Gradient-based optimization algorithms for networks of reconfigurable sensors

• Gradient-based optimization for non-linear non-convex multi-resource allocation.
• Algorithms are based on (modified) gradient-search and (de)centralized principles.
• They are not necessarily providing the optimal solution, but some are very efficient.
• The improved centralized (i.e. synchronous) is better than the distributed approach.

A variety of optimization algorithms has been developed for non-linear and non-convex problems in which numerous reconfigurable sensors need to be assigned to many tasks. The algorithms are based on modified gradient-search methods and inspired by centralized/distributed principles. Numerical evaluation of these algorithms on a statistically large set of optimization problems has shown that while each particular algorithm does not necessarily provide the optimal solution in all possible cases, some are very efficient in solving them. Distributed (agent-based) approaches are usually advocated because of their scalability and speed, but the developed centralized (synchronous) algorithms are shown to be better in terms of speed, and simultaneously in terms of effectiveness, and therefore, in terms of efficiency.