基於高斯過程迴歸的梯度提升演算法

Abstract

高斯過程迴歸 (Gaussian Process Regression) 是機器學習中的一種方法，此方法具有良好的預測結果、且容易實作，但在運算時時間及空間的複雜度高，使得此方法難以被實際運用在大量資料集上。本研究提供一個基於梯度提升演算法的估計方法，實驗結果顯示此方法能夠在訓練時使用較低的時間及記憶體成本來達到良好的估計效果。

Gaussian process regression (GPR) is an important model in the field of machine learning. GPR model is flexible, robust, and easy to implement. However, it suffers from expensive computational cost: O(n^3) for training time, O(n^2) for training memory and O(n) for testing time, where n is the number of observations in training data. In this work, we develop a fast approximation method to reduce the time and space complexity. The proposed method is related to the design of gradient boosting algorithm. We conduct experiments using real-world dataset and demonstrate that the proposed method can achieve comparable prediction performance compared to the standard GPR model and some state-of-the-art regression methods.