Talks

UK Research and Innovation is committed to achieving net zero emissions and to supporting environmental sustainability within the research it funds. This includes improving the sustainability of UKRI's Digital Research Infrastructure (DRI) and use of digital resources. This presentation will outline UKRI's approach to achieving this, focussing on the action being taken in four key areas: Funding, Procurement, Training and Engagement, and Monitoring and Reporting."

00:00:00 Slide 1
00:00:18 Slide 2
00:00:40 Slide 3
00:01:14 Slide 4
00:01:52 Slide 5
00:02:48 Slide 6
00:03:40 Slide 7
00:03:56 Slide 8
00:05:41 Slide 9
00:06:28 Slide 10
00:07:08 Slide 11
00:09:14 Slide 12
00:10:46 Slide 13
00:11:58 Slide 14
00:12:38 Slide 15
00:13:20 Slide 16
00:13:43 Slide 17
00:20:09 Slide 18

One of the main challenges in adopting green software practices is the limited understanding and resources on the topic. Which components have the most emissions? How can researchers minimise their footprint and engage others? We argue that structured community-building is a critical yet underexplored mechanism for accelerating green computing adoption. This talk highlights recent community-building activities in green computing, including the Environmentally Sustainable Computational Science forum that connects 170+ members on an online platform, the Green Algorithms Initiative that aims to study and improve carbon calculators and impact measurement tools, and the Green DiSC certification scheme providing a sustainability roadmap for digital research groups. Drawing on these experiences, we reflect on what has worked to grow and sustain engagement, share lessons on lowering participation barriers, and outline priorities for strengthening the green computing ecosystem ahead.

00:00:00 Slide 1
00:00:15 Slide 2
00:02:05 Slide 3
00:03:54 Slide 4
00:05:15 Slide 5
00:06:41 Slide 6
00:09:53 Slide 7
00:12:17 Slide 8
00:14:43 Slide 9
00:15:05 Slide 10
00:15:34 Slide 11
00:15:54 Slide 12
00:16:17 Slide 13

In the face of growing environmental impacts of computing, there is a legitimate request from researchers to identify what they can do about it. Green DiSC is an open access sustainability certification framework enabling researchers, labs and institutions to tackle the environmental impacts of their computing activities. It will be an opportunity to discuss the scheme, demonstrate how it can work, and showcase how it supports more environmentally sustainable computational research. We will also start by reflecting back on the inception of the scheme, and look ahead what's coming next.

00:00:00 Slide 1
00:00:18 Slide 2
00:02:08 Slide 3
00:02:54 Slide 4
00:03:27 Slide 5
00:04:24 Slide 6
00:05:05 Slide 7
00:05:41 Slide 8
00:07:18 Slide 9
00:07:36 Slide 10
00:07:54 Slide 11
00:08:54 Slide 12
00:09:23 Slide 13
00:10:45 Slide 14
00:11:06 Slide 15
00:11:31 Slide 16
00:12:01 Slide 17
00:12:38 Slide 18
00:13:02 Slide 19
00:13:41 Slide 20
00:13:58 Slide 21
00:14:31 Slide 22
00:14:53 Slide 23
00:18:57 Slide 24
00:19:16 Slide 25
00:23:52 Slide 26

UK Research and Innovation's Digital Research Infrastructure (UKRI DRI) is inherently heterogeneous as the computing resources have grown organically and are specialised for various disciplines. Sustainability approaches for UKRI DRI must therefore span this heterogeneous landscape. The IRIS Carbon Audit SnapshoT (IRISCAST) project estimated the total carbon costs of 6 UKRI DRI services over a 24 hour snapshot period. The IRIS Carbon Mapping Project (IRIS-CMP) then allocated carbon costs at 2 UKRI DRI services to individual service users. UKRI is currently funding the NetDRIVE project to gather evidence, make progress and make recommendation towards a NetZero UKRI DRI. This presentation highlights the key findings of the IRISCAST and IRIS-CMP Projects and re-interprets these in the light of the subsequent work of NetDRIVE and the NetDRIVE Working Group on Metrics and Reporting all in the context of the UKRI DRI.

