Biomedical Engineering Reference
In-Depth Information
FIgURE 3.13:
Standard deviation in training MSE for 100 Monte Carlo simulations.
are presented. This figure is characterized by a large standard deviation in the initial epochs result-
ing from the initial condition and a small standard deviation in the final epochs signifying that the
models are converging to the same solution. These relationships are quantified in Table
3.3
where
the average final MSE had a value of 0.0203 ± 0.0009. Again, a small training standard deviation
indicates the networks repeatedly achieved the same level of performance. Of all the Monte Carlo
simulations, the network with the smallest error achieved an MSE of 0.0186 (Table
3.4
).
We can see in the testing outputs for the RMLP shown in Figure
3.14
that the performance
of a parsimonious nonlinear dynamical network can improve the testing performance for trajectory
reconstruction in a hand-reaching BMI task. The trajectories are clearly smoother and distinctly
reach each of the three peaks in the movement (CC = 0.84 ± 0.15).
TaBlE 3.3:
Training performance for 100 Monte Carlo simulations
TRaININg STaNdaRd
dEVIaTIoN
all RUNS
TRaININg MINIMUM
Average of minimum MSEs
0.020328853
0.000923483
Average of final MSEs
0.020340851
0.000920456
