Consensus speed optimisation with finite leadership perturbation in k-nearest neighbour networks

Clark, Ruaridh; Punzo, Giuliano; Macdonald, Malcolm

doi:10.1109/cdc.2016.7798378

Cited by 5 publications

(6 citation statements)

References 18 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…In practice it turns out that adding weights [58] to sample distances improves the classification precision and there have been many other improvements and applications in the years since the first introduction of the algorithm [59][60][61][62][63].…”

Section: K Nearest Neighborsmentioning

confidence: 99%

Predicting particle accelerator failures using binary classifiers

Reščič

Seviour

Blokland

2020

Nuclear Instruments and Methods in Physics Research Section A:

View full text Add to dashboard Cite

Particle accelerator failures lead to unscheduled downtime and lower reliability. Although simple to mitigate while they are actually happening such failures are difficult to predict or identify beforehand. In this work we propose using machine learning approaches to predict machine failures via beam current measurements before they actual occur. To demonstrate this technique in this paper we examine beam pulses from the Oakridge Spallation Neutron Source (SNS). By evaluating a pulse against a set of common classification techniques we show that accelerator failure can be identified prior to actually failing with almost 80% accuracy. We also show that tuning classifier parameters and using pulse properties for refining datasets can further lead to almost 92% accuracy in classification of bad pulses. Most importantly, in the paper we establish there is information about the failure encoded in the pulses prior to it, so we also present a list of feasible next steps for increasing pulse classification accuracy.

show abstract

Section: K Nearest Neighborsmentioning

confidence: 99%

Predicting particle accelerator failures using binary classifiers

Reščič

Seviour

Blokland

2020

Nuclear Instruments and Methods in Physics Research Section A:

View full text Add to dashboard Cite

show abstract

“…The CoI method generates optimised perturbations by detecting influence, using the first left eigenvector, and investigating how this influence changes when key vertices are removed from the network. This was shown to be effective in k-outdegree networks where the CoI method, using 5 input vectors, produced similar results to the output of a numerical optimiser 31 . In Fig.…”

Section: Validation Of Community Influencementioning

confidence: 89%

“…For such a case, the first left eigenvector of the Laplacian matrix was identified as a sub-optimal resource allocation (equivalent to an input perturbation) for achieving fast convergence to consensus 26 . An improvement in this allocation has since been developed for directed k-outdegree graphs, where a near-optimal perturbation vector can be produced by combining first left eigenvectors from manipulated versions of the adjacency matrix 31 . It is worth noting that the globally bounded perturbation optimisation problem, to maximise convergence rate to consensus, has no verifiable solution.…”

Section: Influence Of Network Perturbationsmentioning

confidence: 99%

“…It should be noted that the k-means clustering requires the number of divisions to be defined and, hence, is set to detect the same number of communities as found by the CDI algorithm. The results of these optimisations are compared with the Communities of Influence (CoI) method 31 and a numerical optimiser 32 in Fig. 2.…”

Section: Validation Of Community Influencementioning

confidence: 99%

See 1 more Smart Citation

Network Communities of Dynamical Influence

Clark

Punzo

Macdonald

2019

Sci Rep

Self Cite

View full text Add to dashboard Cite

Fuelled by a desire for greater connectivity, networked systems now pervade our society at an unprecedented level that will affect it in ways we do not yet understand. In contrast, nature has already developed efficient networks that can instigate rapid response and consensus when key elements are stimulated. We present a technique for identifying these key elements by investigating the relationships between a system’s most dominant eigenvectors. This approach reveals the most effective vertices for leading a network to rapid consensus when stimulated, as well as the communities that form under their dynamical influence. In applying this technique, the effectiveness of starling flocks was found to be due, in part, to the low outdegree of every bird, where increasing the number of outgoing connections can produce a less responsive flock. A larger outdegree also affects the location of the birds with the most influence, where these influentially connected birds become more centrally located and in a poorer position to observe a predator and, hence, instigate an evasion manoeuvre. Finally, the technique was found to be effective in large voxel-wise brain connectomes where subjects can be identified from their influential communities.

show abstract

“…If the direction of information travel along an edge was reversed then the FLE would identify the vertices that are most effective sources for spreading information quickly across the whole network. This knowledge has been used previously to allocate resources that drive a network to a fast convergence to consensus [23], [24].…”

Section: Methodsmentioning

confidence: 99%

Eigenvector-based community detection for identifying information hubs in neuronal networks

Clark¹,

Macdonald²

2018

Preprint

Self Cite

View full text Add to dashboard Cite

Eigenvectors of networked systems are known to reveal central, well-connected, network vertices. Here we expand upon the known applications of eigenvectors to define well-connected communities where each is associated with a prominent vertex. This form of community detection provides an analytical approach for analysing the dynamics of information flow in a network. When applied to the neuronal network of the nematode Caenorhabditis elegans, known circuitry can be identified as separate eigenvector-based communities. For the macaque's neuronal network, community detection can expose the hippocampus as an information hub; this result contradicts current thinking that the analysis of static graphs cannot reveal such insights. The application of community detection on a large scale human connectome (∼ 1.8 million vertices) reveals the most prominent information carrying pathways present during a magnetic resonance imaging scan. We demonstrate that these pathways can act as an effective unique identifier for a subject's brain by assessing the number of matching pathways present in any two connectomes. Author summaryThe dynamic response of a network to stimulus can be understood by investigating that system's eigenvectors. The eigenvectors highlight the most prominent nodes; those that are either a major source or destination for information in the network. Moreover by defining a coordinate system based on multiple eigenvectors, the most prominent communities can be detected with the most prominent node detected alongside those in the community that funnel information towards it. These methods are applied to a variety of brain networks to highlight the circuitry present in a flatworm (Caenorhabditis elegans), the macaque and human subjects. Static graphs representing the connectomes are analysed to provide insights that were previously believed to only be detectable by numerically modelling information flow. Finally, we discovered that brain networks created for human subjects at different times can be identified as belonging to the same subject by investigating the similarity of the prominent communities.

show abstract

Consensus speed optimisation with finite leadership perturbation in k-nearest neighbour networks

Cited by 5 publications

References 18 publications

Predicting particle accelerator failures using binary classifiers

Predicting particle accelerator failures using binary classifiers

Network Communities of Dynamical Influence

Eigenvector-based community detection for identifying information hubs in neuronal networks

Contact Info

Product

Resources

About