00:00:00 Slide 1
00:01:13 Slide 2
00:02:20 Slide 3
00:02:44 Slide 4
00:03:58 Slide 5
00:06:49 Slide 6
00:07:15 Slide 7
00:08:09 Slide 8
00:09:15 Slide 9
00:10:08 Slide 10
00:11:01 Slide 11
00:13:37 Slide 12
00:15:08 Slide 13
00:15:29 Slide 14
00:15:56 Slide 15
00:16:33 Slide 16
00:16:43 Slide 17
00:17:08 Slide 18
00:17:32 Slide 19
00:18:35 Slide 20
00:20:00 Slide 21
00:28:00 Slide 22

As the global energy demand of data centers continues to increase, reducing their environmental impacts has become a major challenge. This presentation provides an environmental assessment of the LHCb data centers using a Life Cycle Assessment (LCA) approach, in line with CERN's sustainability objectives. The presentation will include the contribution of the different components of the data centers and a comparison of several data center cooling systems currently in development or already deployed in the LHCb experiment. The objective is to assess the environmental impacts associated with these systems and to identify the main contributors to the overall environmental performance of the infrastructure.

00:00:00 Slide 1
00:00:25 Slide 2
00:00:51 Slide 3
00:01:30 Slide 4
00:02:43 Slide 5
00:03:24 Slide 6
00:03:57 Slide 7
00:04:52 Slide 8
00:05:46 Slide 9
00:06:26 Slide 10
00:07:28 Slide 11
00:07:49 Slide 12
00:08:20 Slide 13
00:09:10 Slide 14
00:10:32 Slide 15
00:10:55 Slide 16
00:11:12 Slide 17
00:12:15 Slide 18
00:13:12 Slide 19
00:14:02 Slide 20
00:14:44 Slide 21
00:16:00 Slide 22
00:16:40 Slide 23
00:17:26 Slide 24
00:18:01 Slide 25
00:18:34 Slide 26
00:19:55 Slide 27
00:20:05 Slide 28
00:20:49 Slide 29
00:21:58 Slide 30
00:22:27 Slide 31
00:23:40 Slide 32

It is encouraging to see more and more studies published about the environmental footprint of the ICT sector. Unfortunately, the outcomes of those studies are often misinterpreted. In fact, one can look at the footprint of a product or activity in many different ways which all make sense but serve different purposes. It is very easy to mistake one purpose for another and thus derive completely wrong conclusions, which may lead to harmful—albeit well-intentioned—decision-making. I believe we can avoid those misunderstandings by clarifying the different methodological choices and their corresponding purpose. This mental framework helps draw correct conclusions from the growing corpus of sustainability studies. In this presentation, I summarize what I currently see as three of the most important methodological choices. Then, I'll discuss a couple of examples from computer networks (my area of research) to illustrate how easy it is to misinterpret footprint numbers.

00:00:00 Slide 1
00:00:59 Slide 2
00:01:32 Slide 3
00:01:45 Slide 4
00:02:05 Slide 5
00:02:37 Slide 6
00:03:15 Slide 7
00:03:44 Slide 8
00:04:07 Slide 9
00:04:47 Slide 10
00:05:04 Slide 11
00:05:31 Slide 12
00:07:08 Slide 13
00:07:35 Slide 14
00:09:04 Slide 15
00:10:04 Slide 16
00:11:34 Slide 17
00:12:49 Slide 18

This interactive workshop, "IT in Your Research – What Impacts?", aims to raise awareness among researchers about the environmental footprint of their research activities. Targeting group of 20–30 participants, the workshop combines theoretical foundations with hands on practice, encouraging attendees to analyze their own research projects or publications. The session begins with an introduction to Life Cycle Assessment (LCA), covering manufacturing, distribution, use, and end of life phases. Participants explore current challenges in environmental data availability, especially for IT hardware, and discuss differences between self hosted and cloud based computing. Through guided group work and the use of dedicated tools — such as Green Algorithms, cloud impact estimators, and GenAI footprint tools — participants assess their research-related IT impacts and identify opportunities to reduce them.

