Stephen Grossberg is Wang Professor of Cognitive and Neural Systems; Director of the Center for Adaptive Systems; and Emeritus Professor of Mathematics and Statistics, Psychological and Brain Sciences, and Biomedical Engineering at Boston University. He is a principal founder and current research leader of the fields of computational neuroscience, theoretical psychology and cognitive science, and biologically-inspired engineering, technology, and AI. In 1957-1958, he introduced the paradigm of using systems of nonlinear differential equations to develop neural network models that link brain mechanisms to mental functions, including widely used equations for short-term memory (STM), or neuronal activation; medium-term memory (MTM), or activity-dependent habituation; and long-term memory (LTM), or neuronal learning. His work focuses upon how individuals, algorithms, or machines adapt autonomously in real-time to unexpected environmental challenges. These discoveries together provide a blueprint for developing autonomous adaptive intelligence. They includes models of vision and visual cognition; object, scene, and event learning and recognition; audition, speech, and language learning and recognition; brain development; cognitive information processing; reinforcement learning, motivation, and cognitive-emotional interactions; multiple kinds of consciousness; learning to navigate using vision and path integration; social cognition and imitation learning; sensory-motor learning, control, and planning; brain dysfunctions that cause symptoms of Alzheimer’s disease, autism, medial temporal amnesia, visual and auditory neglect, and sleep disorders; mathematical analysis of neural networks; and large-scale applications of these discoveries. Grossberg founded key infrastructure of the field of neural networks, including the International Neural Network Society and the journal Neural Networks, and has served on the editorial boards of 30 journals. His lecture series at MIT Lincoln Lab led to the national DARPA Study of Neural Networks. He is a fellow of AERA, APA, APS, IEEE, INNS, MDRS, and SEP. He has published 17 books or journal special issues, over 550 research articles, and has 7 patents. He was most recently awarded the 2015 Norman Anderson Lifetime Achievement Award of the Society of Experimental Psychologists (SEP), the 2017 Frank Rosenblatt computational neuroscience award of the Institute for Electrical and Electronics Engineers (IEEE), and the 2019 Donald O. Hebb award for his work in biological learning by the International Neural Network Society (INNS).
Keynote Title: Explainable and Reliable AI: Comparing Deep Learning with Adaptive Resonance
Abstract: This lecture compares and contrasts Deep Learning with Adaptive Resonance Theory, or ART. Deep Learning is often used to classify data. However, Deep Learning can experience catastrophic forgetting: At any stage of learning, an unpredictable part of its memory can collapse. It is thus unreliable. Even if it makes some accurate classifications, they are not explainable. It is thus untrustworthy. Deep Learning has these properties because it uses the back propagation algorithm, whose computational problems due to nonlocal weight transport during mismatch learning were described in the 1980s. Deep Learning became popular after very fast computers and huge online databases became available that enabled new applications despite these problems. ART models overcome 17 foundational computational problems of back propagation and Deep Learning. ART is a self-organizing, explainable production system that incrementally learns, using arbitrary combinations of unsupervised and supervised learning, to rapidly attend, classify, and predict objects and events in a changing world, without experiencing catastrophic forgetting. ART has also successfully explained and predicted many psychological and neurobiological data, and can be derived from a thought experiment about how any system can autonomously learn to correct predictive errors in a changing world. It hereby forms a foundation for designing algorithms for any adaptively intelligent system that is truly autonomous. ART has been successfully used in hundreds of large-scale real world applications, including remote sensing, medical database prediction, and social media data clustering.