Time Series, Complex Networks, Algorithms
Work description
The research work to be carried out will involve the following steps: Study and analyze existing methods for synthetic data generation in multivariate time series contexts, particularly those based on GANs, noise-based perturbation, and graph-based representations. Explore the theoretical foundations of quantile graphs, multilayer network models, and Markov models for representing the temporal dynamics of time series data. Adapt, refine, and expand existing inverse quantile graph mapping methods by incorporating higher-order temporal lags and higher-order structural dependencies, based on information from Multilayer Quantile Graphs. Compare the developed method with state-of-the-art techniques for synthetic data generation, such as TimeGAN, DoppelGANger, and noise-based perturbation techniques. Evaluate performance based on metrics of distributional similarity, correlations, dynamic behavior preservation, and privacy evaluation measures. Test the developed method and demonstrate its applicability on different real-world datasets (e.g., energy consumption and physiological data), where temporal dynamics, inter-variable interactions, and the presence of sensitive information are critical factors. Compile findings into a final report.
Academic Qualifications
Bachelor degree in Data Science, Mathematics, Computer Science, or a related field.
Minimum profile required
Familiarity with methods of time series modelling, Markov models, and graph theory concepts.Ability to work independently, critical thinking, and interest in scientific research.
Preference factors
Previous work involving time series analysis and/or graph analysis. Knowledge of tensor analysis and algebraic manipulation methods, as well as time series mapping techniques into graph structures. Experience with the R or Python programming languages.
Application Period
Since 26 Jun 2025 to 09 Jul 2025
Centre
Artificial Intelligence and Decision Support