00:00:00 Slide 1
00:00:52 Slide 2
00:11:11 Slide 3
00:14:04 Slide 4
00:37:57 Slide 5

Evaluating the energy consumption of software is inherently complex, as software itself does not consume energy directly; rather, it is the hardware on which it runs that does. Over the past few years, numerous tools and models have emerged to estimate energy consumption at various levels, such as servers, containers, processes, and transactions. When assessing the energy consumption of software, it is crucial to understand the accuracy of these tools and models. This session presents the findings of experimental evaluations that assess the precision of various tools and models for this purpose. We particularly focus on those that estimate energy consumption at the container, process, and transaction levels.

00:00:00 Slide 1
00:01:44 Slide 2
00:05:16 Slide 3
00:07:20 Slide 4
00:12:02 Slide 5
00:13:35 Slide 6
00:14:10 Slide 7
00:14:42 Slide 8
00:15:22 Slide 9
00:18:22 Slide 10
00:19:32 Slide 11
00:22:34 Slide 12
00:23:07 Slide 13
00:23:23 Slide 14
00:23:34 Slide 15
00:24:06 Slide 16
00:24:44 Slide 17
00:25:19 Slide 18
00:26:18 Slide 19
00:26:35 Slide 20
00:27:30 Slide 21
00:28:13 Slide 22
00:28:35 Slide 23
00:29:00 Slide 24
00:30:02 Slide 25
00:30:30 Slide 26
00:30:46 Slide 27
00:31:12 Slide 28
00:31:29 Slide 29
00:31:44 Slide 30

Third-generation ground-based gravitational wave detectors such as the Einstein Telescope are expected to significantly advance our understanding of compact binary coalescences. One of the most critical challenges in data analysis for the Einstein Telescope is that of overlapping signals. With a tenfold improvement in sensitivity, the Einstein Telescope will be able to detect binary black hole and binary neutron star coalescences with expected rates of up to ~10⁵ events per year. Moreover, the extended range toward lower frequencies will allow the detector to observe these signals for longer durations compared to current-generation detectors. While this creates the opportunity to deepen our knowledge of these sources, detectable signals will inevitably overlap. This poses a severe challenge to parameter estimation analysis pipelines. We need a faster, unbiased parameter estimation strategy. In this talk, we will describe a promising solution to address this challenge: a deep learning approach that combines the power of two state-of-the-art machine learning architectures, Transformers and Normalizing Flows. In particular, we present the first application of a Transformer encoder for gravitational wave data analysis. This architecture is capable of capturing complex, varying-range dependencies, and we use it to extract the information in the data. We then employ Normalizing Flows to estimate the high-dimensional posterior distributions of the overlapped signals. We will present the results from training this network architecture, demonstrating its effectiveness in handling three overlapping signals simultaneously, and discussing how this deep learning method represents a promising solution to the problem, along with its potential extensions and improvements.

00:00:00 Slide 1
00:00:39 Slide 2
00:01:07 Slide 3
00:02:47 Slide 4
00:03:55 Slide 5
00:06:06 Slide 6
00:06:58 Slide 7
00:07:09 Slide 8
00:07:36 Slide 9
00:08:37 Slide 10
00:09:23 Slide 11
00:10:13 Slide 12
00:11:15 Slide 13
00:11:32 Slide 14
00:12:07 Slide 15
00:12:58 Slide 16
00:13:42 Slide 17
00:14:07 Slide 18
00:14:41 Slide 19
00:15:25 Slide 20
00:16:25 Slide 21
00:16:50 Slide 22
00:18:11 Slide 23
00:18:27 Slide 24
00:19:06 Slide 25
00:19:58 Slide 26
00:20:28 Slide 27
00:22:00 Slide 28
00:23:38 Slide 29

We present a framework for probing the full geometry of Bayesian posteriors in inverse problems through a noise-conditioned homotopy. By embedding the likelihood in a one-parameter family controlled by a noise-scaling parameter, we construct a continuous deformation from an almost deterministic posterior concentrated at the true parameters to the full noisy posterior. Traversing this path reveals how posterior structure evolves with measurement quality: when multi-modality emerges, where Gaussian approximations break down, and how parameter degeneracies develop. We argue this constitutes a more integrated alternative to Fisher-information analyses, which becomes beneficial especially in multimodal geometries. Additionally, deviations from smooth homotopy behaviour provide direct diagnostics of inference pipelines, allowing identification of spurious correlations, mode-collapse artefacts, and approximation breakdowns. We discuss the framework as a general validation and benchmarking tool for simulation-based inference methods.

