Talks

Pulsar Timing Array data analysis faces severe computational challenges as parameter spaces scale with the number of pulsars. I present two Normalizing Flows (NFs) based strategies to accelerate and improve Bayesian inference for stochastic gravitational wave background (SGWB). First, integrating NFs into the importance nested sampling framework i-nessai yields speedups of one to three orders of magnitude over standard methods, with robust posteriors and reliable evidence estimates. Second, a dual NFs architecture implementing parameter decorrelation via orthogonal projection disentangles pulsar noise from hyperprior parameters in hierarchical Bayesian modeling, enhancing noise constraining power even in the presence of a SGWB signal.

00:00:00 Slide 1
00:00:08 Slide 2
00:00:19 Slide 3
00:00:43 Slide 4
00:01:49 Slide 5
00:02:29 Slide 6
00:02:48 Slide 7
00:03:44 Slide 8
00:05:14 Slide 9
00:06:00 Slide 10
00:06:49 Slide 11
00:07:52 Slide 12
00:08:59 Slide 13
00:09:36 Slide 14
00:10:24 Slide 15
00:11:04 Slide 16
00:13:42 Slide 17
00:15:19 Slide 18
00:16:24 Slide 19

Scientific progress increasingly depends on powerful computing infrastructure, yet its environmental impacts across the full life cycle are often overlooked. This study evaluates the energy and resource efficiency of computing hardware and software within the Faculty of Science and Engineering (FSE) at the University of Manchester, informing the university's sustainability targets, including Net Zero by 2050. Using life cycle assessment (LCA), hardware audits, and software energy profiling, the study compares three computing setups: desktop computers, laptops, and high-throughput computing cluster nodes. System boundaries cover component production and operational use across 15-year scenarios. Specialised compute nodes dominate most impact categories, particularly those driven by manufacturing. Across nearly all indicators, embodied impacts exceed operational electricity use, highlighting the growing significance of manufacturing burdens despite grid decarbonisation. Scenario analyses demonstrate that replacement strategies strongly influence outcomes, underlining the importance of demand management and procurement policies in reducing academic computing's environmental footprint.

00:00:00 Slide 1
00:00:37 Slide 2
00:01:13 Slide 3
00:02:18 Slide 4
00:03:22 Slide 5
00:04:47 Slide 6
00:06:06 Slide 7
00:07:09 Slide 8
00:08:05 Slide 9
00:08:24 Slide 10
00:09:02 Slide 11
00:09:49 Slide 12
00:11:56 Slide 13
00:13:23 Slide 14
00:15:05 Slide 15
00:16:02 Slide 16
00:17:19 Slide 17
00:18:35 Slide 18
00:19:56 Slide 19
00:20:51 Slide 20
00:21:15 Slide 21
00:21:27 Slide 22
00:23:03 Slide 23
00:23:22 Slide 24
00:23:35 Slide 25
00:24:12 Slide 26
00:25:54 Slide 27
00:26:47 Slide 28
00:26:58 Slide 29
00:27:51 Slide 30

The generation of accurate waveforms from binary black hole (BBH) mergers is a major effort in Gravitational-Wave Astronomy. In recent years, machine-learning-based surrogate models for BBH waveforms have been proposed. Those offer the potential to dramatically accelerate waveform generation while maintaining accuracy competitive with that of traditional waveform approximants. In this work, we investigate the viability of autoencoders as generative models for gravitational-wave signals from quasi-circular BBH mergers. We introduce AESur3dq8, a novel surrogate waveform model based on autoencoders that enables the rapid and accurate construction of large template banks, producing millions of waveforms in under a second using modest computational resources. The model is trained on the numerical-relativity-informed surrogate NRHybSur3dq8 and subsequently fine-tuned using the SXS catalog of BBH simulations. We demonstrate that waveforms generated by AESur3dq8 achieve mismatches of order $10^{-4}$ with respect to Numerical Relativity waveforms, and that parameter estimation performed with these templates yields results fully consistent with those reported by the LIGO-Virgo-KAGRA Collaboration for observed gravitational-wave events.

