Backpropagation Algorithm in ANN

Consider, a dataset with 1000 2-dimensional data with labels which can be obtained from the ‘i-folio’ website (file name is “[login to view URL]”). The dataset consists of 1000 lines of x, y, and its class value. Please download the reference paper that discussed the NetTalk algorithm for you to refer (file name is “Ref_1 Parallel ….. pdf”). Using the BPA algorithm, the smoothing parameter α can be set to 0 and ignore the smoothing effect. 1. Implement a multilayer perceptron (MLP) according to the update equations mentioned in the NetTalk paper to classify the dataset and consider using a MLP with a 2-4-3-1 structure. Please show the plot of the output error curve during training and discuss your results. 2. After training, you can construct the hidden tree for the obtained network. Show the hidden tree for all layers except input to the first hidden layer and identify if there exists redundant neurons or mixed regions. Discuss your results. 3. Repeat the experiment but this time please use the 2-8-6-1 structure on the same dataset and briefly state your findings. Hint: The 2-dimensional dataset are distributed within x ∈ [−1, 1] and y ∈ [−1, 1]. The initial weight range, the learning rate, and number of iterations should be carefully considered.