Guido Zarella

MITRE Corporation

Abstract

Recurrent neural networks are effective tools for processing natural language, and can be trained to perform sequence processing tasks such as translation, classification, language modeling, and paraphrase detection. But despite major gains in the training and application of artificial neural networks, it remains difficult to construct biologically-inspired models of cognition and language understanding. This talk will discuss recent work to bridge the gap between these fields. We will show how deep neural networks are being used to solve language understanding tasks, and demonstrate that many of these networks can be adapted to run on ultra-low power neuromorphic hardware which simulates the spiking of individual neurons. The resulting proof-of-concept, developed in collaboration at the 2015 Telluride Neuromorphic Engineering Workshop, is an interactive embedded system that uses recurrent neural networks to process language while consuming an estimated .00005 watts.

Biography

Guido Zarrella is a Principal Artificial Intelligence Engineer at the MITRE Corporation in Denver, Colorado.  He leads an R&D team pursuing advances in deep learning for language understanding.  He first began building automatic language learning systems at Carnegie Mellon University for his undergraduate research advisor Herbert A. Simon.  His work today still focuses on unsupervised learning of meaning and intent in informal language.