This repository contains coursework, including assignments, projects, and tutorials from the Neural Networks (EEE443) course at Bilkent University.
-
Project Name: Text-to-Image Synthesis using Conditional Latent Spaces and Generative Adversarial Networks
Grade: 95/100
Prof. Feedback -> (Good overall results and report. Formatting could be improved.)
Abstract: Our project develops a deep learning model that turns textual descriptions into matching visual representations. Our model architecture combines Conditional Latent Spaces (CLSs) with Generative Adversarial Networks (GANs), producing images directly from textual descriptions. An iterative refinement method is also applied to incrementally enhance the quality of the synthetic images. For further information and access to the trained model, please refer to the provided Google Drive Link.
Grade: 51/100
TA Feedback -> ( 3.1 figures are not clear -2. 8.1 you can just put the figures of cost is decreasing instead of writing all costs. Try to visualize your results -2 20.1 no comparision of d and p values -10.)
-
Autoencoder Neural Network: Implementing an autoencoder neural network for unsupervised feature extraction from natural images.
-
Natural Language Processing: Constructing a neural network architecture for examining sequences of words with an aim to predict the fourth word in sequence given the preceding trigram.
-
Recurrent Neural Network Architectures: Classifying human activity from movement signals measured with three sensors simultaneously using different recurrent neural network architectures such as a single layer RNN, LSTM, and GRU.
Grade: 100/100
- One Hidden Layer Networks
- L-layer Networks ReLU
- Gradient Checking
- Residual Networks
- Optimizers
The datasets for the mini projects are available in the repository or via the provided links.
This repository includes materials such as quizzes from the course.
- Special thanks to the course instructor and TA for their guidance throughout the course.