KI 2025: 48TH GERMAN CONFERENCE ON ARTIFICIAL INTELLIGENCE
PROGRAM

Days: Tuesday, September 16th Wednesday, September 17th Thursday, September 18th Friday, September 19th

Tuesday, September 16th

View this program: with abstractssession overviewtalk overview

09:00-10:30 Session 1A: DC: Keynote 1 & Applied XAI

The doctoral consortium aims to offer early stage researchers from any AI subject area a unique opportunity to present their planned research, and to connect to fellow PhD students as well as senior researchers in AI.

09:00
Welcome
09:05
Why is Networking Important in Research? A Short Introduction to the AI Grid
09:30
AI-Driven Chest Radiography Report Generation: Integrating LLMs, CLIP, Tree-of-Thoughts, Multimodal Retrieval-Augmented Generation, Classification and Direct Preference Optimization (abstract)
09:50
Analyzing Deep Generative Models for Steel Microstructures (abstract)
10:10
Economically-Driven AI Process for Quality Assurance: Analysis in Optics Manufacturing (abstract)
09:00-10:30 Session 1B: CHAI & FCR Workshop

Joint workshop on Humanities-Centered AI (CHAI 2025) and Formal and Cognitive Reasoning (FCR 2025); Individual workshop pages:

09:00-10:30 Session 1D: AI in Production Workshop
10:30-11:00Coffee Break
11:00-12:30 Session 2A: DC: XAI & DNN Efficiency

The doctoral consortium aims to offer early stage researchers from any AI subject area a unique opportunity to present their planned research, and to connect to fellow PhD students as well as senior researchers in AI.

11:00
Explainable Artificial Intelligence for Multivariate Sensor Data: Towards Transparency and Correctness in Model Explanations (abstract)
11:20
Learning Interpretable Disentangled Concepts for Neurosymbolic Integration (abstract)
11:40
Promoting Flatness of Representation Manifolds to Improve Deep Network Training (abstract)
12:00
Efficient Graph-Based Neural Architectures for Multimodal Learning (abstract)
11:00-12:30 Session 2B: CHAI & FCR Workshop

Joint workshop on Humanities-Centered AI (CHAI 2025) and Formal and Cognitive Reasoning (FCR 2025); Individual workshop pages:

11:00-12:30 Session 2D: AI in Production Workshop
12:30-14:00Lunch Break
14:00-15:30 Session 3A: DC: Keynote 2 & DNN Efficiency

The doctoral consortium aims to offer early stage researchers from any AI subject area a unique opportunity to present their planned research, and to connect to fellow PhD students as well as senior researchers in AI.

14:00
How (not) to PhD
14:40
Model Efficiency Techniques in Multimodal Learning (abstract)
15:00
Data-Efficient Multimodal Training Strategies (abstract)
14:00-15:30 Session 3B: CHAI & FCR Workshop

Joint workshop on Humanities-Centered AI (CHAI 2025) and Formal and Cognitive Reasoning (FCR 2025); Individual workshop pages:

14:00-15:30 Session 3D: AI in Production Workshop
15:30-16:30Coffee Break
16:30-18:00 Session 4A: DC: Formal Methods & Closing

The doctoral consortium aims to offer early stage researchers from any AI subject area a unique opportunity to present their planned research, and to connect to fellow PhD students as well as senior researchers in AI.

16:30
Privacy Risk Assessment in Federated Learning: Extracting and Protecting Sensitive Information from Vision Language Models in Manufacturing Applications (abstract)
16:50
Verification of Neural Networks (abstract)
17:10
Research Proposal: Runtime Verification on Spatial Objects (abstract)
17:30
Closing, Open Discussion Round
16:30-18:00 Session 4B: CHAI & FCR Workshop

Joint workshop on Humanities-Centered AI (CHAI 2025) and Formal and Cognitive Reasoning (FCR 2025); Individual workshop pages:

16:30-18:00 Session 4D: AI in Production Workshop
16:30-18:00 Session 4E: FAIR4ML Tutorial
  • Tutorial: Towards FAIR4ML Rohitha Ravinder, Nelson Quiñones, Dietrich Rebholz-Schuhmann, Leyla Jael Castro
18:00-23:00 Welcome Reception

Together with the co-located conferences, the day ends in the welcome reception, starting at 6pm in the rooms of the Hasso-Plattner-Institute. The evening includes the award ceremony for the GI junior fellows and the Balzert prize.

More details in German: https://informatik2025.gi.de/abendveranstaltungen.html 

Location: HPI House L
Wednesday, September 17th

View this program: with abstractssession overviewtalk overview

08:00-09:00 Women in AI Breakfast

Before the conference day starts, we invite all who identify as women to have breakfast with us. The goal is to strengthen female networks, and share in small discussion rounds personal experiences, strategies, and best practices regarding challenges faced by women in academia. There will be coffee, tea, and an assortments of breakfast items.

09:30-10:30 Session 5: Informatik Festival 2025 Opening & Panel

This session opens the Informatik Festival and includes the first panel on openness and computer science (in German).

10:30-11:00Coffee Break
11:00-12:30 Session 6: Informatik Festival 2025 Talk & Panel

This session includes a talk and a panel on computer science and education.

12:30-14:00Lunch Break
14:00-14:15 Session 7: KI 2025 Opening

This session officially opens the 48th German Conference on Artificial Intelligence, organized in cooperation with the Fachbereich Künstliche Intelligenz der Gesellschaft für Informatik (GI-SIG AI). KI2025 is organized in combination with INFORMATIK 2025.

In this session, we present an overview of the upcoming days and reveal the nominees for the best paper award.

14:00
KI 2025 - Opening
PRESENTER: Tanya Braun
14:15-15:15 Session 8: Invited Talk 1

This invited talk opens the KI 2025.

14:15
Tackling Challenges in Critical Infrastructure through Machine Learning
15:15-15:30 Session 9: DAAD - German Academic Exchange Service

With its program IFI - Internationale Forschungsaufenthalte für Informatiker*innen / International Research Stays for Computer Scientists -, the DAAD (German Academic Exchange Service) promotes research-oriented stays abroad for doctoral students and young researchers with a doctorate in the field of computer science, with a special focus on artificial intelligence.

15:15
International Research Stays for Computer Scientists
15:30-16:30Coffee Break
16:30-18:00 Session 10: Knowledge Representation

This session offers varying topics that have a basis in knowledge representation, ranging from constraint programming and argumentation to answer set programming, game theory, and the combination of logic and probabilities

16:30
A Hybrid Constraint-Based, Greedy, and Local Search Approach for the Transshipment Problem (abstract)
PRESENTER: Sven Löffler
16:50
Numbers Don't Lie: Hybrid Extraction and Validation of Quantitative Statements in Arguments with Semi-Structured Information (abstract)
PRESENTER: Mirko Lenz
17:10
Towards Observing the Effect of Abstraction on Understandability of Explanations in Answer Set Programming (abstract)
17:20
Balanced Reciprocity for Data Sharing - Axiomatization and Mechanism Design (abstract)
PRESENTER: Björn Filter
17:40
Probabilities of the Third Type: Statistical Relational Learning and Reasoning with Relative Frequencies (Extended Abstract) (abstract)
19:00-23:00 Community Dinner

The first day of the conference ends in the community dinner, starting at 6pm with a small reception, followed by food and drinks in a relaxing atmosphere. The dinner includes the award ceremony for the GI fellows and the Konrad Zuse medal.

More details in German here: https://informatik2025.gi.de/abendveranstaltungen.html

Thursday, September 18th

View this program: with abstractssession overviewtalk overview

09:00-10:30 Session 11: Vision, Explanations, and Privacy&Security

This session starts with vision and anonymisation, moving on to interpreting vision transformers and explanations for object detection, and then morphing into an explanation session with talks on explainable move selection for chess and a new attack technique against explanations, circling back to the privacy and security topic of the first talk.

09:00
Towards Systematic Evaluation of Computer Vision Models under Data Anonymization (abstract)
PRESENTER: Sarah Weiß
09:20
Visualizing and Interpreting Neural Network Focus Regions: A Comparative Study of Vision Transformers on Synthetic and Real Data (abstract)
PRESENTER: Waldemar Haag
09:30
ODExAI: A Comprehensive Object Detection Explainable AI Evaluation (abstract)
09:50
Caïssa AI: A Neuro-Symbolic Chess Agent for Explainable Move Suggestion and Grounded Commentary (abstract)
PRESENTER: Nourhan Ehab
10:10
Makrut Attacks Against Black-Box Explanations (abstract)
PRESENTER: Achyut Hegde
10:30-11:00Coffee Break
11:00-12:30 Session 12: Explanations and Robustness

This session focusses on robustness (and resilience) in various forms, for feature selection, object detection, image classification, and vision-language models.

