Single trial analysis using non-negative matrix factorizations

• We found that NMF may help to uncover hidden structure in single trial data. In the fMRI data reported here, three clusters of trials were found. • We found that the city recognition value correlated with cluster membership so that we can assume that the decomposition revealed meaningful information. • We also found that a low recognition value correlated with a deactivation in aFMC. A possible interpretation might be that a deactivation of the Recognition Heuristic leads to a suppression of the default mode network. REFERENCES [1]Volz, et al. (in press), Why you think Milan is larger than Modena: The neural correlates of the recognition heuristic, J Cogn Neurosci. [2] Lee & Seung (1999), Learning the parts of objects by non-negative matrix factorization, Nature, 401:788-791. The reaction to a stimulus may differ across trials of the same stimulus type. For instance, the stimuli may consist of sentences that evoke different reactions depending on their content. If those reactions are not predictable then the differences between trials cannot be described by the experimental design. Rather, they can only be inferred by differences in the BOLD responses that they evoke. Idea: 1. Use a new matrix decomposition method (NMF) to uncover such hidden struc ture in the data. 2. Use the NMF decomposition to investigate the interaction between trials in across different brain regions.