The BaCAI general meeting will take place immediately after the Research Day.

Agenda

All contributions were produced in both text and video formats and published across various social media platforms. The first installment, titled "What is AI, anyway?", was also translated into English and submitted to the AAAI’s Call for Educational AI Videos. It won the Educational AI Video Award and is now available on the official AAAI YouTube channel.

Computational Photography and Low-level Computer Vision are research areas with significant impact on both academia and industry.
This talk is about recent advances in computational photography and imaging. We will cover neural image signal processors (ISPs), advances in image restoration and enhancement, and image domain mapping problems. Our focus will be on challenges, findings, quality of the results, complexity and readiness of the solutions for real-world applications.

About Prof. Dr. Timofte

Radu Timofte is a Full Professor (W3) and the head of the Computer Vision Laboratory (Chair for Computer Science IV) at the University of Wurzburg. Previously, he held a postdoc position from 2013 until 2016 and a lecturer and research group leader position from 2016 until 2022, both at ETH Zurich, working with Prof. Luc Van Gool. He obtained a PhD degree in Electrical Engineering from the KU Leuven, Belgium, in 2013. He serves(d) as an associate editor for top journals: Elsevier CVIU, IEEE Trans. PAMI , Elsevier Neurocomputing and SIAM Journal on Imaging Sciences. He regularly serves(d) as an (Senior) Area Chair/SPC for top vision and machine learning venues: ICCV, ECCV, CVPR, IJCAI, NeurIPS, AAAI, ICLR, ICML. Radu Timofte is part of the organizing teams of ICIP'26 (Tampere) and ECCV'28 (Bucharest). He and his team received multiple awards, including a 2022 Alexander von Humboldt Professorship Award and a 2021 Romanian Academy Award. He is co-founder of Merantix, co-organizer of NTIRE, CLIC, AIM, MAI, AIS and PIRM events, member of IEEE, CVF, and an ELLIS Fellow. His current research interests include augmented perception, mobile AI, multimodal learning, image/video compression, restoration, manipulation, and enhancement. 

As of January 2026, he has published over 250 papers, receiving over 74,500 citations with an h-index of 112. 

Das Spektrum menschlicher Annotationen

Reflexionen aus dem letzen Jahrzehnt im Natural Language Processing

Menschliche Annotation ist inhärent variabel und spiegelt eher Ambiguität, Subjektivität und vielfältige Interpretationen wider als bloßes Rauschen. Dieser Vortrag untersucht die Variation menschlicher Labels als Spektrum und beleuchtet ihre Implikationen für das Design von Datensätzen, das Modelltraining und die Evaluation im NLP sowie für das Verständnis, wie LLMs mit menschlichen Urteilen übereinstimmen. Ich diskutiere neuere Ansätze, die diese Variation modellieren und nutzbar machen. Diese gehen über einzelne „Gold“-Labels hinaus und führen zu Repräsentationen, die den Reichtum menschlicher linguistischer Urteile besser erfassen.

Human annotation is inherently variable, reflecting ambiguity, subjectivity, and diverse interpretations rather than mere noise. This talk explores human label variation as a spectrum - highlighting its implications for dataset design, model training, and evaluation in NLP and for understanding how LLMs align with human judgments. I discuss recent approaches to model and leverage this variation, moving beyond single “gold” labels toward representations that better capture the richness of human linguistic judgment.

Über Prof. Dr. Barbara Plank:

Barbara Plank ist ordentliche Professorin und Lehrstuhlinhaberin für KI und Computerlinguistik an der LMU München, Leiterin des Munich AI and NLP (MaiNLP) Lab und Co-Direktorin des Center for Information and Language Processing (CIS). Sie ist außerdem Gastprofessorin an der IT University of Copenhagen.

More on this at uni.blog

Translated with DeepL.com (free version)

A Unified Theory of Perception, Memory and Semantic Decoding

In our proposed model, the cognitive brain state reflects the dynamic interplay between sensory inputs and higher-order cognitive processes. It is continuously shaped by subsymbolic (bottom-up) signals—originating from sensory modalities, internal processing modules, and reward systems—and guided by symbolic (top-down) influences, which encode elements such as recognition, goals, decisions, temporal markers, predicates, and actions. 
 

Speaker: Prof. Dr. Beishui Liao (Zhejiang University, Hangzhou, China)

Title: LLM-ASPIC+: A Neuro-Symbolic Framework for Defeasible Reasoning

Abstract: Large language models (LLMs) have demonstrated remarkable capabilities in complex reasoning, achieving human-like performance in various nature language processing tasks. However, despite their impressive performance in many domains, LLMs often struggle with reasoning under conditions of inconsistency or contradiction. This capability gap raises significant concerns for real-world applications where reliable decision-making depends on reconciling conflicting evidence, such as legal analysis, medical diagnosis, and commonsense reasoning. In this paper, we focus on defeasible reasoning in natural language, a task that require LLMs to handle contradictory information. To improve the defeasible reasoning capability of LLMs, we propose LLM-ASPIC+, a framework combining neural language understanding with formal argumentation. Our framework harnesses LLMs’ capacity for grounding and contextual reasoning while integrating formal argumentation frameworks to establish systematic conflict resolution mechanisms lacking in LLMs. The LLM-ASPIC+ achieves state-of-the-art performance on multi- step defeasible reasoning benchmarks, attaining 87.1% accuracy on BoardGameQA-2 and 82.6% accuracy on BoardGameQA-3. These results demonstrate that combining neural language processing with formal argumentation offers an effective approach to defeasible reasoning in natural language.

Join in person: WE/05.017 
or via Zoom: https://uni-bamberg.zoom-x.de/j/63070351891 (Password: AISE-2025)