JLR study group problem: Community detection in car electronics networks A modern car is a system with complex architecture that can be represented by a hierarchy of layers of subsystems (typically 9), ECUs (electronic control units, typically around 100), functions (around 20000). At the user end, this network links (on the function level) to a list of features, which is basically the functionality of the car documented in the driver manual. Every feature is therefore determined by several functions, and the main question of this project is how to 'optimally' group these functions together in ECUs and subsystems. The current approach is to basically represent all functions in a network, write down an adjacency matrix A with non-negative integer entries. Each function has a fixed number of outputs and a_{ij} =2 means that two of the outputs of function i feed into j. One then has to identify community structures for that network, which is mostly done by trying to identify blocks in the adjacency matrix (called the N^2 method) using optimization methods that break down for matrices with more than about 30 functions. The goal of the MSc project is to identify methods for community detection that are well suited to the particular context and test them on realistic function networks, which can contain up to several hundred functions. This is of major importance for JLR since they plan to remodel their car design on a quite fundamental level within the next 10 years. A reference to start with: S. Treviño III, A. Nyberg, C. I. Del Genio and K. E. Bassler
, J. Stat. Mech. Theory E. (2015) P02003 http://www2.warwick.ac.uk/fac/cross_fac/complexity/people/staff/delgenio/publ ist/zscore JLR contact: Alexandros Mouzakitis Academic leader: Sam Johnson