Reliable AI for Smart Systems Engineering

RAISSE-Methodology

Four conceptual priorities

  • Development of interpretable, robust analytic AI tools for system diagnostics/ prediction/ control with low-cost constraints for the machine learning (ML) model
  • Determination of smarter hardware configurations in engineering, sensoric or measurement devices by means of ML-analytics
  • Integration of engineering and/or domain expert knowledge into the design of AI systems to achieve lower complexity and, hence, sparser hardware constraints to run a ML-powered diagnostic/control engineering system  
  • Transfer of information or data from well-defined and controlled settings to new environments by means of reliable machine learning models
     

Challenges and requirements for the ML-models 

  • ML-models with low computationally complexity specifically adapted to the application/task
  • ML-models trainable with maybe low data amount
  • ML-models with mathematical guarantees regarding convergence and model/decision robustness
  • Interpretable ML-models for understandable model inspection and decision/prediction control = white-box-ML

RAISSE Methodology based publications

M. Riedel, F. Rossi, M. Kästner, T. Villmann:
Regularization in Relevance Learning Vector Quantization Using l1-Norms
European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning. April 2013

T. Pfaff, M. Kaden, T. Villmann:
Addressing Domain Adaptation through Bias-Mitigated Representation
Mittweida Workshop in Computational Intelligence (MiWoCI). August 2026

in press

T. Pfaff, D. Möbius, T. Villmann, M. Kaden:
Gait-Phase Classification using Environmental Dependent Measurements and Transfer Learning– From High- to Low-monitored Scenarios
IEEE MetroXRAIN. October 2026

in press

T. Pfaff, M. Kaden, T. Villmann:
RAISSE-Methodology for Interpretable Transfer Learning in Restricted Environments with Illustrative Examples
GFaI Future Tech Days. November 2026

in press