DL-DI: A Deep Learning Framework for Distributed, Incremental Image Classification
Metadata[+] Show full item record
Deep Learning technologies show promise for dramatic advances in fields such as image classification and speech recognition. Deep Learning (DL) is a class of Machine Learning algorithms that involves learning of multiple levels of features from data to build a model. One of the open questions in DL is whether up-to-date models can be built and provide dealing with dynamic and large volumes of new data created. This requires addressing how models can be consistently constructed and updated (incremental learning) in a scalable manner. Current research and practices of DL do not fully support these important features, such as distributed learning or incremental learning to an extent that is required. The objective of this thesis is to provide a solution to this problem by building a framework that is distributed and incremental in nature. In the DL-DI framework, a learning problem is composed of two stages: Local Learning and Global Learning. In the local learning stage, a learning problem is divided into several smaller problems. These smaller problems are solved using an optimized original solution for a better local performance. The learning outcomes from the local learning stage, such as predictions and activations, will feed into the global learning. A feed forward deep neural network is used in global learning. The presented framework focuses mainly on image classification problems, but this can be applied to several other learning problems. The proposed framework is implemented in TensorFlow, an open source machine learning library developed by Google, with the capability of building deep neural networks using parallel GPU computations. To support the effectiveness of the DL-DI framework, we have evaluated the DL-DI framework on image classification using Softmax Regression and Convolutional Neural Networks on MNIST, CIFAR10 datasets. The evaluation results have verified that the DL-DIS framework supports distributed incremental Deep Learning while achieving a reasonably high rate of prediction accuracy.
Table of Contents
Introduction -- Background and related work -- The DL-DI framework -- Results and evaluation -- Conclusion and future work