Daniel George is a PhD student in astronomy, with a fellowship in computational science and engineering, at the University of Illinois at Urbana-Champaign. He obtained his bachelor’s degree in engineering physics from IIT Bombay. He is currently a research assistant in the Gravity Group at the National Center for Supercomputing Applications (NCSA), a member of the LIGO and Dark Energy Survey collaborations and an LSST Data Science Fellow, working at the interface of deep learning, high-performance computing and gravitational wave and multi-messenger astrophysics. His long-term interests lie in applying cutting-edge computer science and technology, especially artificial intelligence, to accelerate discoveries in the fundamental sciences.
Introduction to Transfer Learning »
Project: Neural Wolfram|Alpha
Goal of the project:
Generate Wolfram Language code given natural language inputs using deep learning with artificial neural networks.
Summary of work:
Trained bi-directional LSTM recurrent neural networks, with and without attention, to generate code given English queries about indefinite integrals. Tried both character-to-character and character-to-word models.
Results and future work:
Obtained excellent performance beyond our expectations. The neural net learned to always generate valid Wolfram Language syntax (even character by character). We will train on more types of queries in the future.