Developing next-generation algorithms and computing substrates that leverage neurobiology to revolutionize intelligent systems
Researchers in APL’s Intelligent Systems Center (ISC) are extracting and translating design principles of neural connectivity and function from multiple species to create the next generation of robust, efficient intelligent systems operating in the real world. Jump to a section to learn more about this research.
Explore Our Work
ISC researchers believe that the wiring diagrams of the biological nervous system have a lot to teach us about building more efficient learning algorithms. Those wiring diagrams, often referred to as connectomes, are built from very detailed images captured via electron microscopy. ISC researchers are building cloud tools to store and analyze connectomic data, and searching for patterns (or motifs) in the brain’s network structure. They have used connectomic data to improve the efficiency of transformer model architectures and to implement learning mechanisms from the fly.
Building on the ISC’s expertise in neuroscience and neuroengineering, researchers at APL are working to incorporate neurobiological inspiration into artificial neural networks. In doing so, our team is learning how to create more efficient algorithms and how neurobiological processes lead to brain function. One recent effort involved implementing artificial sleep to mitigate catastrophic forgetting, where networks forget the earliest information they were trained on as new data is introduced. APL has also developed evolutionary algorithms as an alternative to training through backpropagation, incorporating biological principles of connectivity and hardware constraints. Additionally, ISC researchers have investigated processing of dynamic signals with novel reservoir topologies and developed new tools to investigate learned representations in transformers.
APL researchers are investigating the computational capabilities of biological neural networks in addition to biologically inspired computing substrates. In partnership with the Johns Hopkins Bloomberg School of Public Health, our team is investigating whether reinforcement learning algorithms can be implemented in brain organoid systems. Organoid-based intelligence would potentially be more efficient than current silicon-based computing, and could serve as a model of biological intelligence to permit more in-depth study. We are also building implementations of artificial neural networks on chips that spike like biological neurons, and investigating novel uses of field-programmable gate arrays to mimic the efficient processing of the biological nervous system.
