From 25 to 28 March 2025, the RILEM Spring Convention 2025 took place in Mendrisio, Switzerland, gathering researchers and industry stakeholders to exchange knowledge on resilient and low-carbon construction materials.
As part of the Reincarnate project, Ghezal Ahmad Jan Zia from our consortium partner Bundesanstalt für Materialforschung und -prüfung presented the study “Meta-Learning for Adaptive Mix Design of Alkali-Activated Concrete.” The work introduces Meta-Design, an AI-powered tool developed to support the discovery and optimization of alkali-activated concrete (AAC) mixtures, a promising alternative to traditional cement-based materials.
Meta-Design applies meta-learning algorithms, specifically Model-Agnostic Meta-Learning (MAML) and Reptile, to enable the reuse of limited experimental data across different material systems. This allows for real-time predictions of compressive strength, reducing the need for repeated lab testing and accelerating the formulation of sustainable AAC mixes. The study explores how AI can be used to shift the design process from trial-and-error towards more data-efficient and environmentally responsible methods.
The tool adopts an iterative, model-guided optimization strategy inspired by SLAMD (Sequential Learning App for Materials Discovery) ), but introduces a distinct decision logic tailored to meta-learning workflows. It is implemented as a user-friendly dashboard application. Key features of the MetaDesign platform include:
- Dynamic model switching between MAML and Reptile
- Backend-integrated hyperparameter optimization
- Acquisition strategies (e.g., EI, UCB, PI) applied automatically during model-driven optimization.
- Interactive visualizations using Plotly for analyzing model predictions and performance.
- Robust handling of missing data with support for iterative dataset updates
The application is available as an open-source project and comes with full documentation, example datasets, and setup instructions. A live demo version is also available online.
📂 GitHub repository
🧪 Live demo
This contribution reflects Reincarnate’s commitment to the integration of AI and data-driven methodologies into the material lifecycle, with the aim of enabling more sustainable and circular building practices.