Information Technology Reference
In-Depth Information
simplified version of 2T-ALOPEX. Finally, Haykin et al. [63] improved conver-
gence by combining the original formulation with that of Bia. Moreover, Haykin
et al. provide a good contextual introduction and derivation of ALOPEX, while
Sastry et al. prove that 2T-ALOPEX behaves asymptotically as a gradient-descent
method. Also, Meissimilly et al. [103] introduced parallel and pipelined implemen-
tations of ALOPEX applied to template matching with corresponding computational
and temporal complexities of calculating the global response function R k .
13.7.3 Genetic Algorithms and DBS
Feng et al. [39] use a model by Terman et al. [149] to test a method of stimulus
administration where each stimulus parameter is obtained from a distribution of such
measures, thus incorporating a degree of randomness in the stimulus waveform.
Moreover, in this method, the shape of each distribution curve is a piecewise linear
model where the model parameters are modified by a genetic algorithm that seeks
to reduce the cross-correlation and/or autocorrelation of measurements taken from
multiple sensors. Figure 13.9 shows a diagram of the method proposed by Feng
et al.
Fig. 13.9: The method proposed by Feng et al. [39] to draw deep brain stimula-
tion parameters ( I i DBS ) from distributions whose shape descriptors ( a i ) are selected
by a genetic algorithm that seeks to minimize correlations in measured data ( x i ).
Constraints ( R ) on the genetic algorithm may be imposed externally.
13.7.4 Hardware Implementations
Various components of a closed-loop system have been implemented as a
microelectronic design, including power and telemetry components [159], and
stimulus/recording circuits interfacing with an external computing platform [90].
A typical setup for the real-time transmission of biosignals from a neural implant is
shown in Fig. 13.10 [159].
Search WWH ::




Custom Search