Projects
Project ideas for the course. The project is worth 60% of the grade.
- Implement Bayesian unlearning
- Bayesian neural network pruning: study the accuracy and calibration of pruned BNN models
- Implement Stein variational inference in JAX
- Automatic kernel combinations for GPs
- Deep GPs for energy disaggregation
- Deep GPs for air quality
- Bayesian neural networks for air quality
- Active learning for energy disaggregation
- Bayesian air-quality forecasting
- Implement techniques from the Laplace Redux paper in BiJAX
- How good is the automatic differentiation Laplace approximation (the Laplace approximation with a change of variables)?
- How good is a low-rank VI approximation? Check on standard UCI datasets using BiJAX or your own code.
- Conditional neural processes for energy disaggregation
- Conditional neural processes for air quality