Abstract
We consider an online decision problem over a discrete space in which the loss function is submodular. We give algorithms which are computationally efficient and are Hannan-consistent in both the full information and partial feedback settings.
| Original language | English (US) |
|---|---|
| Pages (from-to) | 2903-2922 |
| Number of pages | 20 |
| Journal | Journal of Machine Learning Research |
| Volume | 13 |
| State | Published - Oct 2012 |
| Externally published | Yes |
All Science Journal Classification (ASJC) codes
- Software
- Control and Systems Engineering
- Statistics and Probability
- Artificial Intelligence
Keywords
- Online learning
- Regret minimization
- Submodular optimization