Question (4): State whether (true) or (false) and justify your answer. a- Neural network classification always give the best performance. b- If training a neural network does not do well, we need to add a softmax layer. C- If training a neural network does very well but testing is bad, we should change the optimization strategy. d- The tanh activation function is always the best choice. e- Non-CNN based image classification uses machine learning technques giving good results regardless how the image feature is designed. f- CNN image classification models are easily trained and do not need special computer configurations. g- SIFT descriptors feature extraction is one of the hand crafted design techniques. h- PCA is used to decrease hand crafted features' size. i. Non separable data points cannot be classificed even if SVMs is used. j- Bayes' classifier using multi-variate Gaussian models is the best segmentation technique when image feature size is large