• 💬 Ask me about: My post-university plans!

• 📫 How to reach me: jessie.kurtz@mail.mcgill.ca

👨‍💻 Reinforcement Learning - COMP 579 @McGill:

1. Trading Deep Q-Networks (TDQN)

• Enhanced a trading Deep Q-Network with components from the Rainbow DQN architecture (Dueling Networks, Noisy Nets, Prioritized Experience Replay) and implemented an Actor-Critic algorithm.
• Demonstrated improved trading performance on Tesla and Apple stocks, maintaining profitability even during market downturns.

2. Deep Q-Network and Expected SARSA for Atari Games and Robotic Control (Acrobot)

• Applied Deep Q-Network (DQN) and Expected SARSA algorithms to an Atari game and an Acrobot robotic arm system.
• Demonstrated effective learning in both environments.

👨‍💻 Applied Machine Learning - COMP 551 @McGill:

1. Emotion Classification

• Evaluated DistilGPT-2, Multinomial Naive Bayes, and Random Forest models for classifying the GoEmotions dataset, focusing on accuracy, recall, and efficiency.
• DistilGPT-2 performed best, with a test accuracy of 0.6168.

2. Image Classification - OrganAMNIST dataset

• Asessed MLPs, CNNs, and MobileNetV2 for classifying the OrganAMNIST dataset, emphasizing accuracy, recall, and computational efficiency.
• MobileNetV2 with frozen convolutional layers and trainable fully connected layers achieved the best results, with a test accuracy of 0.9249.

3. Oral Temperature Predictor - Linear Regression

• Compared mini-batch stochastic gradient descent (SGD) to full-batch SGD to predict average oral temperatures.
• Mini-batch SGD presented quick convergence and a high R^2 score

4. Diabetes Predictor - Logistic Regression

• Compared mini-batch stochastic gradient (SGD) to full-batch SGD to predict presence or absence of diabetes.
• Mini-batch SGD presented quick convergence and a high F1 score

👨‍💻 Older Projects:

1. Sports Center Website

• Created a full-stack web application in a team of 6 following an Agile methodology.
• Modified the UI according to three different user roles, here Owner, Instructor and Customer.
• Implemented 403 and 404 error pages for security and usability of the application.
• Designed the frontend in Vue, using Vuetify and Bootstrap frameworks.
• Configured the backend in Java using the Spring Boot framework.

2. Firefighting Robot

• Assembled and programmed a robot to deliver the correct fire suppressant cubes to three hypothetical fires located on a grid, according to user input.
• Utilized control sensors, motors, a BrickPi and a Raspberry Pi.
• Used Breadth-First-Search (BFS) Algorithm, coded in Python, to find the quickest path to the hypothetical fires on the map.