Theoretical neuroscience research at Sandia focuses primarily on developing theories for different biological neural circuits and understanding basic neural computation concepts that may eventually impact neural machine learning.
Formal characterization of neural computation
While much of the attention on neural machine learning today focuses on deep artificial neural networks, such as deep convolutional networks and deep belief networks, we are looking towards understanding fundamental features of neural systems that will underlie the next generation of brain-like algorithms. In particular, we are focusing on formally developing novel algorithms leveraging neural properties including
- Spike-based representions and communication
- Dynamical network-based computation
- Synaptic and neuronal learning
Understanding computational primatives of key neural regions
In addition to characterizing the potential of neural circuits broadly, we also are focusing on theoretically characterizing specific brain regions which are well suited for formal analysis, again with the of how these insights may influence either neural computing or broader neuroscience understanding.
Much of our theory effort focuses on the hippocampus, a region of the brain that has long attracted the attention of theoretical neuroscientists. While generally thought to be critical for declarative and episodic memory formation (i.e., memories of facts, events, etc) and presumed to be important for spatial processing, the underlying computational mechanisms are still widely debated.
Notably, our theoretical work has focused significantly on the dentate gyrus (red in figure), a region that is unique among brain regions in that it incorporates new neurons throughout life (see sidebar). Ongoing theory efforts are also considering the downstream hippocampal regions CA3 (blue) and CA1 (brown), which are experimentally well studied regions against which extensive place cell and other in vivo data can be used to validate theories.
Counter to long-held beliefs about the brain, some neural circuits actually produce new neurons throughout life. This process, known as adult neurogenesis, is limited to a handful of regions such as the dentate gyrus region of the hippocampus. Hippocampal neurogenesis is particularly intriguing, because it is thought to potentially play a role in the formation of episodic memories – those memories of specific events in our lives.
Sandia has a number of research efforts related to exploring the implications of continual neurogenesis on neural computation, including the implementation of human-scale dentate gyrus simulations investigating new neurons, theoretical studies of how new neurons impact the neural code, and incorporation of new neurons into deep learning models.
For more information, see Draelos et al., IJCNN 2017
Severa W, Parekh O, James CD, and Aimone JB– “A Combinatorial Model for Dentate Gyrus Sparse Coding” – Neural Computation 2017
Aimone JB – “Computational Modeling of Adult Neurogenesis.” Adult Neurogenesis Cold Spring Harbor Perspectives, Cold Spring Harbor, NY, 2016.
Rangel LM, Quinn L, Chiba AA, Gage FH and Aimone JB –“A Hypothesis for Temporal Coding of Young and Mature Granule Cells” Frontiers in Neurogenesis. 7(75),May 2013