MCML Industry Pitchtalks With SAP

15:30        

Open Doors

16:00

Quick Welcome by SAP and MCML

16:05 - 16:20

Ivica Obadic

Explainable AI for Earth Observation

Earth Observation (EO) is a valuable data source in tackling pressing challenges such as climate extreme events, agriculture monitoring or urban planning. In recent years, the deep learning approaches became a dominant approach for knowledge extraction in EO as they are capable of automated feature extraction that yields high predictive performance. However, their black-box nature hinders intuitive understanding of the model decisions, which represents an obstacle for the wider acceptance of these models in critical applications. In this talk, I will present diverse explainable approaches that provide interpretation to deep learning approaches models for monitoring socioeconomic outcomes, scene classification, and agriculture monitoring.

16:20 - 16:35

Maximilian Mutschalik

shapiq: Shapley Interactions for Machine Learning

Recently, we presented `shapiq`, an open-source library for explainable AI with Shapley interactions. Shapley interactions extend the well-known concept of model explanations via Shapley values and allows finding interactions of any-order. Next to Shapley-based explanations, `shapiq` also offers methods to compute a wide collection of different game-theoretic concepts, which can be of use in different settings apart from explainable AI, such as data valuation or feature selection. The pitchtalk will present the library and the research work it is built upon.

16:35 - 16:50

Yunpu Ma

Agentic Neural Network

Modern challenges increasingly demand collaborative intelligence, yet most multi-agent systems remain statically designed and manually coordinated. In this talk, we introduce the Agentic Neural Network, a conceptual framework that treats agent collaboration like layered computation in a neural network. Each agent functions as a dynamic node, forming structured teams that specialize in different subtasks and evolve through iterative feedback. This approach enables scalable, adaptive agentic systems that self-organize and improve over time—without relying on fixed configurations.

16:50 - 17:05

Talk by SAP

17:05 - 17:20

Talk by SAP

17:20

Time for Networking