Public defence in Computer Science, M.Sc. Alexander Nikitin

Public defence from the Aalto University School of Science, Department of Computer Science
Doctoral hat floating above a speaker's podium with a microphone

Title of the thesis: Probabilistic methods for predictive maintenance and beyond: graph and human-in-the-loop machine learning 

Doctoral student: Alexander Nikitin
Opponent: Prof. Stefan Feuerriegel, Ludwig Maximilian University of Munich, Germany
Custos: Prof. Samuel Kaski, Aalto University School of Science, Department of Computer Science

Inspired by real-world challenges in telecom predictive maintenance, this thesis introduces probabilistic methods with wide-ranging applications. The doctoral thesis encompasses three key areas: 1. Spatio-temporal Graph Kernels: The thesis introduces a pioneering approach to deriving spatio-temporal kernels on graphs from stochastic partial differential equations. These kernels, in particular, can be used with Gaussian processes and enable predictive methods for many applications (such as epidemiology or traffic forecasting, not to mention predictive maintenance). 2. Human-in-the-Loop Methods: The research delves into the integration of human expertise into machine learning methods. It proposes decision rule elicitation, a method that uses explicit feedback from human experts to enhance the performance of data-driven models. This method improves domain adaptation and is effective in predictive maintenance applications. 3. Synthetic Data Generation: A crucial aspect of collaborating with industrial organizations is data availability. This thesis contributes to the field of synthetic data generation by developing a tool for synthetic time series generation and evaluation called TSGM. In particular, the framework proposes a versatile set of metrics for synthetic data evaluation, which allows practitioners to compare different synthetic data approaches with respect to a given application. 

The practical implementation of these methods in the telecom domain serves as a testament to their applicability in real-world scenarios. The results are not only academically significant but also hold practical value for industries seeking to enhance their predictive maintenance capabilities. Several patents were granted based on this research. This doctoral defense is of particular interest to machine learning researchers and practitioners exploring application-driven machine learning systems or engaging in probabilistic methods research. Moreover, this thesis might be valuable to anyone interested in predictive maintenance or in telecom applications of machine learning.

Key words: Machine Learning; Deep Learning; Gaussian Processes; Stochastic Differential Equations; Generative Models; Predictive Maintenance; Human-in-the-loop; Open Source

Thesis available for public display 10 days prior to the defence at: 

Doctoral theses in the School of Science: 

  • Published:
  • Updated:
URL copied!