Student Projects in LAS Group

We offer Semester Projects, Bachelor’s and Master’s theses in our group. Depending on your preference, there are opportunities for working on theory, methods, and applications. M.Sc. projects at LAS often result in publications at leading conferences.
A list of topics for which we are actively recruiting students is given below. If you don’t see a project below that fits well, but you are interested in the kind of research our lab does, feel free to still reach out. To learn more about the research done in the group, visit our recent publications. You can also learn more about research by individual group members.
If you are interested in working with us, you can send an application by clicking the button below. Make sure that your email includes: a résumé, a recent transcript of records, your intended start date, and we highly recommend that you mention the projects you are interested in, members of the group with whom you would like to work, or recent publications by the group relevant to your interests.

If you are a bachelors or masters student but not a student at ETH, please see the opportunity listed in Applications for Summer Research Fellowships.

Current Topics

The detailed project proposals can be downloaded only from the ETH domain.

Active Causal Discovery

Design and analyse objectives for active causal structure discovery using recent techniques in the uncertainty estimation of challenging probabilistic models.

Resources


Keywords: causality, sequential decision making, uncertainty quantification

Learning to Schedule Energy Generation

Use machine learning to solve mixed integer programs for energy scheduling faster, in collaboration with Hitachi Energy.

Resources


Keywords: applied, discrete optimization, integer programming, graph neural networks

Local Causal Models in Model-Free Reinforcement Learning

Explore how recent techniques using local casual models in RL can be applied to policy gradient methods.

Resources

People


Keywords: reinforcement learning, causality

Experimental Design: Learning to Learn

Find a near-optimal policy that can learn on simple problems where the optimal policy is unintuitive.

Resources


Keywords: active learning, meta-learning

Maching Learning for Protein Design

Learn to find the right mutation that leads to the right enzyme.

Resources


Keywords: applied, sequence-data

Applications of Machine Learning for Choosing Crop Varieties

Learning crop variety selection and management policies from data.

Resources


Keywords: applied, uncertainty quantification, active learning, reinforcement learning, remote sensing

Stabilizing Recurrent Neural Networks

Preventing exploding gradients in RNNs on long time-series data.

Resources


Keywords: differential equations, vanishing/exploding gradients, time series

Meta-Learning Function Priors for Bayesian Optimization

Accelerate Bayesian optimization by meta-learning priors.

Resources


Keywords: meta-learning, active learning

Machine Learning for Converter Control

Algorithms for control of power electronics converters, in collaboration with Hitachi Energy.

Resources

People


Keywords: applied, reinforcement learning, control

Safe Bayesian Optimization for Quadruped Locomotion Control

Sample-efficient reinforcement learning for safely tuning control parameters of a real robot.

Resources


Keywords: bayesian optimization, reinforcement learning, robotic control  

Urban Vehicle Trajectory Learning with Transformer

Short-term trajectory prediction of multi-vehicle trajectories in complex urban driving contexts.

Resources


Keywords: sequential learning, Transformer, high-resolution vehicle trajectory  

General Areas

We offer projects in several general areas.
  • Probabilistic Approaches (Gaussian processes, Bayesian Deep Learning)
  • Discrete Optimization in ML
  • Online learning
  • Large-Scale Machine Learning
  • Causality
  • Active Learning
  • Bayesian Optimization
  • Reinforcement Learning
  • Meta Learning
  • Learning Theory

Examples of Previous Master Theses

Near-Optimal Multi-Perturbation Experimental Design for Causal Structure Learning
Scott Sussex with Andreas Krause and Caroline Uhler. NeurIPS 2021. [paper] [blog]
Neural Contextual Bandits without Regret
Parnian Kassraie with Andreas Krause. AISTATS 2022. [paper]
DiBS: Differentiable Bayesian Structure Learning
Lars Lorch with Jonas Rothfuss. NeurIPS 2021. [paper] [blog]
PopSkipJump: Decision-Based Attack for Probabilistic Classifiers
Noman Ahmed Sheikh with Carl-Johann Simon-Gabriel. ICML 2021. [paper]

Icons on this page are by www.flaticon.com.