 |
|
|
|
|
|
|
|
|
|
|
|
|
 |
|
|||||||||||
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
 |
|
|
 |
 |
|
||||||||||
 |
|
|||||||||||
 |
|
|||||||||||
 |
 |
|
||||||||||
| Seminar Schedule (Seminar Organizer: Prof. Ziv Bar-Joseph) | |
Machine Learning Lunchtime Chats
|
Spring 2007 Seminar
Date: May 8, 2007 Time: 3:00 PM - 4:00 PM Location: 3305 Newell-Simon Hall Speaker: Dan Li Tepper PhD Student, ML MS Student Title: The Cluster-Squared Algorithm for Combining Economic Forecasts Abstract: We propose a “Cluster-Squared” algorithm to efficiently combine economic forecasts. Forecasters are first clustered into groups, which implicitly maximize in-group similarities and inter-group disagreement. This step resolves the weighting instability problem in most dynamic weighting schemes. Secondly, to account for non-stationarity in the time series, we allow the groups’ combination weights to be state-dependent. The states are uncovered using subsequence clustering. Experimental results show that the algorithm has smaller prediction error compared with median forecasts for a broad set of economic indicators in the Survey of Professional Forecasters (Federal Reserve bank of Philadelphia). Our algorithm also compare favorably against other popular dynamic weighting algorithms.