Privacy-Preserving Machine Learning for Energy Prediction in Industrial Robotics

Strategic Insights -

Researchers at the AIT Austrian Institute of Technology GmbH have developed a privacy-preserving, data-driven cloud service for predicting the energy consumption of industrial robots. By integrating machine learning techniques with secure multi-party computation (SMPC), they have created a system that ensures confidentiality of proprietary model weights and confidential input trajectories. The study explores the feasibility and implications of this approach, evaluating the performance impact of SMPC on different network types and optimizing strategies to reduce inference overhead.

Key Takeaways:

  • The research explores the feasibility and implications of developing a privacy-preserving, data-driven cloud service for predicting the energy consumption of industrial robots.
  • The system uses machine learning to evaluate three neural network architectures: dense, LSTM, and convolutional-LSTM hybrids, to model energy usage based on robot trajectory data.
  • Models incorporating manually engineered features (angles, velocities, and accelerations) significantly improve prediction accuracy.
  • The researchers integrate privacy-preserving machine learning (ppML) techniques based on secure multi-party computation (SMPC) to ensure secure collaboration in industrial environments.
  • The study analyzes the performance impact of SMPC on different network types and evaluates two optimization strategies to reduce inference overhead.
  • The results highlight that network architecture plays a larger role in encrypted inference efficiency than feature dimensionality, with dense networks being the most SMPC-efficient.
  • The research identifies and discusses specific stages in the MLOps workflow, particularly model serving and monitoring, that require adaptation to support ppML.
  • The study provides insights useful for integrating ppML into modern machine learning pipelines.

Statistics:

  • The study evaluates three neural network architectures: dense, LSTM, and convolutional-LSTM hybrids.
  • The models incorporating manually engineered features (angles, velocities, and accelerations) improve prediction accuracy by x%.
  • The SMPC optimization strategies reduce inference overhead by y%.
  • The performance impact of SMPC on network types is analyzed and presented in a table.
  • The study uses Europian Research Council funding (EU Horizon Europe Work Programme; Austrian Research Promotion Agency Ffg Within The Present Project).

Sources:

  • Research article: Towards Privacy-Preserving Machine Learning for Energy Prediction in Industrial Robotics: Modeling, Evaluation and Integration, Machines, 2025,13(9):780. (Machines - http://www.mdpi.com/journal/machines)
  • Publisher: MDPI AG
  • DOI: https://doi-org.sdpl.idm.oclc.org/10.3390/machines13090780
  • News article: NewsRx, AIT Austrian Institute of Technology GmbH Researchers Provide Details of New Studies and Findings in the Area of Machine Learning, Journal of Engineering, October 13, 2025; p 18.