11:00
Unsupervised Selection of Features by their Resilience to the Curse of Dimensionality (abstract)
PRESENTER: Tom Hanika
11:20
XAIRob --- An Explainable-AI-Based Relative Robustness Measure for Object Detection (abstract)
11:30
Re-Evaluating the Robustness and Interpretability of the Contrastive Explanation Method for Image Classification (abstract)
PRESENTER: Luisa Schneider
11:40
On the Domain Robustness of Contrastive Vision-Language Models (abstract)
12:00
Toward Short and Robust Contrastive Explanations for Image Classification by Leveraging Instance Similarity and Concept Relevance (abstract)
12:20
Learn, Optimize, Explain: A Neuro-Symbolic Advisor for Personal Finance (abstract)
PRESENTER: Nourhan Ehab
12:30-14:00Lunch Break
14:00-14:45 Session 13: Invited Talk 2 (together with Informatik Festival 2025)

This session is part of the Informatik Festival main program and is thematically linked with the following panel.

14:00
Magic wands and digital zombies: some promises and risks of AI for digital legacy
14:45-15:30 Session 14: Panel Discussion (together with Informatik Festival 2025)

What Norms Do We Want for AI?

AI is increasingly permeating all areas of life. This panel discussion focusses on AI and ethics, touching on changes and risks, rules and regulations, as well as the human in all of it.

15:30-16:30Coffee Break
16:30-17:00 Session 15: Special Session: On the Origins of AI in Germany

To celebrate the 50th anniversary of the first meeting that started the KI conference as it is known today, we present a special session on the origins of AI in Germany with a presentation by Rudolf Seising, Deutsches Museum Munich.

16:30
The origins of AI research in the Federal Republic of Germany (abstract)
17:00-18:30 Session 16: GI-FBKI General Assembly

The Artificial Intelligence Chapter (FBKI, in German: Fachbereich Künstliche Intelligenz) of the Gesellschaft für Informatik invites its members to its annual general assembly (in German: Mitgliederversammlung).

18:00-23:59 Junge GI Festival Night

The festival night offers retro-gaming, live music, a science slam, and many opportunities to talk to new people.

More details in German here: https://informatik2025.gi.de/abendveranstaltungen.html

Location: Siggi
Friday, September 19th

View this program: with abstractssession overviewtalk overview

09:00-10:00 Session 17: Invited Talk 3

The third invited talk opens the last day of the conference.

09:00
Theory of mind - Between natural and artificial intelligence (abstract)
10:00-10:30 Session 18: Large Language Models for Easy Language

This sessions presents papers using large language models to generate easy German.

10:00
Accessible Language Simplification: Large Language Models for Generating Easy German (abstract)
PRESENTER: Raeesa Yousaf
10:20
LLMs for Easy Language Translation: A Case Study on German Public Authorities Web Pages (abstract)
10:30-11:00Coffee Break
11:00-12:30 Session 19: Machine Learning and its Applications

This session presents two papers on meta-features and the combination of synthetic and real-world data as well as papers on the application domains of passive acoustic monitoring, flood inundation, steel microstructures, and personal finance.

11:00
Enhancing Semi-Supervised Learning with a Meta-Feature Based Safeguard System (abstract)
11:20
Development of Hybrid Artificial Intelligence Training on Real and Synthetic Data (abstract)
PRESENTER: Paul Wachter
11:40
Intermediate-Task Transfer Learning for Bioacoustic Data (abstract)
PRESENTER: Hannes Kath
12:00
Deep learning emulators for large-scale, high-resolution urban pluvial flood prediction (abstract)
12:10
Comparing the visual quality of deep generative models for steel microstructures (abstract)
PRESENTER: Marcel Wentzien
12:30-14:00Lunch Break
14:00-15:30 Session 20: Large Language Models & Closing

This session focusses on large language models, specifically set encodings to improve LLMs and using LLMs for learning linear functions, code completion, and in a human-AI framework for collaborative systematic literative review.

14:00
Positional Overload: Positional Debiasing and Context Window Extension for Large Language Models using Set Encoding (abstract)
PRESENTER: Lukas Kinder
14:20
Re-examining learning linear functions in context (abstract)
PRESENTER: Omar Naim
14:40
Exploiting Contexts of LLM-based Code-Completion (abstract)
15:00
Augmenting Systematic Literature Reviews in Information Systems: A Human-AI Collaborative Framework (abstract)
15:20
KI 2025 - Closing
PRESENTER: Tanya Braun
15:30-16:30Farewell Coffee Break