Multilayer Feedforward Neural Networks
- Variation of error as a function of time-step/epoch
(a) Observe the variation in case of pattern mode and batch mode. In the former case, the variation of error would be less smoother than in the latter case.
(b) In each case, observe the number of epochs required for the error to converge to a given value.
- Convergence of weight values
(a) Observe the plot of individual weight values as functions of time-step/epoch. Each weight value may converge over time, or oscillate within limits, indicating that it is near a minimum in the error surface.
(b) Observe the number of epochs required for convergence in the case of pattern mode and batch mode.
- Cross-validation
(a) For the subset of input patterns, observe the values of the corresponding outputs, and compare the obtained outputs with desired outputs. What is the error between the two?
(b) Is the cross-validation error lesser for batch mode or for the pattern mode?
