The presentation broadly focuses on two aspects. The first is my research experience on cognitive classification applying various machine learning techniques, and the second is my research plan as part of postdoc activity at UFAL.
In the real world scenario, functional magnetic resonance imaging (FMRI) data are highly complex and high dimensional, severely under-constrained and interspersed with a vast quantity of irrelevant or redundant features. Therefore, the problem of understanding the intentions of scanned subjects is a difficult pattern recognition task of FMRI analysis. At the same time, machine learning algorithms based on stochastic and probabilistic approaches such as genetic algorithms, naive Bayesian classifier, decision trees, etc. are well established for pattern recognition tasks in the medical domain. Hence, the task of decoding the cognitive states instantaneously through FMRI analysis based on machine learning approaches is a natural and interesting goal. The presentation will focus on related machine learning techniques, my proposed framework and various ensemble and non-ensemble methodologies for classifying different cognitive states.
My future research plan (a postdoc at UFAL) includes creating a Czech and Hindi variation of Visual Genome, a multi-modal network (a visual "wordnet", if you wish), a dataset that links visual and textual information and is potentially very useful for training of multi-modal neural machine translation.