General Profile:

Having recently graduated with a bachelor’s degree in mathematics and physics, I am excited to continue my academic journey within the IAI CDT program. My first exposure to the world of Machine Learning occurred during my role as a data scientist at a consulting firm, where I used tree models to classify clients according to their behaviour. During my final year project, I had the opportunity to design Convolutional Neural Networks capable of analysing weather radar data, an experience that kindled my passion for the intersection of data science, mathematics, and computer science.

Since that pivotal moment, I've been drawn to the field of Artificial Intelligence, which has driven me to set my sights on pursuing a Ph.D. in Interactive AI. My primary goal is to develop AI systems that are not just intelligent but also transparent, reliable, and capable of collaborating with humans. I am dedicated to enhancing the interpretability and adaptability of AI systems, avoiding the "black-box" technology stereotype.

 Beyond my academic pursuits, I have a deep interest in sports and outdoor activities, such as climbing, surfing, running, and cycling.

Research Project Summary:

This project aims to make reinforcement learning (RL) agents' decision-making more understandable, reliable, and amendable by creating methods that generate programs—or sets of instructions—that guide how an AI agent acts. Unlike traditional approaches to RL that depend heavily on complex neural networks, this project focuses on developing programmatic policies—essentially, programs that define what actions an agent should take in various situations. The main question it seeks to answer is:  Can we create more interpretable and adaptable decision-making processes for RL agents by using program-based approaches rather than purely neural methods? To answer this question, I aim to fulfil the following objectives:

- Design a framework for generating programmatic policies that guide agent decision-making, making the AI's behaviour more transparent.

- Evaluate and compare the effectiveness and adaptability of these program-based policies with standard neural network-based methods across different tasks.

- Explore how human input can be incorporated into these policies, allowing users to modify or refine the AI's decision-making when necessary.