00:00:00 Slide 1
00:00:25 Slide 2
00:02:14 Slide 3
00:02:27 Slide 4
00:03:10 Slide 5
00:04:56 Slide 6
00:05:06 Slide 7
00:05:53 Slide 8
00:06:59 Slide 9
00:07:48 Slide 10
00:09:25 Slide 11
00:10:46 Slide 12
00:11:49 Slide 13
00:13:26 Slide 14
00:14:07 Slide 15
00:14:35 Slide 16
00:16:14 Slide 17

The detection of gravitational waves by the Laser Interferometer Gravitational-Wave Observatory (LIGO) has opened a new window onto the universe, but the sensitivity of these detectors is fundamentally limited by a complex and evolving landscape of instrumental and environmental noise. In recent years, machine learning has emerged as a powerful tool for understanding, characterizing, and mitigating these noise sources. In this talk, I will present current applications of machine learning in LIGO detector characterization, with a focus on transient noise identification and its impact on search analyses. By leveraging techniques from modern computer vision, including object detection and segmentation models such as YOLO, we are able to automatically identify and localize noise artifacts in time–frequency representations of detector data. These approaches enable scalable classification of noise transients, improved data quality vetting, and more robust separation of astrophysical signals from noise, directly enhancing the sensitivity and reliability of gravitational-wave searches. Looking ahead, I will discuss emerging directions where machine learning can play a transformative role beyond data analysis. These include integrating ML into low-latency pipelines, developing physics-informed models for noise prediction and subtraction, and exploring the use of ML-driven optimization in the design of future interferometers. In particular, data-driven approaches may help guide the selection of design parameters for next-generation cavities and instruments, bridging the gap between detector characterization and instrument development. Together, these efforts highlight how machine learning is becoming an integral component of both current LIGO operations and the future of gravitational-wave instrumentation.

00:00:00 Slide 1
00:00:44 Slide 2
00:01:05 Slide 3
00:01:58 Slide 4
00:03:40 Slide 5
00:04:18 Slide 6
00:04:57 Slide 7
00:05:59 Slide 8
00:07:08 Slide 9
00:08:04 Slide 10
00:08:46 Slide 11
00:09:54 Slide 12
00:10:22 Slide 13
00:11:21 Slide 14
00:11:34 Slide 15
00:11:55 Slide 16
00:12:59 Slide 17
00:14:06 Slide 18
00:14:53 Slide 19
00:15:56 Slide 20
00:16:29 Slide 21
00:16:57 Slide 22
00:17:31 Slide 23

This project builds on DeepClean, a machine learning framework for subtracting instrumental and environmental noise from gravitational-wave detector data using auxiliary "witness" channels. While effective, DeepClean's convolutional architecture limits its ability to scale across varying channel configurations and frequency bands. We propose a hybrid CNN–transformer architecture that maintains permutation invariance across channels. The model combines a per-channel CNN for independent feature extraction with a transformer that treats time steps as tokens and models cross-channel interactions via self-attention, while preserving channel-specific representations. This design aims to reduce retraining requirements under changing detector configurations, such as varying channel sets or frequency bands. Preliminary results show performance comparable to DeepClean on standard denoising benchmarks, indicating that this added flexibility does not compromise accuracy. Ongoing work explores improved performance across broader frequency bands and increased robustness to imperfect or noisy channel selection. Ultimately, this project aims to move toward a more automated version of DeepClean, incorporating learning-based channel selection tailored to the denoising task.

00:00:00 Slide 1
00:00:44 Slide 2
00:01:05 Slide 3
00:01:58 Slide 4
00:03:40 Slide 5
00:04:18 Slide 6
00:04:57 Slide 7
00:05:59 Slide 8
00:07:08 Slide 9
00:08:04 Slide 10
00:08:46 Slide 11
00:09:54 Slide 12
00:10:22 Slide 13
00:11:21 Slide 14
00:11:34 Slide 15
00:11:55 Slide 16
00:12:59 Slide 17
00:14:06 Slide 18
00:14:53 Slide 19
00:15:56 Slide 20
00:16:29 Slide 21
00:16:57 Slide 22
00:17:31 Slide 23

The coming decade will be crucial for determining the final design and configuration of a global network of next-generation (XG) gravitational-wave (GW) detectors, including the Einstein Telescope (ET) and Cosmic Explorer (CE). In this study and for the first time, we assess the performance of various network configurations using neural posterior estimation (NPE) implemented in Dingo-IS-a method based on normalizing flows and importance sampling that enables fast and accurate inference. We focus on a specific science case involving short-duration, massive and high-redshift binary black hole (BBH) mergers with detector-frame chirp masses ($\mathcal{M}_\mathrm{d}$) > 100 M$_\odot$. These systems encompass early-Universe stellar and primordial black holes, as well as intermediate-mass black-hole binaries, for which XG observatories are expected to deliver major discoveries. Validation against standard Bayesian inference demonstrates that NPE robustly reproduces complex and disconnected posterior structures across all network configurations. For a network of two misaligned L-shaped ET detectors (2L MisA), the posterior distributions on luminosity distance can become multimodal and degenerate with the sky position, leading to less precise distance estimates compared to the triangular ET configuration. However, the number of sky-location multimodalities is substantially lower than the eight expected with the triangular ET, resulting in improved sky and volume localization. Adding CE to the network further reduces sky-position degeneracies, and the better performance of the 2L MisA configuration over the triangle remains evident.

