A repository of assignments performed during the Advanced Machine Learning course.
The assignment consists in the prediction of default payments using a feed Forward Neural Network.
The provided dataset contains information on default payments, demographic factors, credit data, history of payment, and bill statements of credit card clients in Taiwan from April 2005 to September 2005 (X_train, Y_train). This two dataset can be used for training the FFNN, and eventually for the validation. This dataset can be used to test model prediction.
Regularization&Autoencoder Folder
The assignment consists in the prediction of grayscale images of handwritten letters P - Z with a traditional neural network (FFNN) with the right regularization. It's also require to build a simple auto-encoder model and investigating visually the reconstruction abilities of such an architecture, and optional the use and evaluation of the encoded representation generated by the auto-encoder to solve the problem of supervised classification.
The Dataset consists of 14000 training labelled images and 8800 test images.
The assignment consists in the prediction of grayscale images of handwritten numbers 0-9 with a convolutional neural network (CNN). The dataset is the classical MNIST dataset. The only requirment is to implement a CNN with up to 6000 trainable parameters.
The task of this assignment is Transfer Learning using a CNN pre-trained on IMAGENET, the suggested architecture is the VGG16. The CNN should be used as fixed feature extractor on a new task of your choice containing a number of classes in the range from 2 to 10. It is also required to consider three different layers as cutting point and analyze the performace for each choice
⊜ Riccardo Confalonieri
- Enrolled in: Master degree in Data Science at University of Milano-Bicocca.
- Previously enrolled in: BSCS in Information and Communication Technologies at University of Milano-Bicocca.
