Deep Learning

This is a collection of projects by Dallin Stewart that I worked on for my Deep Learning class.

Table of Contents

Welcome
Cancer Detection
Cartpole with RL and PPO
RNN Text Generation
Stable Diffusion
Style Transfer
English to Spanish Translator

Welcome

This repository is a collection of projects that I started as an assignment for my computer science classes that is still under development. Credit for a lot of the ideas in these projects goes to the CS 474 curriculum. Each of the projects are distinct, but most use tools from linear algebra, calculus, and probability including:

Convolutional Neural Networks

Recurrent Neural Networks

Reinforcement Learning

Q Learning

Proximal Policy Optimization

Transformers

Generators

Markov Chains

Transfer Learning>/li>

Working with GPT-2

Working with Stable Diffusion

Applied and Computational Math Emphasis

Brigham Young University’s Applied and Computational Math Emphasis (ACME) program is a major designed for solving the problems of the 21st century. Mathematics provides the foundation of modern technology and science, it is the key to building successful algorithms in artificial intelligence and machine learning, and it provides the analytical power needed to process, evaluate, and take full advantage of the ever-growing flood of data and information in the world. ACME is a new educational model that teaches both the theory and the practical skills in mathematics, statistics, and computation needed to solve the problems of the modern world.

Project Descriptions

Cancer Detection

This project uses a U-Net convolutional neural network structure to detect cancerous cells in images

Cartpole with RL and PPO

This project uses a Proximal Policy Optimization (PPO) algorithm to train a reinforcement learning agent to balance a cartpole. The agent is trained using a custom environment that is based on the OpenAI Gym Cartpole environment. The environment is modified to allow for a more complex reward system and to allow for the agent to be trained on a continuous action space.

RNN Text Generation

This project uses a Recurrent Neural Network (RNN) to generate text.

Stable Diffusion

This project uses the Stable Diffusion model to generate images. The model is trained on the CIFAR-10 dataset. The model is then used to generate images that are similar to the images in the dataset.

Style Transfer

This project uses a convolutional neural network to transfer the style of one image to another image.

English to Spanish Translator

This project uses a Transformer model to translate English to Spanish. The model is trained on a dataset of English and Spanish General Conference taks.

(back to top)