Given-
- no. of input node - 784
- no. of hidden node - 100 (for training purpose , MNIST data set the researcher has taken 100 as it gives good accuracy suggested by the researcher.)
- no. of output node - 10 (Because in MNIST handwritten data set it only has 10 label numbers )
- Using the probability normal distribution function to initialise the weights of input_hidden and output_hidden.
- Using Sigmoid Activation Function to as it has smooth curve and it is widely implemented.
- And in neural network learning is done through weights , so to update the weights we have formula.
- And we have training data set of 100 length and testing data set of 10 len. Given epoch-2 so it will train the data twice and while querying we are getting good result .
- The output_node with the most value gives the correct results as it is giving [0.96681712] value for label (1).
- Giving an overall accuracy of 70% .