EAPSI: Generating Word Embeddings using Extreme Learning Machines for Classifying Clinical Texts

Lauren Paula, Royal Oak MI

Investigators

Abstract

In 1950, the computer scientist Alan Turing proposed a test for true artificial intelligence. In Turing's view, a computer must be considered intelligent if it could understand human language. After more than 60 years of research, this is still an ongoing effort. Recent methods, including neural language models that use advanced statistics, have made great strides towards realizing Turing's vision. This study builds on existing research to explore methods for improving computer-based natural language understanding. The research will be conducted under the mentorship of Professor Guang-bin Huang, a noted expert on machine learning, of Nanyang Technological University. Natural language processing (NLP) involves the development of computer-based algorithms to understand natural language. Statistical language models are typically used for various NLP tasks, including machine translation and text categorization. Language models based on neural networks, also known as neural embeddings, map words (or phrases) to a numerical representation in a low-dimensional space. Typically, neural networks use back-propagation for training a neural network, which results in slow training. Extreme Learning Machines (ELM) is a type of neural network, where hidden neurons are randomly generated hidden nodes. This study involves the use of ELM for faster training in the generation of neural embeddings. This award under the East Asia and Pacific Summer Institutes program supports summer research by a U.S. graduate student and is jointly funded by NSF and the National Research Foundation of Singapore.