We are uniquely positioned to develop next-generation artificial intelligence systems.
We are uncovering the fundamental computational dynamics and physics that underlie neural computation in the biological brain. We leverage that knowledge to create algorithms, software, and hardware for next-generation artificial intelligence systems that can be used to solve our industry partners’ tough problems.
Our success, so far, has come from doing things differently. We work in areas not traditionally associated with artificial-intelligence research, including pure mathematics, cognitive neuroscience, and experimental neurobiology. This expertise complements our deep roots in the established artificial-intelligence fields of computer science, computer engineering, electrical engineering, computational neuroscience, neural engineering, physics, and applied mathematics.
With this wide, and unique, breadth of expertise, we build the interdisciplinary teams that generate new insights on how the biological brain works and then use those insights to solve hard challenges necessary to advance next-generation artificial intelligence, computer learning and computation.
The new systems we are building depart from most of today’s artificial intelligence methods, which are limited by a set of fundamental engineering challenges inherent to statistical learning, including the need for very large training sets, huge computational and energy resources, and an almost complete lack of robustness and ability to adapt beyond the training sets. These existing artificial neural networks represent 1950’s neuroscience combined with present day computing power.
Instead of relying on 1950 understanding, we recognize the distinct advantages of designing algorithms derived from the most updated insights of how the biological brain learns and manipulates data and information. We are harnessing the computational flexibility, adaptation, and robustness of the biological brain – qualities that are beyond the capabilities of today’s artificial neural networks and machine learning techniques.
Our approach allows us to make observations and measurements on the real biological brain, and to use this information to guide our thinking and algorithm development. Such experimental and empirical neurobiological data motivates and informs our development of mathematical theory and its associated algorithms and software.