Abstract
Finding optimal paths in connected graphs requires determining the smallesttotal cost for traveling along the graph's edges. This problem can be solved byseveral classical algorithms where, usually, costs are predefined for alledges. Conventional planning methods can, thus, normally not be used whenwanting to change costs in an adaptive way following the requirements of sometask. Here we show that one can define a neural network representation of pathfinding problems by transforming cost values into synaptic weights, whichallows for online weight adaptation using network learning mechanisms. Whenstarting with an initial activity value of one, activity propagation in thisnetwork will lead to solutions, which are identical to those found by theBellman Ford algorithm. The neural network has the same algorithmic complexityas Bellman Ford and, in addition, we can show that network learning mechanisms(such as Hebbian learning) can adapt the weights in the network augmenting theresulting paths according to some task at hand. We demonstrate this by learningto navigate in an environment with obstacles as well as by learning to followcertain sequences of path nodes. Hence, the here-presented novel algorithm mayopen up a different regime of applications where path-augmentation (bylearning) is directly coupled with path finding in a natural way.