We wanted to thank Damien Fay from HPE for presenting at our 2024 Seminar Series yesterday. Demian gave an interesting presentation: “A self-organizing eigenspace map for time series clustering”
His talk focused on a time series clustering technique they developed as well as the underlying time series analysis techniques used which can handle non-stationary chaotic and semi-seasonal time series. He spent some time during the talk giving a tutorial on this before launching into the research which extends the theory substantially (see abstract below).
This paper presents a novel time series clustering method, the self-
organising eigenspace map (SOEM), based on a generalisation of the well-known self-organising feature map (SOFM). The SOEM operates on the eigenspaces of the embedded covariance structures of time series which are related directly to modes in those time series. Approximate joint diagonalisation acts as a pseudo-metric across these spaces allowing us to generalise the SOFM to a neural network with matrix input. The technique is empirically validated against three sets of experiments; univariate and multivariate time series clustering, and application to (clustered) multivariate time series forecasting. Results indicate that the technique performs a valid topologically ordered clustering of the time series. The clustering is superior in comparison to standard benchmarks when the data is non-aligned, gives the best clustering stage for when used in forecasting, and can be used with partial/non-overlapping time series, multivariate clustering and produces a topological representation of the time series objects.
organising eigenspace map (SOEM), based on a generalisation of the well-known self-organising feature map (SOFM). The SOEM operates on the eigenspaces of the embedded covariance structures of time series which are related directly to modes in those time series. Approximate joint diagonalisation acts as a pseudo-metric across these spaces allowing us to generalise the SOFM to a neural network with matrix input. The technique is empirically validated against three sets of experiments; univariate and multivariate time series clustering, and application to (clustered) multivariate time series forecasting. Results indicate that the technique performs a valid topologically ordered clustering of the time series. The clustering is superior in comparison to standard benchmarks when the data is non-aligned, gives the best clustering stage for when used in forecasting, and can be used with partial/non-overlapping time series, multivariate clustering and produces a topological representation of the time series objects.
Bio:
Damien Fay recently joined HPE labs Ireland to build out the data centre energy usage and modelling group. He obtained his PhD in time series analysis under J. Ringwood in 2003, was a lecturer in stats in NUIG 2003-2007, researcher from 2008-2012 in Cambridge, McGill and UCC. From 2012 to 2016 he was a lecturer at the data science institute in Bournemouth university. In 2016 he joined Infor as a senior principal involved in developing models for retail F&R. His research interests include applied graph theory, time series analysis and applied statistics/ML.