Đang chuẩn bị liên kết để tải về tài liệu:
A new approach for continous learning

This paper presents a new method for continous learning based on data transformation. The proposed approach is applicable where individual training datasets are separated and not sharable. This approach includes a long short term memory network combined with a pooling process. | Nguyễn Đình Hóa A NEW APPROACH FOR CONTINOUS LEARNING Nguyễn Đình Hóa Khoa Công nghệ thông tin 1 Học viện Công nghệ Bưu chính Viễn thông Abstract This paper presents a new method for continous learning based on data transformation. The proposed approach is applicable where individual training datasets are separated and not sharable. This approach includes a long short term memory network combined with a pooling process. The data must be transformed to a new feature space such that it cannot be converted back to the originals while it can still keep the same prediction performance. In _this method it is assumed that label data is sharable. The method is evaluated based on real data on permeability prediction. The experimental results show that this approach is sufficient for continous learning that is useful for combining the knowledge from different data sources. Key words knowledge combination data transformation continous learning neural network estimation. I. INTRODUCTION Permeability 1 is an important reservoir property that represents the capacity to transmit gas and fluids and plays an important role in oil well investigation. This property cannot be measured with conventional loggings but only can be achieved through SCAL in cored intervals. The conventional workflow is trying to get porosity and cored permeability relationship in cored section then applying the empirical function to the estimate permeability log. However in most cases porosity and permeability relationship cannot be described in a single empirical function and machine learning approaches such as Neural Networks are proven for better permeability prediction. In machine learning theory larger size of training data is promising to provide better estimation models. However companies cannot share their SCAL data to others. An efficient approach must be introduced to combine the knowledge from different core dataset for permeability prediction without sharing its local original dataset. Online .