Hello! I want to let the people know that my Master’s thesis is now publicly available at digital commons. My defense slides are also available there (link). I previously made the project code freely available, too.

The thesis poses financial forecasting (specifically: realized volatility forecasting) as a meta-learning problem. I’ve learned a lot since doing this work, but I can say it was a tremendous learning experience in more ways than I can briefly recount here. It’s also an invaluable reference point to look back on for the work I’m doing now.

I believe the main pedagogic benefit for readers is in helping them to understand how to apply a model-based / black-box meta-learning approach to a real-world time series problem. The thesis provides a detailed mechanistic description of the meta-learning techniques used.

I’m going through Chelsea Finn’s excellent meta-learning lectures (Stanford CS330: Deep Multi-Task and Meta Learning), and doing the thesis prior definitely enhanced my experience of it. It not only provided me relevant background, but it also gave me a tangible and engaging problem context to think about the ideas.

Message me about my thesis (or to talk about anything interesting) here.