Description

Achieving high performance at small size and high energy efficiency is a central challenge of computer engineering. Neuromorphic computing has shown great potential to use insights from spiking neural networks to reduce the power footprint of computing systems; however, such systems are still eclipsed by the power efficiency and computational capacity of the miniaturized animal brain. In particular, animals achieve remarkable feats of learning and accuracy from large to small brain sizes at energy costs very low relative to human-engineered systems. Even in the taxonomically diverse group of tiny-brained stingless bees, brains miniaturize from 4 to 0.2 mg without measurable loss of sensory and learning capacities. However, little is known about the mechanisms that facilitate this miniaturization. This project will study several species of stingless bees to uncover rules of brain miniaturization and incorporate these rules into novel modes of neuromorphic computing. Methods from neuroanatomy and physiology will be used to uncover novel topologies as well as modulation of spiking patterns that may improve performance of artificial spiking neural networks. These studies will provide the first measures of how neuron size, synapse density, energy capacity, and brain structure change during miniaturization with demonstrated preservation of cognitive function. Because tiny brain sizes have evolved several times independently in this group, the project will determine whether some fundamental mechanism of miniaturization and energetic optimization exists or if there are multiple solutions. Computational models will incorporate these findings to link ion channel, cellular, and brain system function, and new principles will be used to reducing the energy cost of neuromorphic computing.
Integrative Value and Transformative Potential: This project integrates across neuroscience, physiology, computer science, electrical engineering, and operations research and is transformative as opposed to merely translational. Along with uncovering new neuroscientific insights about the miniaturized brain, the taxonomic comparison will lead to insights about the natural evolution of brain architectures. The projects improved computational models of spiking neural networks may also lead to new tools for understanding cognitive systems outside of the studys model organisms. Novel brain-inspired computing architectures from this project will aid in the advancement of artificial cognitive systems that come closer to natural systems in terms of performance. The questions motivated by engineering will lead to new lines of scientific inquiry just as those scientific results will generate novel technologies. This project is only possible with interdisciplinary collaboration, and it will contribute to a further expansion of the unification of animal cognitive science and artificial intelligence.
This project will uncover new mechanisms responsible for brain performance in insect systems. By studying the differences in these mechanisms across diverse, but still closely related, taxa, it will also reveal new insights into the evolution of those neural mechanisms. The focus on the energetic consequences of neuroanatomy will assist in achieving high energy efficiency in artificially intelligent systems, which currently require tremendous amounts of power to operate.
StatusActive
Effective start/end date3/19/199/18/20

Funding

  • DOD: Defense Advanced Research Projects Agency (DARPA): $1,000,000.00

Fingerprint

Brain
Cognitive systems
Animals
Physiology
Neural networks
Energy efficiency
Operations research
Engineering research
Electrical engineering
Intelligent systems
Computer science
Neurons
Artificial intelligence
Costs
Topology
Modulation
Ions