TY - JOUR
T1 - Dimensionally-reduced visual cortical network model predicts network response and connects system- and cellular-level descriptions
AU - Tao, Louis
AU - Sornborger, Andrew T.
N1 - Funding Information:
Fig. 8 Comparison of sub sampled, pre-synaptic connec tion matrices. (a) Top row, pre-synaptic excitatory connec tion kernels for select neurons sub-sampled from a LSS. Bot tom row, pre-synaptic excitatory connection kernels for the same neurons from the corresponding dimensionally-reduced simula tion. (b) Top row, pre-synaptic inhibitory connection kernels for select neurons sub-sampled from an LSS. Bottom row, pre synaptic inhibitory connection kernels for the same neurons from the corresponding DRM. (c) Top row, the sub-sampled connection matrices projected into the dimensionally-reduced space (left panel, excitatory; right panel, inhibitory). Bottom row, the effective connection matrices from a fit to sub sampled LSS data (left panel, excitatory; right panel, inhibito ry) (see text) Acknowledgments This work was supported by (ATS) NIH NIBIB005432 and (LT) NSF DMS-0506257. ATS would like to thank Liping Wei and the Center for Bioinformatics at the College of Life Sciences at Peking University for their hospitality. Most of the large-scale numerical computations were performed on the NJIT Hydra cluster obtained under NSF MRI-0420590.
PY - 2010/2
Y1 - 2010/2
N2 - Systems-level neurophysiological data reveal coherent activity that is distributed across large regions of cortex. This activity is often thought of as an emergent property of recurrently connected networks. The fact that this activity is coherent means that populations of neurons may be thought of as the carriers of information, not individual neurons. Therefore, systems-level descriptions of functional activity in the network often find their simplest form as combinations of the underlying neuronal variables. In this paper, we provide a general framework for constructing low-dimensional dynamical systems that capture the essential systems-level information contained in large-scale networks of neurons. We demonstrate that these dimensionally-reduced models are capable of predicting the response to previously un-encountered input and that the coupling between systems-level variables can be used to reconstruct cellular-level functional connectivities. Furthermore, we show that these models may be constructed even in the absence of complete information about the underlying network.
AB - Systems-level neurophysiological data reveal coherent activity that is distributed across large regions of cortex. This activity is often thought of as an emergent property of recurrently connected networks. The fact that this activity is coherent means that populations of neurons may be thought of as the carriers of information, not individual neurons. Therefore, systems-level descriptions of functional activity in the network often find their simplest form as combinations of the underlying neuronal variables. In this paper, we provide a general framework for constructing low-dimensional dynamical systems that capture the essential systems-level information contained in large-scale networks of neurons. We demonstrate that these dimensionally-reduced models are capable of predicting the response to previously un-encountered input and that the coupling between systems-level variables can be used to reconstruct cellular-level functional connectivities. Furthermore, we show that these models may be constructed even in the absence of complete information about the underlying network.
KW - Low-dimensional characterization
KW - Network connectivity
KW - Primary visual cortex
UR - http://www.scopus.com/inward/record.url?scp=77549085155&partnerID=8YFLogxK
U2 - 10.1007/s10827-009-0189-8
DO - 10.1007/s10827-009-0189-8
M3 - Article
C2 - 19806444
AN - SCOPUS:77549085155
SN - 0929-5313
VL - 28
SP - 91
EP - 106
JO - Journal of Computational Neuroscience
JF - Journal of Computational Neuroscience
IS - 1
ER -