00:00:00 Slide 1
00:00:45 Slide 2
00:01:06 Slide 3
00:01:26 Slide 4
00:01:50 Slide 5
00:02:58 Slide 6
00:03:57 Slide 7
00:04:41 Slide 8
00:05:54 Slide 9
00:06:45 Slide 10
00:07:38 Slide 11
00:08:23 Slide 12
00:08:42 Slide 13
00:09:26 Slide 14
00:09:55 Slide 15
00:10:08 Slide 16
00:10:30 Slide 17
00:11:28 Slide 18
00:12:34 Slide 19

The number of GW events have increased from two real-time detections in the LIGO first observing run, to over two hundred in the LIGO-Virgo-KAGRA fourth observing run. In parallel, the last decade has also seen the increased use of machine learning, especially neural networks, in science. For the first time, after a decade of discovery, binary black holes (BBHs) are routinely detected by neural-networks as a part of the LVK data analysis. A total of 23 BBH events were detected in real-time by neural-network based search, Aframe, between late August to mid November 2025. Sky-localization and chirp mass estimates have been distributed in low latency using neural network based parameter estimation algorithm AMPLFI. I will be talking about Aframe and AMPLFI with a view toward this new paradigm of doing low latency GW science using AI.

00:00:00 Slide 1
00:00:04 Slide 2
00:00:33 Slide 3
00:02:18 Slide 4
00:03:19 Slide 5
00:04:50 Slide 6
00:05:25 Slide 7
00:05:58 Slide 8
00:08:30 Slide 9
00:10:19 Slide 10
00:11:36 Slide 11
00:12:27 Slide 12
00:13:59 Slide 13
00:14:17 Slide 14
00:15:22 Slide 15
00:16:24 Slide 16
00:17:59 Slide 17
00:19:24 Slide 18
00:20:12 Slide 19
00:21:32 Slide 20
00:24:50 Slide 21
00:25:18 Slide 22
00:25:52 Slide 23
00:26:27 Slide 24
00:26:55 Slide 25
00:27:13 Slide 26
00:28:01 Slide 27
00:28:21 Slide 28
00:28:59 Slide 29
00:29:13 Slide 30
00:29:34 Slide 31
00:31:02 Slide 32
00:33:58 Slide 33
00:34:41 Slide 34
00:35:02 Slide 35
00:35:32 Slide 36
00:36:28 Slide 37
00:37:02 Slide 38
00:38:26 Slide 39
00:39:11 Slide 40
00:39:27 Slide 41
00:40:27 Slide 42
00:40:43 Slide 43
00:42:03 Slide 44
00:44:28 Slide 45
00:45:06 Slide 46
00:45:30 Slide 47
00:45:53 Slide 48
00:49:05 Slide 49
00:49:40 Slide 50
00:50:04 Slide 51
00:58:29 Slide 52
00:58:46 Slide 53
00:59:38 Slide 54
01:01:15 Slide 55
01:02:22 Slide 56
01:02:51 Slide 57
01:03:37 Slide 58
01:03:55 Slide 59
01:04:43 Slide 60
01:05:30 Slide 61
01:05:46 Slide 62
01:06:06 Slide 63
01:06:28 Slide 64
01:06:40 Slide 65
01:07:18 Slide 66
01:07:49 Slide 67
01:11:27 Slide 68

There is a consensus in the sustainable computing community that understanding the carbon footprint of computing is vital. Despite this, wide-spread adoption of emissions monitoring has not yet been realised. In this interactive session, we will share recent work from STFC's Scientific Computing Department which has focused on measuring and reporting user carbon emissions in different contexts, as well as plans for an upcoming NetDRIVE-supported (https://uknetdrive.org/) community project to develop training materials on this topic. We will invite participants to discuss existing initiatives, share experiences, and help identify key education gaps and barriers with respect to adopting emissions monitoring tools. The aim of this session will be to consolidate ideas to maximise the impact of future training courses and materials for all stakeholders across compute infrastructure.

