Ultimately, CELEST seeks to advance technologies that interact with humans in real time. Without real-time capabilities, most intelligent learning technologies would be useless and hazardous. For example, it would be difficult for an impaired patient to use a motor prosthesis if there is a lag of even a few fractions of a second between command and action, producing an unstable control system. As more unmanned aerial vehicles (UAVs) are developed, even brief delays in processing could result in mid-air collisions and crashes. The design of tomorrow’s most useful neural prosthesis or assistive neuromorphic agents will be more robust and resilient if their designs are informed by a rigorous understanding of anatomical and physiological substrates of actual human or animal learning behavior. It is thus critical for CELEST to understand both the fundamental spatial and temporal constraints of learning in brains, whether for the sake of furthering our fundamental understanding of learning, emulating the performance of human brains, or enhancing technological designs for devices that complement and extend the abilities of human operators.
A whole brain system does not learn in isolation in space or in time, but instead integrates information over a range of spatial and temporal scales. The brain contains billions of neurons and trillions of synapses, all of which are highly interconnected and organized to efficiently transport information, overcome communication bottlenecks, and support real-time adaptive decision-making and behavior. These brain processes evolved to interact with a dynamically changing environment and learn via plasticity mechanisms from the scale of the single neuron up to the whole-brain system. The interplay of these different mechanisms, operating across this broad range of organizational scales, is what enables a rich repertoire of adaptive behaviors. CELEST identified four strategic initiatives to organize its scientific research, two that focus on spatial factors (Functional Connections and Processing Bottlenecks) and two on temporal factors (Dynamic Coding and Neural Plasticity) that enable successful learning processes in whole brain systems. In Figure 1, the third level from the top of the pyramid depicts these four initiatives, which in turn feed the Capstone Projects at the next level of the pyramid.
By bridging research across the four scientific initiatives, CELEST researchers can tackle complex problems that would be impossible without center-level support; it is the existence of the center that allows integration of research across the four initiatives to address questions that could not be attacked by considering any one of these mechanisms in isolation. For example, one cannot properly understand dynamic coding that spans several brain areas without understanding the functional connections between those areas. Similar statements tie together all the initiatives. Together, the four initiatives support the larger purpose of understanding the neural basis of learning at a sufficiently mechanistic level to support the development of two Capstone Projects.
CELEST’s scientific research initiatives have evolved over the years as the center’s research has registered successes, confronted and overcome obstacles, and adapted to breakthroughs generated within CELEST and in the broader scientific community. The research initiatives in CELEST’s first years had names of learning modalities such as “vision” or “audition” or “memory.” As CELEST evolved, it became increasingly clear that in order to fulfill its mission to understand real-time autonomous systems, as initially stated, CELEST first had develop a functional understanding of how the brain learns as a whole system. This realization stood in contrast to the approach CELEST had adopted in its first five years, which was based on classical learning functions examined within traditional disciplinary enclaves.
During Year 5, CELEST began to transition away from an organization based on learning modalities (often focusing on individual brain regions). Instead, CELEST began reorganized its research around specific, mechanistic substrates of learning that were common across all of the traditional fields that defined CELEST research in its first five years. This shift enabled CELEST researchers to address fundamental research questions at a systems level. These research questions probe the fundamental limits of both spatial and temporal parameters of learning at the whole brain level. Going into Year 9, the focus of CELEST has transcended classic categories of neuroscience that rely on piecemeal accumulation of data within isolated brain areas. Instead researchers within CELEST work together to ask larger questions about learning within the framework of two Capstone Projects.