Learning Performance in the Video Quality Problem

Here, we give the results of the GD and AM-LM to learn the problem of video quality assessment presented in Chapter 6. The training database contains 80 samples (each is composed of 5 inputs and one output). The topology of the RNN is the three-layer feedforward architecture having 5 neurons in the input layer, 5 hidden neurons and one output neuron. We trained this network using the two algorithms; the stop criterion was MSE=0.0025. For GD, we used $\eta=0.1$ and for AM-LM, we used $dP=0.7$ and $\zeta=0.90$. We depict in Figure 10.7 the variation of the error with respect to the number of iterations. As we can see, AM-LM gives better performance than GD in terms of speed: it takes only 7 iterations in 47.37 sec., while GD reaches the same error after 5870 iterations in 58538.5 sec.

Figure: Comparsion between the performance of GD and that of AM-LM on the video quality database presented in Chapter 6.

[GD training algorithm] [AM-LM training algorithm]