Issues to combat
1. Issues to combat
- NB: Performance issues: overfitting
- Too-complex-NN
- has too many weights
- will overfit training set, therefore generalize badly
- Optimization
Remove irrelevant & redundant…
- input variables
- hidden units
- weights
2. Architecture selection approaches:
- By trial-and-error
- Regularization (penalty methods)
- Pruning
- Growing
- Meta-Heuristic Algorithms:
- optimization techniques that explore and exploit the search space efficiently.
