1 / 11

Relational Learning, and Structured Output

Relational Learning, and Structured Output. Lei Tang May.04, 2007. Relational Learning. Typical Classification task: IID assumption Relational Learning: instances are interrelated. Some Examples: Hypertext Classification Identifying topics in scientific literature.

trista
Download Presentation

Relational Learning, and Structured Output

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. Relational Learning, and Structured Output Lei Tang May.04, 2007

  2. Relational Learning • Typical Classification task: IID assumption • Relational Learning: instances are interrelated. • Some Examples: • Hypertext Classification • Identifying topics in scientific literature. • Prediction of Movie Box Sales

  3. Strategies for relational learning • How to use the link information? • Make it flat, several models (Jens-etal04, kdd)

  4. Ensemble Method • Consider relational learning as datasets with heterogonous attributes. • Another strategy is to build classifier on each type of attribute and do the ensemble learning.

  5. Iterative Strategy for Prediction • For prediction, the class labels are unknown for the neighbors of the test data. • So, iteratively predict until converge. (Nevi-Jens00, AAAI; Li-Geto03, ICML) • This like label propagation in semi-supervised learning. • Baseline models: • Use only attributes • Use only link information (This actually works pretty well on most benchmark datasets used in literature)

  6. Interdependent & Structured Output • The output space is not a “flat” list of labels, but some structure. • E.g. Multi-label/class classification, Hierarchical Classification

  7. Collective Multi-label Classification • Takes into consideration of interrelations between labels rather than instances.

  8. Large-Margin Methods • Use Joint Feature Map.

  9. Discussions • Interdependent features (graphical model) • Interdependent instances (not well studied yet) • Interdependent classes (The future)

More Related