00:00:00 Slide 1
00:00:28 Slide 2
00:00:49 Slide 3
00:01:13 Slide 4
00:01:46 Slide 5
00:02:18 Slide 6
00:03:39 Slide 7
00:04:50 Slide 8
00:06:24 Slide 9
00:07:02 Slide 10
00:07:48 Slide 11
00:08:54 Slide 12
00:10:57 Slide 13
00:12:43 Slide 14
00:13:47 Slide 15
00:14:22 Slide 16
00:16:27 Slide 17
00:17:16 Slide 18
00:17:51 Slide 19
00:18:12 Slide 20
00:18:40 Slide 21
00:19:01 Slide 22
00:20:58 Slide 23
00:21:43 Slide 24

Extreme-mass-ratio inspirals (EMRIs) are key gravitational-wave sources for the Laser Interferometer Space Antenna (LISA), but their detection and parameter inference are computationally challenging due to the extreme concentration of posterior distributions within vast prior volumes. In this work, we introduce a novel divide-and-conquer strategy that reformulates global inference as a hierarchical identification problem. Our approach iteratively localizes the posterior mode through a coarse-to-fine procedure based on ordinal classification, progressively restricting the parameter space while preserving the true signal region. A transformer-based neural network is trained at each stage to identify the most probable parameter subregions, enabling exponential reduction of the search volume with only a few refinement steps. Once the parameter space is reduced to the Fisher-information scale, standard local sampling methods efficiently recover the full joint posterior. We demonstrate that this method achieves rapid and accurate intrinsic parameter estimation for EMRIs in simulated LISA data, dramatically reducing computational costs compared to traditional global sampling techniques. This framework provides a scalable and efficient pathway for real-time EMRI detection and inference in the LISA era.

00:00:00 Slide 1
00:00:17 Slide 2
00:00:32 Slide 3
00:02:01 Slide 4
00:02:37 Slide 5
00:03:09 Slide 6
00:03:25 Slide 7
00:04:13 Slide 8
00:04:34 Slide 9
00:05:33 Slide 10
00:06:11 Slide 11
00:07:13 Slide 12
00:07:47 Slide 13
00:08:22 Slide 14
00:09:15 Slide 15
00:10:14 Slide 16
00:10:43 Slide 17
00:11:09 Slide 18
00:11:22 Slide 19
00:12:24 Slide 20
00:12:39 Slide 21
00:13:05 Slide 22
00:17:26 Slide 23

Gravitational-wave (GW) astronomy promises groundbreaking discoveries in the coming decades, but its progress is bottlenecked by the computational challenges of large-scale and real-time data analysis. I will present DINGO, a machine learning approach for fast and accurate GW inference that addresses these challenges. DINGO trains generative neural networks to directly estimate probability distributions over GW source parameters. I will explain the core ideas behind DINGO and highlight several machine learning techniques that we developed to adapt modern simulation-based inference to the challenging field of GW data analysis.

00:00:00 Slide 1
00:00:29 Slide 2
00:00:44 Slide 3
00:01:03 Slide 4
00:01:29 Slide 5
00:03:07 Slide 6
00:04:59 Slide 7
00:06:06 Slide 8
00:07:31 Slide 9
00:07:57 Slide 10
00:09:09 Slide 11
00:09:50 Slide 12
00:10:03 Slide 13
00:10:54 Slide 14
00:11:28 Slide 15
00:12:01 Slide 16
00:12:14 Slide 17
00:13:37 Slide 18
00:14:52 Slide 19
00:15:30 Slide 20
00:15:52 Slide 21
00:16:39 Slide 22
00:16:59 Slide 23
00:19:04 Slide 24