Figure 1: Methods and behavioral task.

A, Mapping of local field potentials from continuous space to point process space. Using the generative model, a dictionary of typical oscillatory patterns is learnt from bandpassed LFP traces (training). Construction of the point process space for LFPs follows from a convolution between LFP traces and the learnt dictionary (testing).B, Example evaluation of joint entropy term in directed information (equation 16) using matrix formulation of Renyi’s entropy, α-entropies. At time i, each time-shifted spiking matrix is projected onto a high–dimensional RKHS defined by the kernel, κ. Eigen-spectrum of the Hadamard product between the normalized Gram matrices captures the joint entropy between the two time series. C, (Top) Experimental setup. Following a 1s-fixation period, the subject is shown a sample stimulus for 1s. At the end of a delay period (fixed/variable), the subject saccades to the sampled stimulus that reappears at one of four randomized locations along with distractor images. D, Exemplar spiking activity and LFPs from layer 2/3 of V4 visualized in the same space of point processes, subject 2. For visualization purposes, every peak of the detected oscillatory burst has been demarcated. E, Multi-scale average rate plots obtained via kernel smoothing methods, subject 2. All activity is aligned with the trial intervals in C.