Limitations

The RBN Toolbox is a powerful tool for the simulation of Random Boolean Networks with low connectivity. However, if we choose to simulate large networks with high connectivity (typically N~K), we rapidly have to deal with huge matrices of some megabytes and even more.

Especially the Rules-Matrix grows exponentially in K as it is of dimensions 2^K x N. For a fully connected network with parameters N=K=30, we already have (considering that one matrix-element needs one byte of memory) a matrix of 2^30*30/(2^20) = 30'720 megabytes! Unfortunately, these memory ressources cannot be provided by Matlab.

As about 50% of the Rules-Matrix are non-zero entries, using the Matlab sparse-matrices would not reduce the memory needed for the storage of the matrix; on the contrary, some tests showed that by using sparse-matrices the memory ressources even grow.

The most effective way to solve this problem, would probably be to change to model / internal representation of networks with high connectivity. By randomly generating the rules for each node ad hoc, instead of saving them permanently in a matrix, we could avoid dealing with huge matrices. However, this approach would generate some additional computational overhead while evolving the network. Furthermore, by randomly choosing the logic transition rules, we loose the property that a node has a fixed rule during evolution and it is unclear how this would affect the evolution af a network.