Publications

Journals

  1.  Bayesian Optimal Design for Ordinary Differential Equation Models With Application in Biological Science, Journal of the American Statistical Association, Antony M. Overstall, David C. Woods & Ben M. Parker (2019)

    DOI: 10.1080/01621459.2019.1617154

    Importance: Differential Equations are used throughout the biological sciences, and more widely in the natural sciences and engineering. This paper is one of the first to set out an effective method for optimising the design of these experiments, and shows that astonishing gains in experimental efficiency can be made by considering experiments within this framework. These efficiencies will allow scientists to perform quicker and/or more precise experiments. Examples are given for a wide class of biological experiments, and code and examples given using the acebayes R package which allows practitioners to quickly access the benefits of this academic research in their own experiments.

  2. Optimal design of experiments on connected units with application to social networks, Parker, B. M., Gilmour, S. G. and Schormans J. (2017), J. R. Stat. Soc. C.

    DOI: 10.1111/rssc.12170

    Importance: Statistical analysis has traditionally been used on data which do not have any more than a simple connection structure, but we now live in a connected world, and recent work in a range of fields has sought to examine the network patterns in nature (complexity science or network science). Experimental design has been used with non-connected experimental units and provides a strong framework to help scientists perform effective experiments. In this paper, we investigate how network connections between experimental units affect the design of experiments. As well as providing applications to social networks, presents a general method to design experiments on experimental units connected in non-trivial (networked) method. We show that large efficiency gains can be used for well designed experiments.

  3. Likelihood-based assessment of graph evolution models. Complex Networks, 2016. Richard G. Clegg, Ben Parker, and Miguel Rio

    DOI: 10.1093/comnet/cnv031

    http://www.richardclegg.org/sites/default/files/papers/feta_comnet_2015.pdf

    Importance: Rigorous statistical analysis exists for data where there is no connection between elements in the data set, but how do we do adapt this for the networked world? This paper provides the basis for a likelihood-based framework for studying dynamic graph evolution- seemingly a gap in the literature. This paper acts as a foundation for other researchers to apply statistics which is already developed to networks which are growing, and generalises traditional methods in statistics for use with networks.

  4. Optimal Design of Measurements on Queueing Systems. Queueing Systems (2015): Volume 79,Issue 3-4,pp 365-390. Ben M Parker, Steven Gilmour, John Schormans , Hugo Maruri-Aguilar.

    DOI: 10.1007/s11134-014-9421-y

    http://eprints.soton.ac.uk/383915/2/MM1QueueDesignv20.pdf

    Importance: Queues occur in many human environments, but are also an important way of controlling traffic in data communication networks. Despite their ubiquity, measuring queues effectively (inference) is not well understood. Previous work on inference for queues has shown some of the problems in finding estimators of the queues with low variance, in the case where measurement increases traffic in the queue. In this paper, we set out how to optimally measure the arrival and departure rate for an M/M/1 queue. This is one of the first papers for optimal design of queueing systems, and can provide people who work in measuring queues- for example telecommunications providers- a framework for more effective measurement.

  5. Increasing throughput in IEEE 802.11 by optimal selection of backoff parameters; M. Parker, J. Schormans and S.G. Gilmour; IET Networks 2014

    DOI: 10.1049/iet-net.2013.0021

    Importance: Wireless Ethernet uses a “media access control” algorithm which controls when a wireless network device can transmit and receive information. By modelling the problem as a novel Markov chain, based on the de facto standard of the Bianchi model, we are able to find optimal parameters for use in the access algorithm to increase throughput. The relevance of this is that we are able to improve throughput in wireless networks by at least 5%, as well as demonstrating a rigorous statistical method in an electronic engineering problem.

  6. A utility based framework for optimal network measurement; Ben M. Parker, John Schormans, Steven G. Gilmour; IET Networks 2014

    DOI: 10.1049/iet-net.2013.0014

    http://eprints.soton.ac.uk/363697/1/paperv8-revisionToIET.pdf

    Importance:  When measuring communications networks, we use (active) probes. These add to the traffic in the network. Thus by measuring with more probes, we decrease the variance, but potentially increase bias in the measurement. This paper presents a utility based method for quantifying this trade off, which may provide a superior measurement technique for practitioners.

  7. Design of Experiments for Categorical Repeated Measurements in Packet Communication Networks; Technometrics Volume 53, Issue 4, 2011 Ben M. Parker, John Schormans, Steven G. Gilmour.

    DOI: 10.1198/TECH.2011.10052

    Importance: In this paper, we generalise results in paper 1, and show how we can optimally measure any system that evolves according to the Markov principle. This has wide application in measuring communications networks. We present new techniques in optimal design of experiments, for example using automatic differentiation in optimisation of our design criterion.

  8. Measurement of Packet Loss Probability by Optimal Design of Packet Probing Experiments; IET Communications Special Issue, June 2009, Volume 3, Issue 6. Ben M Parker, Steven G Gilmour, John Schormans.

    Importance: One of the first papers to apply statistical principles of design of experiments to measurement problems in packet communication networks, which account for vast majority of global information exchange, so being able to measure them correctly is important. We introduce a novel way of optimally measuring Markov chains for a particular example of a network buffer