EEE443 Neural Networks

The assignment and project implementations for EEE443 course, Bilkent University.

Coverage of the repository

Assignment 1

• Finding the probability distribution of a network with a number of neurons.
• Building a neural network to perform a simple logic formula.
• Building a single layer perceptron to learn the representations of letters in the alphabet.
• Implemented in Python (Jupyter Notebook)

Assignment 2

• Proving the solution of mean squared error loss.
• Design and implement a neural network from scratch, and train it for a binary classification (car vs cat) problem.
• Implement a neural network that predicts the fourth word given the initial three. A simple NLP application.
• Implemented in Python (Jupyter Notebook)

Assignment 3

• Unsupervised feature extraction using a sparse auto-encoders. MSE loss function with the addition of Tikhonov (L2) regularization and KL-Divergence to ensure sparsity.
• Observation and completion of networks about CNNs, PyTorch and TensorFlow.
• Implemented in Python (Jupyter Notebook)

Project:

• @emredonmez98
• Built an image captioning networks that generates automated captions from an image.
• A custom Flickr dataset is used.
• A pretrained Inception V3 is used to extract features from the images.
• Then, a an attention based RNN decoder (with LSTMs) is trained (initialized with GloVe embeddings) is used to generate captions for the images.