A Parallel Programming Model of Local Processing Units in the Fruit Fly Brain
-
1
Columbia University, United States
The fruit fly’s brain can be subdivided into some 41 neural circuit modules called Local Processing Units (LPUs) [1]. Extensive local processing is achieved by spatially restricted local neurons in the LPUs. Yet little is known about the I/O transformations taking place in LPUs. To identify these transformations, it is necessary to determine the algorithms describing each LPU and the underlying circuit level implementation of these algorithms. Towards that end, we propose here a parallel programming model for exploration of high level LPU algorithms that takes into account existing anatomical observations of the underlying neuronal circuits of the fly brain neuropils. The programing model comprises canonical neural circuit abstractions (cNCAs) and composition rules (CRs) among them [2] (see Figure 1).
cNCAs are fundamental computational units in many LPUs. For example, cartridges and columns are cNCAs in the lamina and medulla LPUs, respectively, of the early visual system, and channels are cNCAs in the antennal lobe of the early olfactory system. Each LPU comprises tens, if not hundreds of its respective cNCAs. The cNCAs, by themselves, implement a particular algorithm that performs local computations. Such an algorithm may utilize multiple neurons that each perform, in turn, elementary operations. They are executed in all cNCAs independently and in parallel. Our focus on using cNCA rather than individual neurons as computational units highlights the preeminence of circuit building blocks underlying neural computation over elementary neuronal operations. It is instructive to compare cNCAs with threads that are widely used in parallel programming models of computer programming.
cNCAs alone can only realize a limited set of overall algorithms due to their independence. In our programming model, communication among cNCAs can be achieved by defining composition rules. CRs are global algorithms that are performed asynchronously and are implemented by a few neurons. By enabling interaction among cNCAs, the CRs facilitate the design of algorithms that can use locally processed information to achieve computation on different spatial brain scales. For example, spatially restricted information that is individually processed by cNCAs can be compared using CRs to implement motion detection algorithms in which computation between spatially displaced visual areas is essential. Thus, CRs are critical for the diversity and functionality of algorithms that can be realized using the programming model proposed here.
The LPU parallel programming model identifies the objects that are necessary in algorithmic I/O design of the neural circuit architecture. It is important then to explore the appropriate transformations that can be efficiently implemented under this model. We demonstrate that the proposed programming model can be applied to a range of sensory LPUs, including those in vision and in olfaction. Furthermore, the neural implementation can flexibly and efficiently be implemented for a range of algorithms that process different sensory inputs. Thus, designing such a programming model facilitates not only the understanding of I/O relationships but also the design of new I/O behaviors for LPUs including odorant preprocessing in early olfaction and motion detection in early vision. Finally, we will demonstrate the compatibility of the designed LPUs in the context of Neurokernel architecture [3, 4, 5].
Acknowledgements
The research reported here was supported in part by AFOSR under grant #FA9550-12-1-0232 and in part by NIH under grant #R021 DC012440001.
References
[1] Chiang, et al. Three-dimensional reconstruction of brain-wide wiring networks in Drosophila at single-cell resolution. Current biology : CB, 21(1):1–11, January 2011.
[2] Aurel A. Lazar, Wenze Li, Nikul H. Ukani, Chung-Heng Yeh, and Yiyin Zhou. Neural circuit abstractions in the fruit fly brain. In Society for Neuroscience Abstracts, November 2013.
[3] Lev E. Givon and Aurel A. Lazar. An open architecture for the massively parallel emulation of the drosophila brain on multiple gpus. BMC Neuroscience, 13:P99, 2012.
[4] Lev E. Givon and Aurel A. Lazar. Neurokernel: An open scalable architecture for emulation and validation of drosophila brain models on multiple gpus. Neurokernel Request for Comments, Neurokernel RFC #1, February 2014.
[5] Aurel A. Lazar, Nikul H. Ukani, and Yiyin Zhou. The cartridge: A canonical neural circuit abstraction of the lamina neuropil – construction and composition rules. Neurokernel Request for Comments, Neurokernel RFC #2, January 2014.
Keywords:
Drosophila,
local processing units,
canonical circuits,
composition rules,
algorithmic IO design
Conference:
Neuroinformatics 2014, Leiden, Netherlands, 25 Aug - 27 Aug, 2014.
Presentation Type:
Poster, not to be considered for oral presentation
Topic:
Large-scale modeling
Citation:
Lazar
AA,
Ukani
NH,
Yeh
C and
Zhou
Y
(2014). A Parallel Programming Model of Local Processing Units in the Fruit Fly Brain.
Front. Neuroinform.
Conference Abstract:
Neuroinformatics 2014.
doi: 10.3389/conf.fninf.2014.18.00024
Copyright:
The abstracts in this collection have not been subject to any Frontiers peer review or checks, and are not endorsed by Frontiers.
They are made available through the Frontiers publishing platform as a service to conference organizers and presenters.
The copyright in the individual abstracts is owned by the author of each abstract or his/her employer unless otherwise stated.
Each abstract, as well as the collection of abstracts, are published under a Creative Commons CC-BY 4.0 (attribution) licence (https://creativecommons.org/licenses/by/4.0/) and may thus be reproduced, translated, adapted and be the subject of derivative works provided the authors and Frontiers are attributed.
For Frontiers’ terms and conditions please see https://www.frontiersin.org/legal/terms-and-conditions.
Received:
04 Apr 2014;
Published Online:
04 Jun 2014.
*
Correspondence:
Prof. Aurel A Lazar, Columbia University, New York, United States, aurel@ee.columbia.edu