. 2019 Apr 8;18:44. doi: 10.1186/s12938-019-0659-9

Table 1.

Features set exploited in the cases of velocity and force custom decoder

	Forces	Velocities
Feature 1	$F e a t 1 (t_{i}) = - 1 + \frac{1}{{(A F R (t_{i}) - 1.5)}^{2}}$	$F e a t 1 (t_{i}) = \frac{A F R (t_{i}) - A F R (t_{\min})}{t_{\min} - t_{i}}$
Feature 2	$\{\begin{matrix} 1, \land F e a t 1 (t_{i}) \geq T_{u 1} \\ 0, \land F e a t 1 (t_{i}) < T_{u 1} \end{matrix}$	$\{\begin{matrix} 1, \land F e a t 1 (t_{i}) \geq T_{u 1} \\ 0, \land F e a t 1 (t_{i}) < T_{u 1} \end{matrix}$
Feature 3	$\{\begin{matrix} 1, \land F e a t 1 (t_{i}) \leq T_{l 1} \\ 0, \land F e a t 1 (t_{i}) > T_{l 1} \end{matrix}$	$\{\begin{matrix} 1, \land F e a t 1 (t_{i}) \leq T_{l 1} \\ 0, \land F e a t 1 (t_{i}) > T_{l 1} \end{matrix}$

The choice of the exponential of the AFR in the case of force decoding was driven by the fact that the FR-force relation (Eq. 9) approximates a logarithmic profile. t_i represents the time instant. AFR = (FRs average). $t_{\min}$ represents the minimum firing rate identified before subjects’ activity. T_u1, T_l1 are thresholds chosen to identify the intervals in which the neuronal activity is high (as during the execution of motions) or low (as during rest). In particular they are used by the decoder to identify the time points in which giving a prediction output