Study and Understanding the Significance of Multilayer-ELM Feature Space

摘要

Multi-layer Extreme Learning Machine (Multi-layer ELM) is one of the most popular deep learning classifiers among other traditional classifiers because of its good characteristics such as being able to manage a huge volume of data, no backpropagation, faster learning speed, maximum level of data abstraction etc. Another distinct feature of Multi-layer ELM is that it can be able to make the input features linearly separable by mapping them non-linearly to an extended feature space. This architecture shows acceptable performance as compared to other deep networks. The paper studies the high dimensional feature space of Multi-layer ELM named as MLELM-HDFS in detail by performing different conventional unsupervised and semi-supervised clustering techniques on it using text data and comparing it with the traditional TF-IDF vector space named as TFIDF- VS in order to show its importance. Results on both unsupervised and semi-supervised clustering techniques show that MLELM-HDFS is more promising than the TFIDF- VS.