AI Strategy of SCCH

Leitmotif of SCCH Agenda: “Smart Transition to AI-based CPS”

SCCH addresses the technological transition from contemporary industrial (software) systems, characterized by

  • designed to perform a specific automation task on a specific platform;
  • restricted flexibility and machine intelligence due to specific data-driven models with a limited view of data and tasks;
  • sensitive to functional design changes (e.g. GUI layout), code changes and system implementation dependencies;
  • sensitive to changes w.r.t. requirements and/or operation conditions;
  • inadequate for small lot-size and personalized operation,

to AI-based cyber-physical systems (CPS), characterized by

  • extended functionality through merge and use of distributed data, cloud computing, transfer learning and service-oriented architectures;
  • designed for adaptability and extensibility to a broader class of tasks;
  • increased flexibility and machine intelligence due to transferable machine learning and leveraging data from a broader scope of tasks;
  • faster and more cost-effective adjustment to changes w.r.t requirements and/or operation conditions.



Smart transition to AI-based CPS

Main SCCH Research Topics

  • (deep) transfer learning for leveraging powerful but data-hungry learning methods for industrial applications where labeled training data is scarce,
  • privacy and security for distributed learning in computing clouds,
  • AI–based software analytics for enabling the innovation cycle in data and software-intensive systems.

Selection of AI related Projects

  • RePhrase (RIA H2020, ICT-09-2014): parallel patterns of machine learning for heterogeneous multicore systems
  • ALOHA (RIA H2020, ICT-05-2017, start January 2018): on runtime-adaptive, secure deep learning on heterogeneous architectures (our role: transfer learning + use case by PKE on critical infrastructure protection)
  • VISIOMICS (FGG/COIN): tumor diagnostics by multiomics data
    IoT4CPS (IKT d. Z., Leitprojekt IoT 2016): trustworthy IoT for cyber-physical-systems, trust in cyber physical (production) systems, security at design, deployment and run-time
  • PV-go-Smart (OÖ 2017 Digitalisierung): deep learning for PV and weather networks
    AutoDetect (OÖ 2017 Digitalisierung): fault diagnosis by robust learning
  • SKIN (FFG/KIRAS): protection of the outer skin of critical infrastructure by predictive behavioral analytics (based on deep learning)

Scientific Excellence (on transfer learning)

Werner Zellinger et al, „Central moment discrepancy for domain invariant representation learning“, International Conference on Learning Representations, (ICLR) [listed as top-tier conference on deep learning] on a metric-based approach for deep transfer learning; the approach outperforms state-of-the-art domain adaptation algorithms on standard sentiment analysis and object recognition benchmark datasets


  • Deep Learning Reading Club on “Applied Deep Learning” (together with Hamid Eghbal-Zadeh from JKU/CP and JKU/FLLL; as ISI course)
  • Currently 5 PhDs (JKU/FLLL, JKU/CP, JKU/FAW, TU Prague/CMP, University of St. Andrews/HWU)

AI-related scientific partners

  • Prof. S. Hochreiter (JKU/BIOINF): deep learning, LSTMs
  • Prof. G. Widmer (JKU/CP): deep learning, particularly in connection with probabilistic methods and trustworthiness
  • Prof. S. Saminger-Platz (JKU/FLLL): transfer learning
  • Prof. W. Wöß (JKU/FAW): continuous data quality management
  • Prof. G. Weiß (Univ. Maastricht, NL): AI, stream data processing
  • Prof. T. Heskes (Raboud University, NL): AI, causal inference
  • Prof. A. Birlutiu (1 December 1918 University of Alba Iulia, Romania): transfer learning
  • Prof. M. Matskin (KTH Sweden): data analysis workflow modeling
  • Prof. J. Matas (TU Prague/CMP): machine learning based computer vision;
  • Prof. B-S. Scholz (Hariot Watt University of St. Andrews, HWU, UK): high-performance computing for deep learning