Project Summary
CELEST conducts basic and translational research on the neuroscience of learning – from bench to application through models – in order to develop brain-machine interfaces and computing applications that can think and act autonomously. Our aim is to develop a new generation of experimentally-tested models that explain how brains and machines can learn to plan, explore, communicate, and remember and to rapidly transition insights from these models to technological applications such as neural prostheses or neuromorphic computers. Learning comes in many forms, with different temporal scales and physical constraints, ranging from the molecular to the behavioral levels. CELEST focuses on problems that are defined at a sufficiently integrated level to promote novel behavioral challenges, but that are circumscribed enough to allow progress in a few years.
Intellectual Merit
Our research in CELEST relies on a synergy between experimentation and computational modeling or between modeling and technological applications. Rather than organizing scientific endeavors according to traditional lines that treat the brain as a set of distinct modules, we organize research according to a set of functional goals that underlie learning in a dynamic and complex world. First we identify four fundamental “organizing functions” (planning, exploring, communicating, and remembering), each of which engages multiple brain regions in tightly coordinated activity. Within this framework we investigate how learning in humans and animals is affected by the “cross-function themes” of processing bottlenecks, dynamic coding, functional connections, and neural plasticity in the brain. Interaction among researchers focused on these themes bridges conceptual barriers and seeds new multi-investigator collaborations not possible outside the center structure of CELEST. Technology and educational efforts are tightly linked to the four organizing functions.
Broader Impacts
Basic neuroscience findings within CELEST are informing new technological applications designed by teams that include graduate students and postdoctoral fellows who cross-train in experimental, modeling, and translational research methods. Pre-doctoral and post-doctoral researchers are already adapting CELEST models for implementation in a new generation of computer chips being developed in the DARPA SyNAPSE program along with industrial partners Hewlett-Packard and HRL Labs. These chips will allow the neural models of learning that CELEST is developing to run in real time on low-power mobile hardware, while having as many artificial neurons and synapses as small mammals have real ones. Finding common neural mechanisms involved in planning, exploring, communicating, and remembering will yield fundamental advances in the neuroscience of learning and help us to emulate (via new computing machines) and enhance (via neural prostheses) those functions for society’s benefit. The CELEST Catalyst is working to facilitate collaborations between academic research and industry partners. CELEST also transfers the results of basic research on learning for instruction of undergraduate and graduate students through the development of courses and materials for the new undergraduate neuroscience major at Boston University, through electronic dissemination on the CELEST web site, and through multi-day workshops on the computational neuroscience of learning. A number of CELEST programs are geared to increase opportunities for groups underrepresented in science through participation in its innovative curriculum and research initiatives. These educational and diversity efforts are supported through graduate fellowships, summer internships for faculty from minority-serving institutions, and a ten-week summer program for undergraduates from underrepresented groups to work in CELEST faculty laboratories and be introduced to the interplay of modeling, experimental, and technological aspects of neuroscience.