基於高斯過程的梯度提升分類器

Abstract

高斯過程(Gaussian process)主要被用在解決迴歸(regression)與分類(classification)問題上，在眾多機器學習方法之中預測表現良好，且是非常彈性的非參數(nonparametric)模型。但其擁有一些明顯的缺點，如時間複雜度高，導致不容易應用在資料量較多的問題上。為了解決這個缺點，本次研究將結合高斯過程(Gaussian process)與梯度提升方法(gradient boosting)應用在分類問題上。根據實驗結果顯示，與原本的高斯過程相比，此結合後產生的演算法可大幅增加模型訓練速度，使模型可以在訓練中使用更多的資料，且可在某些情況下增加模型的表現，此外也會探討此方法與其他分類器在不同資料集與不同訓練資料量上之差別。

Gaussian process (GP) is mainly used to solve regression and classification problems in machine learning. GP is a nonparametric model with good prediction performance and wide applications. However, GP has an obvious drawback: high time complexity. This drawback makes it inappropriate for large data. In this work, we combine Gaussian process and gradient boosting to form Gradient Boosting Gaussian Process Classifier (GBGPC), then apply it to classification problems. The experiment results show that the proposed algorithm can largely improve training efficiency and performance in some situation compare to the standard Gaussian process. We also discuss the difference of performance between GBGPC and other well-known classifiers among different datasets and training data sizes.