Ensemble
# Tag:
- Source/KU_ML2
Ensemble
여러 model(base-learner)들을 각각 train하여 합쳐 더 좋은 accuracy를 얻는 방법.
- Different Learning Algorithm
- Same learning Algorithm with different hyperparameters
- Different subsets of the train data: 이렇게 훈련 받은 model은 다른 model이 된다.
- Different input representations: feature 중 일부만 사용하거나, multimodal에 대해 같은 목적에 대하여 각각 넣는 방법 등이 있다.
- 음성 인식에 대해서도, 제스쳐, 입모양, 음성 등을 각각 사용해 훈련시킨 후 ensemble 할 수도 있다.
여러 model을 결합하는 방법인 Learner fusion, model 중 일부를 선택하는 Learner Selection이 있다.