You are here

J Model: A Simple and Runtime-Adaptive Ensemble Architecture for Classification

SSP Seminar Series
Speaker: 
Jinhan Kim

Many ensemble methods have been applied to classification. Those have been used sophisticated and static approaches to construct an ensemble. In this paper, we suggest a new architecture that is simple and runtime-adaptive called J Model. It is based on the diversity of learning agents and their peer ranked voting. We are able to generate ensemble models that can be adjusted easily and give a good performance for the purposes of classification tasks. We present the details of J Model and empirical studies over the UCI benchmark datasets, and show that the proposed ensemble architecture is effective for classification.

Date and time: 
Tuesday, 18 November, 2008 - 15:00
Length: 
60 minutes
Location: 
AT701