Category: Machine Learning
-
Trading to Protect the Grid: A Reinforcement Learning Trading Agent in Electricity Markets
In this project, I designed an AI trading agent for electricity markets using deep reinforcement learning, trained with curriculum learning. The agent showcases strong performance in day-ahead power markets by taking in price and grid forecasts (that I developed myself using transformer models) and outputting optimal bids for the next…
-
Time-Series Forecasting for Day-Ahead Markets
Modern electricity markets are highly dynamic systems. Across a day, the electricity price can fluctuate rapidly and reach peaks several factors higher than the baseline or reach lows of negative prices. This volatility requires accurate forecasters in order to make decisions in such an uncertain environment. In this project, I…
-
Developing my Own Chess AI
While many chess computers display impressive superhuman performance, they often require massive computational ressources and/or extensive expert level knowledge and finetuning during their development. In this project, I wanted to challenged this supposition and developed a personal chess AI on my own using reinforcement learning together with a Monte Carlo…
-
Power Grid Automation: Bridging Machine Learning and Optimal Control
The operation of modern power grids has become increasingly challenging with the transition to renewable energy sources. As a consequence, the transmission system operator of France has launched a series of competitions, called L2RPN [1], to explore novel methods to operate power grids. During my work at Hitachi Energy, I…
-
Self-Supervised Learning on Wearable Data: Missing Data Completion
Together with the ETH AI Center, I explored a self-supervised learning method to address the issue of missing data in wearables. Wearable devices such as smartwatches suffer from a high rate of missing data, e.g., when the sensor loses contact with the skin. This poses a big problem for downstream…
-
Automated Machine Learning: Training Neural Networks with Evolution
A typical machine learning application often involves a challenging procedure composed of data preprocessing, feature extraction, architecture selection, and hyperparameter tuning. Automated machine learning, or AutoML, aims to simplify this complex process. In this project, I designed an AutoML algorithm using a cascaded architecture comprised of a Nelder-Mead method and a…
-
Blackjack Agent: Combining Reinforcement Learning and Card-Counting
Blackjack is one of the most popular casino games in the world even though it is inherently unfair towards the player and strongly favors the “house”. In order to still make a profit, professional Blackjack players resort to a strategy called card counting. In this project, I combined card counting…
-
Coding an Artificial Neural Network from Scratch
In my second year of my bachelor’s studies, I encountered the field of machine learning and was immediately captivated by it. Eager to learn more, I taught myself more about neural networks and decided to build one from scratch in Java, without the help of any libraries. The result was…