Defence of dissertation in the field of Signal Processing for Telecommunications, M.Sc. (Tech.) Shahab Basiri
The title of thesis is “Robust large scale statistical inference and ICA using bootstrapping”
We live in an era of data deluge, where massive data volumes are generated by satellites,
telescopes, high-throughput instruments, sensor networks, and supercomputers. Yet, as
the size of data grows, so does the chance to involve outlying observations. This in turn
motivates the need for robust (i.e., outlier-resilient) statistical analysis tools which can
scale to big data sets.
The goal of this thesis is to develop robust statistical inference tools that are able to assign
measures of accuracy (such as bias, variance, confidence intervals, or prediction error) to
sample estimates that are calculated from large-scale data. Such tools are required, for
example, for accurate decision making, choosing appropriate data analysis tools, or identifying
relevant variables or features from the data.
The developed methods are founded on non-parametric data resampling method, called
the bootstrap, which is particularly useful tool for large-scale data analysis as it avoids
making parametric model assumptions on the data. The big data bootstrap methods developed
in the thesis are compatible with distributed storage systems and parallel computing
architectures and enable, for the first time, conducting robust statistical inference for largescale
data using the popular bootstrap principle. In addition, also robust statistical inference
tools for the independent component analysis model are developed and applied in
Electroencephalography (EEG) signal processing.
Opponents: Professors Daniel Palomar, Hong Kong University of Science and Technology and Jean-Yves Tourneret, University of Toulouse, France
Supervisor: Professor Esa Ollila, Aalto University School of Electrical Engineering, Department of Signal Processing and Acoustics.