Here are collected some possible project ideas and associated papers. You do not need to choose something on this list; it is just a list of possibilities to get you started. You may structure your project exploration around a general problem type, algorithm, or data set, but should explore around your problem, testing thoroughly or comparing to alternatives.
You should submit a project proposal (1 page description of your group and proposed project concept) by Friday 4/22.
You should turn in a write-up (< 10 pages) describing your project and its outcomes, similar to a research-level publication (journal or CS conference); I suggest the latex styles for NIPS or UAI. This is due by 5pm Friday 6/3, to EEE.
Explore a particular inference algorithm's performance, either for inference itself or when used as part of learning.
- Variational techniques
- Monte Carlo techniques
One project route is to take an interesting high-dimensional dataset and build a graphical model for it. Some possible examples:
- Spatio-temporal traffic counts or flow
- Text & language data
- Protein sequence & structure data
- Image data (segmentation, recognition, etc.)
- Sequence prediction challenge
Another route is to explore a particular technique for building models that we have not explored in depth.
Gaussian and Gaussian-like models:
- Covariance selection in Gaussian models (learning sparse inverse covariance structures)
- Gaussian copula models (Gaussian covariance with non-Gaussian marginal distributions)
- Stable distribution models
- CDF models (model a cumulative distribution rather than a PDF)
- Conditional random fields
- Structured prediction / Max-margin networks
- Deep belief networks
- Contrastive divergence
- Structural EM
- Thin junction trees (structure learning with bounded tree-width)
- Independence testing based structure estimation
Latent variable models
More ideas from other schools: