Fig. 7.

Average mutual information (in bits) between movement direction, θ, and rate estimators averaged across the two subgroups of neurons in the entire population as a function of decomposition level (i.e., kernel size). Solid lines indicate the performance of the EDWT method (dark for the broad tuning group and gray for the sharp tuning group). The two dashed lines represent the Gaussian kernel method (broad tuning and sharp tuning groups), while the two dotted lines represent the rectangular kernel method in a similar way. As expected, sharply tuned neurons require smaller kernel size to estimate their firing rates. Overall, the EDWT method achieves higher mutual information than either the fixed width Gaussian or rectangular kernels for broadly tuned neurons, while slightly less for sharply tuned neurons owing to the relatively more limited response time these neurons have, limiting the amount of data.