News from MLA2S and Make it (net)work!

• Claudius Krause (MBI Vienna (ÖAW))
• Nicki Holighaus (Acoustics Research Institute, Austrian Academy of Sciences)
• Claus Trost (Erich Schmid Institute of Materials Science of theAustrian Academy of Sciences)
• Kati Heinrich (IGF | ÖAW)
• Jan Odstrčilík (IMAFO)
• Elisabeth Eder (Austrian Academy of Sciences)

Including a short round table news update.

From Data-Driven Alloy Design to Characterisation: Machine Learning in Materials Science

• Swati Mahato (Erich Schmid Institute of Materials Science)

The development of advanced materials has traditionally followed a fundamental processing- structure-properties-performance paradigm. Understanding and optimising these links is therefore central to designing materials for specific applications. However, the growing complexity of modern multicomponent alloys has made this task increasingly challenging. The vast compositional space, coupled with the intricate relationships among chemistry, processing, microstructure, and deformation mechanisms, makes conventional trial-and-error approaches inefficient and often requires a prohibitively large number of experiments. Machine learning offers an opportunity to accelerate the materials development cycle by learning these complex relationships directly from data. The talk will present how data-driven approaches can be used to explore and optimise alloy chemistry, predict material properties, and tailor composition to activate different deformation mechanisms, thereby enhancing performance. Rather than replacing experiments, machine learning can narrow the enormous design space and guide experiments towards the most promising alloy compositions, helping to overcome one of the major bottlenecks in conventional alloy development. The same data-driven approach can be further extended from design to characterisation. Advanced electron microscopy techniques such as TEM and 4D- STEM generate rich, spatially resolved datasets that contain information about microstructural heterogeneity, local deformation, and strain distributions. While these datasets provide unprecedented insight into materials, extracting and quantifying such information can be challenging and time-consuming. The talk will address how machine learning can be integrated with microscopy to accelerate image analysis, quantify microstructural features and extract information that would otherwise require extensive manual analysis. The talk will also address the emerging potential of explainable machine learning to move beyond prediction towards materials understanding. By identifying the chemical, structural and microstructural features that control model predictions, explainable machine learning provides a pathway to uncover physically meaningful relationships rather than treating the model as a black box. Overall, the talk will demonstrate how machine learning can be embedded across the materials science paradigm, from alloy design and property prediction to microscopy and mechanistic understanding, creating a data-driven pathway towards faster materials discovery while retaining the fundamental goal of materials science.

Discussion

An AI Framework for Semantic Processing and Natural Language Reasoning in Reactive Heritage Digital Twins: The ARTEMIS Approach to Integrating and Exploring Cultural Heritage Data

• Aida Himmiche (Fondazione PIN, University of Pisa)
• Mariana Somenzi (ARIADNE RI, University of Pisa)

Cultural heritage research increasingly relies on large and heterogeneous collections of resources, ranging from archaeological catalogues and excavation reports to scientific analyses, restoration records, images, sensor observations and other forms of documentation. Artificial Intelligence and Machine Learning offer new possibilities for processing, connecting and accessing this information. At the same time, Digital Twins are emerging as a promising framework for bringing together the tangible and intangible dimensions of cultural heritage: not only the physical characteristics and changing conditions of heritage objects, monuments and sites, but also the historical knowledge, cultural practices, interpretations and narratives associated with them. Developing such systems, however, requires approaches capable of integrating diverse data while remaining sensitive to domain knowledge, provenance, interpretability and the essential role of human expertise. This lecture presents the AI framework developed within ARTEMIS, a project funded by the European Union dedicated to the creation of Reactive Heritage Digital Twins. After introducing Heritage Digital Twins, the project, and discussing how Machine Learning and AI can support the documentation, study, conservation and management of cultural heritage, the lecture will describe the semantic data infrastructure underpinning the ARTEMIS ecosystem. Particular attention will be given to the role of ontologies and Knowledge Graphs in integrating diverse data while explicitly representing their meaning, relationships, and provenance. The central part of the lecture will present two complementary AI contributions. The first is an AI semantic annotation pipeline designed to transform heterogeneous and partly unstructured cultural heritage documentation into structured, ontology- aligned knowledge. The pipeline combines Large Language Models, semantic retrieval and ontology-based entity linking to identify relevant concepts and relations, associate them with formal cultural heritage ontologies and generate RDF statements for Knowledge Graph population. A short demonstration will illustrate this process using related documentation. The second contribution is the Artemisia reasoning engine, which operates in the opposite but complementary direction: rather than using AI to populate the Knowledge Graph, it uses the graph and its ontological structure as a foundation for AI-based querying and reasoning. Through the Artemisia chatbot, users can formulate questions in natural language and explore interconnected cultural heritage data without needing to understand the underlying ontologies, graph structure or SPARQL query language. The Knowledge Graph provides the system with explicit semantic context and traceable evidence, helping it identify relevant resources, traverse meaningful relationships and generate grounded responses. A live showcase will demonstrate how this approach facilitates access to distributed and semantically diverse information. More broadly, the lecture will also address the importance of selecting and configuring the “right” AI for specialised domains. The ARTEMIS approach aims to minimise hallucinations, protect data sovereignty and make AI outputs more transparent, traceable, and verifiable. This approach also responds to the principles of accountability, transparency and human oversight promoted by the European AI Act.

Discussion

Networking with refreshments / further discussions