In this presentation I will touch upon two data-driven approaches to neuroimaging data. In the first part, I will introduce the so-called energy landscape analysis. In this analysis, one regards noisy brain dynamics as those of a "ball" constrained on an energy landscape inferred from multivariate time series data (such as fMRI data recorded from a set of ROIs). A ball tends to go downhill on the energy landscape whereas it sometimes goes uphill to transit from one "local minimum" to another, possibly corresponding to major transitions in the brain. The method is based on the Ising model in statistical physics (also known as Boltzmann machine). I will also show the application of the method to bistable visual perception tasks and some other situations. In the second part, I will overview network analysis methods, drawing upon recent development of network science. In particular, I will discuss some mesoscopic structures of networks and temporal changes in networks.