DIGIT FEEDFORWARD NEURAL NETWORK

Neural networks have been a very powerful tool for machine learning problems. Nowadays there is a considerable amount of neural network architectures, and they vary a lot depending on the problem and applications.

 

We analyze here the feed-forward neural network topology for the digit classification problem, already presented earlier. We start from the simplest networks, moving to the multilayer perceptron.

​

Neural networks pose a huge amount of possible parameters and settings. In our study we focus on analyzing different hyperparameters, evaluating each one individually, in order to reach the best model. We will keep on our studies in next section, with Convolutional Neural Networks.