ABSTRACT
This chapter describes some of the most important architectures and algorithms for committee machines. It discusses three reasons for using committee machines. The first is that a committee can achieve a test set performance unobtainable by a single committee member. Second, with committee machines, one obtains modular solutions, which is advantageous in many applications. The third reason for using committee machines is a reduction in computational complexity. The interest of the machine learning community in committee machines began around the middle of the 1990s, and this field of research are very active. The basic idea is to train a committee of estimators and combine the individual predictions with the goal of achieving improved generalization performance as compared to the performance achievable with a single estimator. The generalization performance for the various committee machines is certainly impressive. Committee machines improve performance when the individual members have low bias and are decorrelated.
