Past events

In this session, three EPCC speakers shared updates on research and infrastructure:

• Kara Moraw - AI-driven climate simulation. AI models such as ECMWF's AIFS can emulate traditional climate simulations at a fraction of the compute cost, running on a single EIDF GPU. Takeaway: AI could make climate and weather modelling significantly more sustainable and accessible.
• Bianca Prodan - Automating historic document scans. A pipeline built with NCAP and the US National Archives extracts metadata from over a million historic aerial-photography plots, compressing a year of manual work into a week. Takeaway: automated pipelines can transform massive manual data-conversion tasks.
• Alistair Grant - Debugging stuck GPU jobs. Recent GPU scheduling improvements make multi-GPU jobs easier to run, and most failures come down to missing PVCs, image pull errors or queue quota. Takeaway: most stuck jobs are quick to diagnose with the right checks.

Full recording: https://edin.ac/4dBtOOX

In this session, three EIDF team members shared service updates:

• Magdalena Drafi-McHugh - New cost estimator and application process. A new cost estimator is live from the EIDF front page, usable without logging in, letting users estimate project costs by organisation type and services. Takeaway: it is now easier to move from a cost estimate to a project application.
• Amy Krause - Usage reporting and budget monitoring. A new portal reporting view shows service usage and costs by month, with budget alerts and CSV exports planned. Takeaway: better tools for tracking usage and managing budgets are on the way.
• Ruairidh MacLeod - Confidential Data Workspace. A new secure service for projects working with private, sensitive or commercially confidential data, accessed by remote desktop only. Takeaway: a secure environment for sensitive data, with GPU access planned for the future.

Full recording: https://edin.ac/43C1GpN

In this session, two speakers shared updates on EIDF services and research:

• Alistair Grant - EIDF service enhancements. New updates to the ECIR (EIDF Container Image Registry) support private repositories, security scans and software bills of materials for stronger compliance and security, alongside progress on InfiniBand in the GPU service. Takeaway: ECIR and InfiniBand are set to improve security and multi-node communication speeds.
• Bianca Prodan - Quantum machine learning for detecting deep fakes. A project using hybrid classical-quantum models to improve deep fake detection, adapting to limited data and compute resources with promising early results. Takeaway: quantum approaches could enhance the generalisation of deep fake detection systems.

Full recording: https://edin.ac/4do4uvW

In this session, two guest speakers shared their research, followed by community Q&A:

• Rosa Filgueira - Semantic enrichment of historical collections. Work on the National Library of Scotland's Gazetteers using machine learning and LLMs to segment text and clean OCR errors in scanned documents, plus a semantic web platform (Frances) to make the data accessible to non-computer scientists. Takeaway: EIDF resources can power practical applications that open up historical data.
• Maurice Jamieson - The 'Era' wellbeing app. A Design Informatics project for the University of Edinburgh Business School using an LLM to support health-issue exploration and personalised recommendations, developed on an A100 VM within EIDF. Takeaway: EIDF provides the compute to build and test real LLM-driven applications.

Discussion highlights: Robust discussion on using LLMs to enrich and access historical collections, plus questions on EIDF service details and documentation. Amy Krause shared future EIDF plans, including Windows VMs and enhanced storage options.

Full recording: https://edin.ac/46pdYUM 

In this session, the EIDF team presented the data publishing service and shared future plans:

• Mario Antonioletti - EIDF free data publishing service. A free service that helps researchers meet journal and funder compliance requirements, using DOIs to boost data accessibility. Takeaway: data publishing is open beyond the University of Edinburgh, extending to businesses, government and other academic institutions.
• Nick Johnson - Future EIDF enhancements. Upcoming upgrades include a Harbor-based container image registry for the GPU service and VMs, a mid-January OpenStack VM service upgrade, a Cerebras update (R2.6.0/R3.0.0) for AI and ML work, enhanced Ceph disk storage, and new encrypted S3 options. Takeaway: EIDF is strengthening both its storage security and container capabilities.

Closing: Attendees encouraged to sign up to the mailing list and provide feedback to help shape future EIDF sessions.

Full recording: https://edin.ac/4j7DSTM

• Emma King-Smith - AI in synthetic chemistry. Shrieking the integration of AI in synthetic chemistry, using machine learning to predict the synthesis and properties of molecules, with the potential to overcome traditional synthetic challenges. Takeaway: ML models could help revolutionise how new molecules are designed and made.

• Alejandro Bordallo - Autonomous vehicles in complex urban environments. A look at the challenges of integrating autonomous cars into the dense, unpredictable urban setting of Edinburgh, covering both practical hurdles and the required technological innovations. Takeaway: autonomous driving in real cities demands advances in perception, planning, and safety.

  Full recording: https://edin.ac/49aDSNx 

In this session, the EIDF team gave an overview of the service and answered community questions:

• Service overview. Welcomed new users, walked through EIDF's service offerings and how to get started with access.
• Q&A highlights. Confirmed EIDF is not formally listed with UKRI but that users can list it in applications; data publishing is free (subject to fair use, around 100GB per item); projects can be extended and can run for as little as a month; and charging is moving toward a pay-for-what-you-use model rather than use-it-or-lose-it.
• What's next. The team is exploring easier ways to access your support tickets across EPCC systems, and working toward an interface to make Docker and Kubernetes container jobs on the GPU service more approachable.

Full recording: https://edin.ac/46Z5Bi6