ZIH-Info No. 182, September 2024

Shibboleth login: new look and feel

Since August 26, 2024, the login window familiar from many TU Dresden web applications has a new visual appearance: The layout corresponds to the corporate design of TU Dresden. In addition, accessibility requirements have been implemented. The login window was changed as part of the software update to Shibboleth v 5.1.2. (Contact: ).

Baramundi makes the PC pools fit

Following the frequent occurrence of technical problems in the four PC pools in the Andreas Pfitzmann Building of TUD due to outdated software for providing the networked working environment, there is now a remedy: Now Baramundi has been procured for the new client management. A new Windows 11 EDU image was created for the TUD and all software packages for the pools and made available on the approx. 120 PCs. The pools are now stable, secure and up-to-date, the support effort has decreased and satisfaction has increased. Some fine-tuning is still required before the changeover can be implemented in the eight other PC pools managed by the ZIH. (Contact: )

Energy efficiency: expanding the use of waste heat

The implementation for the sustainable use of waste heat from the ZIH supercomputing systems, which was agreed last year with the regional energy supplier SachsenEnergie AG, is on the finishing straight. On August 13, 2024, the three large heat pumps were delivered and placed in the new, purpose-built technical building next to the data center (LZR). This brings the goal of processing the left-over waste heat from the data center directly and feeding it into the city's district heating network one step closer. The 24,000 megawatt hours of green heat that will be generated here each year from the end of the year will be able to supply an average of more than 3,500 Dresden households. At the same time, around 2,700 tons of CO² can be avoided, which would be produced each year in the generation of district heating without this project. Energy efficiency was a key planning focus at the LZR from the very beginning. For example, the waste heat from the computers has been reused to heat surrounding campus buildings since commissioning in 2015, see: https: //tu-dresden.de/zih/die-einrichtung/news/einweihung-lrz-und-hrsk-ii.(Contact: Dr. Daniel Hackenberg)

Computer simulations of tumor evolution

A research team led by scientists from the ZIH Department of Innovative Computing Methods has published a study on evolutionary dynamics in tumors, i.e. cell changes. The study, which was conducted in collaboration with researchers from Norway, the United Kingdom and the United Arab Emirates, sheds light on the complex relationship between irreversible genetic changes and reversible adaptations of tumor cells. The different developments of tumor cells lead to a high cell diversity, which makes treatment more difficult if, for example, some cells are resistant and survive treatment. The international research team has now developed a novel model approach to describe the individual genetic and non-genetic properties of cancer cells. The results of the study could have a significant impact on future therapeutic approaches. The study was published in the current issue of the journal PLOS Computational Biology. https://journals.plos.org/ploscompbiol/article?id=10.1371/journal.pcbi.1012003 (Contact: Dr. Simon Syga)

Development of multilingual language models

Language processing has become considerably more important due to the spread of open source language models – so-called Large Language Models (LLMs). The project OpenGPT-X, launched in 2022, aims at the multilingual expansion of this technology. As most of the available benchmarks for evaluating language models exist primarily for the English language, the OpenGPT-X consortium aims to improve language diversity – for a fairer and more effective language technology. To this end, the OpenGPT-X team carried out extensive multilingual training runs and tested the developed AI models on tasks such as logical reasoning, commonsense understanding, multi-task learning, truthfulness and translation. The team has now published a multilingual "European LLM Leaderboard", a ranking that compares several publicly available state-of-the-art language models, each comprising around 7 billion parameters. The plan is to use it to automate the evaluation of models from the central AI platform Hugging Face in order to ensure the comparability and reproducibility of the results. The ZIH provides the necessary infrastructure for the extensive evaluations. The project OpenGPT-X is funded by the German Federal Ministry for Economic Affairs and Climate Action (BMWK). Further information: https: //tu-dresden.de/zih/die-einrichtung/news/european-llm-leaderboard-of-opengptx (Contact: Klaudia-Doris Thellmann)

Interaction of HPC, AI and research data

Data analysis and artificial intelligence (AI) place high demands on both the available computing resources and the corresponding software development. The NHR-funded project "Interaction between HPC, AI, and Research Data" (HAI) therefore focuses not only on computing efficiency but also on strategies and tools for efficient and collaborative code development. Based on use cases on the supercomputing systems of the participating NHR partners, HAI focuses on three aspects: efficient data processing pipelines with LLM-automated data engineering, efficient and resource-saving training of large AI models through monitoring and optimization, and collaborative code development and execution. To ensure integrity across the entire AI project lifecycle, FAIR data management practices are taken into account. The project is carried out together with the NHR partner institutions NHR4CES@RWTH (Aachen), NHR4CES@TUDa (Darmstadt), NHR@FAU (Erlangen), NHR-Nord@Göttingen, NHR@KIT (Karlsruhe) and the NHR@SW network. ScaDS.AI will participate as associated partner. (Contact: Dr. Matthias Lieber)

Plain text: websites in plain language

It is estimated that between 10 and 17 million people in Germany have reading difficulties, including functional illiterates and people with reading and writing disorders and mental impairments. In the course of internationalization, the number of non-native speakers has also increased in recent years. All of these people could benefit from texts written in easy-to-understand language. The current "Klartext" project, led by Professor Michael Färber, will therefore use large language models to translate website content from ScaDS.AI Dresden/Leipzig and TU Dresden into plain or simple language. The aim is to reduce language barriers and promote the inclusion and participation of all citizens in scientific and cultural discourse. The "Klartext" project is funded as part of the SMWK 2024 special funds for inclusion for the period from 01.09. to 31.12.2024. Further information www.scads.ai/en/klartext. (Contact: Prof. Dr.-Ing. Michael Färber)

Podcast: What is artificial intelligence?

The Living Lab of ScaDS.AI Dresden/Leipzig welcomed the "Experts in their own cause" in July. The representatives of this project for people who rely on easy-to-understand language wanted to find out more about AI. The conversation can be listened to in episode 35 "Artificial intelligence" of the podcast "reden; nur mit uns!". In addition to the fundamental opportunities and risks of AI, various areas of application are discussed, including natural language processing, art and music, politics, medicine as well as traffic and transportation. The experts on their own cause are self-advocates and also represent the interests of other people with disabilities. With their commitment, they want to help shape society and politics. Link to the podcast: https://nipb-sachsen.de/podcast-der-expertinnen-in-eigener-sache/. (Contact: Dr. Siavash Ghiasvand)

The events

Open Q&A session for users of the NHR@TUD computing cluster:
02.09., 16.09, 30.09.2024: 13:30-14:30 each time (online)
Softphone consultation hour:
02.09., 09.09., 16.09, 23.09., 30.09.2024: 10:00-11:00 a.m. each (online)
12.09.2024, 11:00-12:00: Living Lab Lecture: Mapping Glacier Sliding with Machine Learning (ScaDS.AI Dresden/Leipzig)
18.09., 25.09. and 2.10.2024, 09:00-17:00: From Zero to Multi-Node GPU Programming (3-Day-Workshop/NHR-Tutorial)
26.09.2024, 09:20-10:50: OPAL basic course (online)

http://www.tu-dresden.de/zih/veranstaltungen