Motifs for processes on networks

Schwarze, Alice C.; Porter, Mason A.

doi:10.48550/arxiv.2007.07447

Cited by 2 publications

(3 citation statements)

References 86 publications

(152 reference statements)

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In a general network of multiple units, we find that X comprises all paths through the unobserved units together with all projections to each readout unit (Figure 5c and Supplemental Material). The analysis is similar to previous works which decompose the network response covariance into structural motifs [56,48].…”

Section: The Implications Of Subpopulation Codes For Divergent Cortic...mentioning

confidence: 81%

“…Assuming boundedness of W UU we have the expansion . Since L U J UU are the effective connection strengths in the network, ( L U J UU ) k denotes the k th step through the unobserved population [48], and consequently X is comprised of all paths through the unobserved population and their projections to the observed population ( J OU ). This expansion has been used before to decompose the structure of the response covariance in a neural network [56, 4].…”

Section: Supplemental Materialsmentioning

confidence: 99%

See 1 more Smart Citation

Subpopulation Codes Permit Information Modulation Across Cortical States

Getz

Huang

Doiron

2022

Preprint

View full text Add to dashboard Cite

Cortical state is modulated by myriad cognitive and physiological mechanisms. Yet it is still unclear how changes in cortical state relate to changes in neuronal processing. Previous studies have reported state dependent changes in response gain or population-wide shared variability, motivated by the fact that both are important determinants of the performance of any population code. However, if the state-conditioned cortical regime is well-captured by a linear input-output response (as is often the case), then the linear Fisher information (FI) about a stimulus available to a decoder is invariant to state changes. In this study we show that by contrast, when one restricts a decoder to a subset of a cortical population, information within the subpopulation can increase through a modulation of cortical state. A clear example of such a subpopulation code is one in which decoders only receive projections from excitatory cells in a recurrent excitatory/inhibitory (E/I) network. We demonstrate the counterintuitive fact that when decoding only from E cells, it is exclusively the I cell response gain and connectivity which govern how information changes. Additionally, we propose a parametrically simplified approach to studying the effect of state change on subpopulation codes. Our results reveal the importance of inhibitory circuitry in modulating information flow in recurrent cortical networks, and establish a framework in which to develop deeper mechanistic insight into the impact of cortical state changes on information processing in these circuits.

show abstract

Section: The Implications Of Subpopulation Codes For Divergent Cortic...mentioning

confidence: 81%

Section: Supplemental Materialsmentioning

confidence: 99%

Subpopulation Codes Permit Information Modulation Across Cortical States

Getz

Huang

Doiron

2022

Preprint

View full text Add to dashboard Cite

show abstract

“…In general, patterns contained in a complex network that occur significantly more than what would happen in a random graph are called motifs. The study of motifs allows us to detect relevant processes taking place on the network and contributes in determining the complex behavior of the dynamical system represented by the network [47,48].…”

Section: B Cyclesmentioning

confidence: 99%

Network analysis of Reynolds number scaling in wall-bounded Lagrangian mixing

Perrone,

Kuerten,

Ridolfi

et al. 2021

Preprint

View full text Add to dashboard Cite

The dispersion and mixing of passive particles in a turbulent channel flow is investigated by means of a network-based representation of their motion. We employ direct numerical simulations at five different Reynolds numbers, from Reτ = 180 up to Reτ = 950, and obtain sets of particle trajectories via numerical integration. By dividing the channel domain into wall-normal levels, the motion of particles across these levels is used to build a time-varying complex network, which is able to capture the transient phase of the wall-normal mixing process and its dependence on the Reynolds number, Reτ . Using network metrics, we observe that the dispersion of clouds of tracers depends highly on both their wall-normal starting position and the time elapsed from their release. We identify two main mechanisms that contribute to the long lasting inhibition of the dispersion of particles released near the walls. We also show how the relative importance of these mechanisms varies with the Reynolds number. In particular, at low Reτ the weaker velocity fluctuations appear dominant in inhibiting dispersion, while at higher Reynolds numbers a larger role is played by cyclic patterns of motion. At the higher Reynolds numbers employed in this work, we find that most network properties are Reynolds-independent when scaled with outer flow variables. Instead, at lower Reτ , the aforementioned scaling is not observed. We explore the meaning of the emergence of this scaling in relation to the features of dispersion and to the network definition.

show abstract

Motifs for processes on networks

Cited by 2 publications

References 86 publications

Subpopulation Codes Permit Information Modulation Across Cortical States

Subpopulation Codes Permit Information Modulation Across Cortical States

Network analysis of Reynolds number scaling in wall-bounded Lagrangian mixing

Contact Info

Product

Resources

About