文章目录
Abstract
- propose gcForest,
- a decision tree ensemble approach with performance highly competitive to deep neural networks
- contrast to dnn
- require great effort in hyper-parameter tuning,
- gcForest much easier to train;
- even when it is applied to different data across different domains in our ex-
periments, excellent performance can be achieved
by almost same settings of hyper-parameters.
- training process of gcForest is efficient,
- users can control training cost according to computational resource available.
- The efficiency may be further enhanced because gcForest is naturally apt
to parallel implementation.
- contrast to deep neural networks which require largescale training data,
- gcForest can work well even when there are only small-scale training data.
1 Introduction
DNN很牛逼
- In recent years, deep neural networks have achieved greatsuccess in various applications, particularly in tasks involv-
ing visual and speech information [Krizhenvsky et al., 2012; Hinton et al., 2012], leading to the hot wave of deep learning
[Goodfellow et al., 2016].
Though deep neural networks are powerful, they have ap-
parent deficiencies. First, it is well known that a huge amount
of training data are usually required for training, disabling
deep neural networks to be directly applied to tasks with
small-scale data. Note that even in the big data era, many
real tasks still lack sufficient amount of labeled data due to
high cost of labeling, leading to inferior performance of deep
neural networks in those tasks. Second, deep neural networks
are very complicated models and powerful computational fa-
cilities are usually required for the training process, encum-
bering individuals outside big companies to fully exploit the
learning ability. More importantly, deep neural networks are
with too many hyper-parameters, and the learning perfor-
mance depends seriously on careful tuning of them. For ex-