Search
Results
-
Optimizing Recommender Systems for Social Networks
(2023)Social media platforms employ recommender systems to engage users with content. The advertised content depends not only on the user’s interest history but also on the interactions with his peers. The state-of-the-art methods use brute-force machine learning approaches that combine features describing personalization, popularity and similarity of interests with other users on the social platform to provide a set of media feed that the user ...Master Thesis -
Data-driven predictive control for optimising heat production and distribution in buildings
(2023)Buildings use a significant amount of energy for heating and cooling. Therefore, Building Automation and Control Systems aim to improve the energy efficiency of buildings by including advanced control approaches, such as Model Predictive Control (MPC). MPC is a widely used optimisation-based control strategy, which can include weather forecasts, building thermodynamics and system constraints into the control problem to optimise the heating ...Master Thesis -
Time-Varying Safe Bayesian Optimization
(2023)Ensuring safety is a key aspect in sequential decision making problems with uncertainty, such as robot controller design and medical therapy development. This is especially challenging when the underlying problem is potentially time-varying. We consider the problem of optimizing an unknown time-varying reward function subject to unknown time-varying safety constraints. We develop Safe Bayesian optimization (SBO) algorithms in two scenarios. ...Master Thesis -
-
Natural Bayesian Statistics
(2023)In this thesis, we develop Natural Bayesian Statistics, a new statistical theory that is intended as a replacement for standard Bayesian statistics in most situations. It is fully defined for families of probability distributions where the transformations between different random variables of that family have a group structure, like location-scale models. Additionally, it also applies to submanifolds of these models, like deep neural ...Master Thesis -
Enhancing toxicity prediction of MLinvitroTox: Prioritizing unidentified compounds in environmental samples based on hazard assessment
(2023)This thesis enhances the capabilities of the MLinvitroTox framework, which is designed to forecast the toxicity of unidentified chemical compounds utilizing High-Resolution Mass Spectrometry (HRMS/MS) data. The framework aims to maximize the detection rate of most hazardous compounds within environmental samples, all while minimizing the occurrence of false alarms, thus channeling resources and efforts into the labor-intensive task of ...Master Thesis -
Revealing the Recent Height Changes of the Great Altesch Glacier Using TanDEM-X DEM Series
(2023)Monitoring glacier mass balance is essential for understanding glacier-climate interactions and predicting water resources management. As the largest glacier in the Alps, the Great Aletsch Glacier has a length of 22 km and covers about of 78 km2. It contains 20% of the entire Swiss ice mass, and thus plays significent role in understanding the dynamics of glacier mass change in this region. Because of its significance, the TanDEM-X satellite ...Master Thesis -
Metering the Meter, or How to Efficiently and Deterministically Charge the Execution of Smart Contracts
(2023)Blockchain systems are rapidly evolving and, fueled by novel use cases and the competition with traditional financial systems, have to be not only secure and decentralized but also performant: processing thousands of transactions per second and having sub-second latency. Under such demanding conditions, the cost models, which are used by blockchains to mitigate attacks on the network and to align the economic incentives between users and ...Master Thesis -
Towards developing a robust and generalizable brain computer interface
(2023)Recent advances in brain computer interfaces (BCI) have been shown to enable control of external prosthesis, restore paralyzed limb movements and to improve communication. In the field of communication, electrocorticography (ECoG), has emerged as one of the most promising techniques, especially when used together with deep learning. However, among the main challenges of this method are the low signal to noise ratio and the difficulty to ...Master Thesis -
Contextual Bandit Optimization with Pre-Trained Neural Networks
(2023)Bandit optimization is a difficult problem, especially if the reward model is high-dimensional. When rewards are modeled by neural networks, sublinear regret has only been shown under strong assumptions, usually when the network is extremely wide. In this thesis, we investigate how pre-training can help us in the regime of smaller models. We consider a stochastic contextual bandit with the rewards modeled by a multi-layer neural network. ...Master Thesis