00:00:00 Slide 1
00:01:46 Slide 2
00:02:11 Slide 3
00:03:04 Slide 4
00:03:18 Slide 5
00:04:30 Slide 6
00:04:56 Slide 7
00:06:03 Slide 8
00:06:28 Slide 9
00:06:56 Slide 10
00:07:27 Slide 11
00:08:56 Slide 12
00:10:08 Slide 13
00:10:57 Slide 14
00:11:39 Slide 15
00:12:07 Slide 16
00:12:58 Slide 17
00:13:11 Slide 18
00:13:28 Slide 19
00:13:56 Slide 20
00:14:47 Slide 21
00:15:55 Slide 22
00:19:31 Slide 23
00:22:56 Slide 24
00:23:26 Slide 25
00:23:44 Slide 26
00:24:14 Slide 27
00:25:21 Slide 28
00:25:46 Slide 29
00:26:23 Slide 30
00:26:49 Slide 31
00:28:22 Slide 32
00:29:11 Slide 33
00:29:44 Slide 34
00:30:14 Slide 35
00:31:29 Slide 36
00:31:41 Slide 37
00:32:19 Slide 38
00:32:49 Slide 39
00:33:02 Slide 40
00:33:32 Slide 41
00:35:11 Slide 42
00:36:03 Slide 43
00:37:54 Slide 44
00:38:20 Slide 45
00:39:20 Slide 46
00:39:34 Slide 47

Building tools that monitor the environmental cost of computational work is only the first challenge. The harder question is: what should they actually show, and to whom, and in what form, to make a genuine difference? This interactive session uses the Green Algorithms Dashboard—a daily monitoring tool for computational resource usage and estimated carbon footprint of HPC systems—as a working example. Participants won't just interrogate what the tool currently does, but what research computing carbon monitoring should do. After a brief introduction of the tool, participants will be grouped and given a scoped, technically grounded problem, working toward a one-page proposal covering the problem, available data, and a path forward. The aim is a community-informed roadmap, openly published after the conference. If you submit jobs to a cluster, write research code, or simply want research computing to have a smaller footprint, this is your chance to shape what comes next.

00:00:00 Slide 1
00:00:44 Slide 2
00:01:36 Slide 3
00:02:56 Slide 4
00:03:07 Slide 5
00:03:20 Slide 6
00:05:52 Slide 7
00:08:23 Slide 8
00:09:22 Slide 9
00:10:15 Slide 10
00:13:10 Slide 11
00:14:56 Slide 12
00:15:34 Slide 13
00:27:07 Slide 14
00:29:22 Slide 15
00:29:46 Slide 16
00:31:07 Slide 17
00:31:29 Slide 18
00:32:59 Slide 19
00:33:26 Slide 20
00:34:08 Slide 21
00:34:57 Slide 22
00:43:37 Slide 23

A sustainable research computing culture is vital to ensure research institutions can meet their sustainability aims while supporting ever-increasing demands for computational capacity driven by modern research practices. This talk, and the associated poster, will outline work underway at Imperial College London to foster a sustainable institutional research computing culture and create models that can be adopted elsewhere. Our approach centres on three pillars: Community: Building community to develop shared understanding among researchers and technical professionals of sustainable computing practices, particularly as AI and high performance computing become increasingly prominent. Education: Underpinned by the development of a new green and sustainable computing training course that delivers core knowledge and highlights best practices, providing a resource for cultural change. Advocacy: Involves working with institutional leaders and sustainability teams to highlight practical improvements, large and small, alongside work with an artist to develop visual outputs that inspire broader engagement with sustainable IT.

00:00:00 Slide 1
00:01:11 Slide 2
00:01:53 Slide 3
00:04:14 Slide 4
00:05:55 Slide 5
00:09:09 Slide 6
00:11:25 Slide 7
00:12:16 Slide 8
00:13:53 Slide 9
00:15:38 Slide 10
00:16:50 Slide 11
00:17:45 Slide 12
00:18:34 Slide 13
00:18:58 Slide 14
00:20:55 Slide 15
00:22:23 Slide 16
00:23:53 Slide 17
00:25:02 Slide 18
00:25:36 Slide 19
00:26:53 Slide 20
00:28:32 Slide 21
00:29:59 Slide 22
00:31:42 Slide 23
00:33:37 Slide 24
00:35:05 Slide 25
00:35:56 Slide 26