Astronomy

Abstract

The field of transient astronomy has seen a revolution with the first gravitational-wave detections and the arrival of multi-messenger observations they enabled. Transformed by the first detection of binary black hole and binary neutron star mergers, computational demands in gravitational-wave astronomy are expected to grow by at least a factor of two over the next five years as the global network of kilometer-scale interferometers are brought to design sensitivity. With the increase in detector sensitivity, real-time delivery of gravitational-wave alerts will become increasingly important as an enabler of multi-messenger followup. In this work, we report a novel implementation and deployment of deep learning inference for real-time gravitational-wave data denoising and astrophysical source identification. This is accomplished using a generic Inference-as-a-Service model that is capable of adapting to the future needs of gravitational-wave data analysis. Our implementation allows seamless incorporation of hardware accelerators and also enables the use of commercial or private (dedicated) as-a-service computing. Based on our results, we propose a paradigm shift in low-latency and offline computing in gravitational-wave astronomy. Such a shift can address key challenges in peak-usage, scalability and reliability, and provide a data analysis platform particularly optimized for deep learning applications. The achieved sub-millisecond scale latency will also be relevant for any machine learning-based real-time control systems that may be invoked in the operation of near-future and next generation ground-based laser interferometers, as well as the front-end collection, distribution and processing of data from such instruments.

Inference-optimized AI and high performance computing for gravitational wave detection at scale

“…For instance, PyTorch models for AI forecasting of binary neutron star and black hole-neutron star systems were quantized to reduce their size by 4X and accelerate their speed 2.5X for rapid inference at the edge [62]. Furthermore, the combination of TensorRT AI models for data cleaning, and AI models for black hole detection under the umbrella of a generic inference as a service model that leverages HPC, private or dedicating computing was introduced in [45]. On the other hand, this work is the first in the literature to combine TensorRT AI models for accelerated signal detection with HPC at scale to process one month of advanced LIGO strain data from Hanford and Livingston within 50 seconds using an ensemble of 4 TensorRT AI models.…”

“…As described in recent reviews [25,26], AI and high performance computing (HPC) as well as edge computing have been showcased to enable gravitational wave detection with the same sensitivity than template-matching algorithms, but orders of magnitude faster and at a fraction of the computational cost. At a glance, recent AI applications for gravitational wave astrophysics includes classification or signal detection [27][28][29][30][31][32][33][34][35][36][37][38][39][40][41][42][43][44][45], signal denoising and data cleaning [46][47][48][49], regression or parameter estimation [50][51][52][53][54][55][56][57], accelerated waveform production [58,59], signal forecasting [60,61], and early warning systems for gravitational wave sources that include matter, such as binary neutron stars or black hole-neutron star systems [62][63][64].…”

A Software Ecosystem for Deploying Deep Learning in Gravitational Wave Physics

“…cloudbreak is not currently used as part of the DeepClean production pipeline, but will form a critical part of future large-scale offline experiments. online averaging of DeepClean's outputs on the server-side, streaming back non-overlapping averaged predictions for each segment and avoiding the drawbacks of the "fully online" inference scenario depicted in figure 14 of [21]. More information on this will be given in section 5.3.…”

“…As described in [21], the streaming nature of gravitational-wave data complicates the use of inference services. Overlapping data in both the input and output kernels introduces extra data transfer overhead which is linear in the inference sampling rate 𝑟 .…”

AI and extreme scale computing to learn and infer the physics of higher order gravitational wave modes of quasi-circular, spinning, non-precessing binary black hole mergers

“…This approach has been embraced and further developed by an international community of researchers [5][6][7]. To this date, AI has been explored for a variety of signal processing tasks, including detection [8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24], gravitational wave denoising and data cleaning [25][26][27], parameter estimation [28][29][30][31][32][33][34], rapid waveform production [35,36], waveform forecasting [37,38], and early warning systems for multi-messenger sources [39][40][41], among others.…”