[Monday, January 24, 2022 at 12:00 PM in Zoom Meeting ID: 945 0413 9035] Abstract: Deep learning has been tremendously successful, especially for solving problems in natural language understanding and computer vision. This is a great achievement, but artificial intelligence (AI) still has a long way to go toward achieving the versatility of humans. My lab works toward overcoming the limitations of todays AI systems.
In this talk, I review my labs work on language guided scene understanding. Second, I discuss my labs work on lifelong continual machine learning. In continual learning, an AI system is progressively updated from streams of data. In contrast, conventional deep neural networks operate in a train and then evaluate paradigm. Unlike humans, conventional networks cannot easily continue to learn more information. If this is attempted, then conventional models will suffer from catastrophic forgetting, resulting in the loss of previously acquired skills. I describe deep neural network systems for online continual learning without catastrophic forgetting. These models use brain-inspired replay mechanisms to avoid catastrophic forgetting, and they achieve state-of-the-art results on image and multi-modal classification datasets. I also discuss my work at Paige, where I have been leading AI research to create deep learning systems to improve cancer detection and treatment.
Biography: Christopher Kanan is a tenured Associate Professor at the Rochester Institute of Technology (RIT), a Visiting Associate Professor at Cornell Tech, and the Director of AI Science at Paige. Prof. Kanan is an NSF CAREER Award winner and an IEEE Senior Member. He has published over 60 refereed papers, many of which are in top venues across AI such as CVPR, ICCV, NeurIPS, AAAI, ICLR, ACL, ICRA, etc. He received a PhD in computer science from the University of California at San Diego, where he worked on brain-inspired algorithms for object recognition, neural networks, active vision, and cognitive modeling. He received an MS in computer science from the University of Southern California. Before joining RIT, he was a postdoctoral scholar at the California Institute of Technology, and later worked as a Research Technologist at NASAs Jet Propulsion Laboratory, where he used deep learning to develop vision systems for autonomous ships.
At RIT, his lab works on lifelong machine learning and language guided computer vision, which has been supported by awards from NSF, AFOSR, ONR, DARPA, and more. He is the co-founder and Associate Director of RIT's Center for Human-aware AI, a university designated Center of Excellence. He is a member of RIT's McNair Scholars advisory board, which aims to increase the number of underrepresented minorities in doctoral programs. He is the recipient of the 2016 Rising Star Award and the 2019 Distinguished Scholarship Award in RIT's College of Science. At Paige, a startup that has raised $220M to improve the diagnosis of cancer, Dr. Kanan has led the AI R&D team for over three years. He joined Paige as one of the first employees, and has seen the company grow to nearly 200 team members. His work led to the first FDA cleared AI system for helping pathologists diagnose cancer. His work at Paige resulted in numerous patent filings, with eight granted patents so far.
January 31, 2022, 12:00 PM
William Schuler: Is syntactic composition a real part of language?
[Monday, January 31, 2022 at 12:00 PM in Zoom Meeting ID: 992 5113 7788] Abstract: People can express novel, precise complex ideas: plans with sophisticated contingencies, predictive models of interrelated uncertain events, and more. This seems to suggest a formal, compositional semantics in which sentences are divided into syntactic categories with associated semantic functions. But state-of-the-art NLP systems -- transformers like BERT and GPT-3 -- dont work like that.
This talk will review evidence about syntactic composition from sentence processing experiments and grammar induction simulations conducted over the past few years in the OSU computational cognitive modeling lab, and make some educated guesses about the cognitive status of syntactic composition.
Bio: William Schuler is a professor at the Ohio State University, specializing in computational psycholinguistics and computational cognitive modeling. His research is funded by the NSF, NIH, and DARPA, and is primarily focused on building human-like sentence processing models both for spoken language interfaces and to evaluate theories of sentence processing in human working memory. He is a PECASE awardee, WIF fellow and author of nearly 100 research papers, including a COLING best-paper award winner.