DeepLearning Homeworks

The repository contains homeworks from Deep Learning course at Unipd 2021/22.

Homework 1 - From Perceptron to DNN

Implementation of the backpropagation algorithm for the perceptron and a two-layer network, with sigmoid and mean squared error.

Training for the handwritten digits dataset using softmax and cross entropy loss function.

Notebook: HW1 | Assessment: 100/100

Homework 2 - Optimize and Train Deep Models

Text classification with Tensorflow and Keras on the AG News Subset, dealing with overfitting.

Regularization methods as L1-norm, L2-norm, early stopping. Tuning of hyper-parameters using model selection and Grid Search.

Notebook : HW2 | Assessment: 90/100

Homework 3 - Convolutional Neural Networks

Development of a simple Convolutional Neural Network for image classification on the CIFAR-10 dataset.

Use of Tensorflow and Keras for hyper-parameters selection to obtain better models.

Notebook : HW3 | Assessment: 100/100

Homework 4 - Recurrent Neural Networks & Transformer

Development of simples Recurrent Neural Networks for sentiment analysis using the IMDB dataset.

Introduction and testing of the Transformer model and its parameters.

Notebook : HW4 | Assessment: 100/100

Homework 5 - Autoencoders

Comparison between Single Value Decomposition and different types of Autoencoders on CIFAR-10 dataset.

Linear and non-Linear Autoencoders for sequences with the IMDB dataset.

Notebook : HW5 | Assessment: 95/100

Homework 6 - Variational Autoencoders

Development of Variational Autoencoders with the MNIST Dataset.

Ad-Hoc layer and custom training step exploration.

Notebook : HW6 | Assessment: 100/100