پیاده سازی single layer perceptron و multi layer perceptron در ماشین لرنینگ

---

شرح پروژه

Part 1) The Single Layer Perceptron (SLP)
a) Implement an SLP where the inputs are 784-dimensional flattened images and the outputs
are 10-dimensional class probabilities.
b) Train the perceptron for a number of epochs (e.g., 25) on the training dataset and plot the
misclassification rate after each epoch on both training and test datasets.
c) Report the confusion matrices on both training and test datasets at the end of training.
d) Each output unit has 784 weights coming to them. Visualize those weights as a 28x28
image for each output unit.
Part 2) The Multi-Layer Perceptron (MLP)
For H=20, 30, 50
a) Implement an MLP where the input is a 784-dimensional flattened image, the hidden
layer has H units, and the outputs are 10-dimensional class probabilities.
b) Train the perceptron for a number of epochs on the training dataset and plot the
misclassification rate after each epoch on both training and test datasets.
c) Report the confusion matrices for both training and test datasets at the end of training.
d) Each hidden unit has 784 weight values coming to them. Visualize those weights as a
28x28 image for each hidden unit.
Part 3):
a) Derive the update rules for an MLP with two hidden layers of H1 and H2 hidden units
using gradient-descent.
b) Implement three MLPs with 784 inputs, (H1=20, H2=20), (H1=30, H2=20) and (H1=40,
H2=20) hidden units in its two hidden layers, and 10 outputs.
c) Train each perceptron for a number of epochs on the training dataset and plot the
misclassification rate after each epoch on both training and test datasets.
d) Report the confusion matrices for both training and test datasets at the end of training.
e) Each hidden unit in the first hidden layer has 784 weight values coming to them.
Visualize those weights as a 28x28 image for each hidden unit.

نکته1) الگوریتم های فوق باید روی دیتاست

MNIST

پیاده شوند.

نکته2) از کتابخانه های ماشین لرنینگ مثل 

SciPy, PyTorch, Keras, Tensorflow
نباید استفاده شود.

نکته3) برای دیتا هندلینگ و محاسبات ماتریس میتوان از کتابخانه های

Numpy and Pandas
استفاده کرد.

مهارت ها و تخصص های مورد نیاز

---

---