Below is a selection of talks, podcasts, and news articles from or about us (some in German) that may be of interest to you. Click here for news about our latest activities. If you have any questions or comments, we look forward to hearing from you!
Sexism in online discussions can discourage women from participating, especially given the overwhelming volume of comments that moderators have to manage. Our method, which emulates the decisions of different moderators, was awarded 2nd place at the GermEval 2024 shared task 2, held at KONVENS in Vienna. For a detailed look at our approach, please view the presentation by our team member Patrick Gustav Blaneck below.
Our team's research on tackling data scarcity in sleep medical settings was selected for an oral presentation at the Time-Series for Health Workshop at ICRL 2024 in Vienna. Niklas presented his findings on pretraining sleep staging models without patient data—opening new doors for deep learning in settings where time series data is limited.
Anfang März 2023 fand an der FH Aachen eine hochschulweite Informationsveranstaltung zu ChatGPT und seinen möglichen Auswirkungen auf die Hochschullehre statt. Der Vortrag erläutert, wie große Sprachmodelle (large language models) wie ChatGPT trainiert werden, welche Auswirkungen solche Systeme auf die Hochschullehre haben und wie zukünftige Entwicklungen aussehen könnten.
Dieser Vortrag wurde im Rahmen der didaktischen Tafelrunde der Hochschule gehalten und zuerst auf dem ZHQ-Blog der FH Aachen veröffentlicht.
In diesem Interview in einem Podcast unserer Hochschule diskutieren wir die möglichen Auswirkungen von Systemen wie ChatGPT auf die Hochschullehre.
Dieses Podcast-Interview wurde als Teil des ZHQ-Podcasts der FH Aachen geführt und wurde zuerst ebenda veröffentlicht.
Dieser Artikel, der Anfang 2023 in der Zeitschrift Die Neue Hochschule veröffentlicht wurde, stellt die Entstehung großer Sprachmodelle wie ChatGPT vor, diskutiert ihre Risiken und Chancen und erörtert mögliche Auswirkungen auf sich verändernde Kompetenzen, Lernziele und Prüfungsformen an Hochschulen.
Stephan was invited to speak at the NetDat 2022 conference hosted by the Max Planck Institute for the Physics of Complex Systems in Dresden, Germany. His talk explained the rapid progress in Deep Learning to a non-machine learning audience, and included discussions of observations that preceded the introduction of ChatGPT, which was released two months later. You can watch his one-hour talk in the video below.
This talk was recorded by the Max-Planck Institute for the Physics of Complex Systems in Dresden, Germany.
The Aachener Zeitung reported that our research group won first place in the international GermEval 2021 competition for automatic hate speech detection. Our method outperformed all other approaches submitted by international university teams. You can read the article here (German).
🏆 Our team won 2nd place in the international competition GermEval 2024 (Subtask 2) on detecting sexism in online news media! Congratulations to all team members! 👏 Helping moderators identify harmful user comments is becoming increasingly important due to the growing number of comments on many platforms. Find out more about our work here: 📄 paper, 👨🏫 presentation, 💡LinkedIn post.
🙌 At the ICLR conference in Vienna, our PhD student Niklas presented his research on overcoming data scarcity in sleep medical settings. Watch his presentation on Pretraining sleep staging models without patient data—a step forward for deep learning applications in limited data environments. ✨
🎉 Our team's research on tackling data scarcity in sleep medical settings has been selected for both an oral presentation and a poster at ICLR 2024. Niklas will present his findings on Pretraining sleep staging models without patient data—opening new doors for deep learning in settings where time series data is limited. 🚀
🤓 Check out our newly published paper on finetuning large language models (LLMs) to decode political conversations in German parliamentary debates. 'Speaker Attribution in German Parliamentary Debates with QLoRA-adapted Large Language Models' offers a novel perspective on automated speaker attribution, facilitating the analysis of political interactions.
🎓 We celebrate Tobias Bornheim who successfully defended his Master thesis on Prompt Engineering vs QLoRA Fine-Tuning of Large Language Models! We are grateful for Tobias' exceptional dedication, which has been instrumental in achieving our three top placings in international NLP competitions - a testament to his invaluable contribution to our team. 🙏
🏆 Our team won the 2nd place in the international competition GermEval 2023 on speaker attribution in German parliamentary debates! Congratulation to all team members! 👏 Our approach is based on LoRA-training of large language models (Llama 2 70B). Find out more.