Adventures in Hybrid Architectures for Intelligent Systems
We are living in a world ripe with data and computing power and so it is no surprise that the current generation of AI systems are leveraging data-driven approaches to yield often surprising advances in such problems as perceptual processing and language understanding. These systems, however, often suffer from key drawbacks, including data inefficiency and lack of interpretability. In this talk I will present diverse examples of general architectures, spanning distributed optimization and reinforcement learning, that leverage a common approach: integrating higher-order knowledge to enhance data-driven processing. Together, this work highlights how bridging the data-knowledge dichotomy may lead to exciting new applications for intelligent systems.
Nate Derbinsky is an Associate Teaching Professor, as well as the Director of Teaching Faculty, in the Khoury College of Computer Sciences at Northeastern University. His research focuses on artificial intelligence, drawing inspiration and techniques from cognitive science, database systems, and optimization. Nate earned his PhD in Computer Science and Engineering from the University of Michigan and was a Postdoctoral Associate at Disney Research. In parallel with undergraduate and graduate work, he founded BitX Solutions, a small corporation based in North Carolina; for 11 years, the firm worked on such diverse project as tracking evidence for law-enforcement agencies; managing supplies at food banks; providing diagnostic training for neurology interns; and supplying comprehensive web-based fellowship/scholarship advising for thousands of students and faculty members at two universities.