NSF Award
9861247
9861247<br/> This Small Business Innovation Research Phase I project details an analog parallel distributed (APD) solution to the shortest path-problem, the time complexity of which depends on the required accuracy of the solution. The algorithm is based on Bellman's formulation of the SPP. Rather than implementing a classic gradient descent in our system as is typical, a new energy minimization technique was developed. The resulting modified gradient descent has the important benefit of not stabilizing at local minima. It also permits a straightforward distributed VLSI implementation which with the currently available technology is expected to have a peak accuracy within 0.5% and time constant of a fraction of a microsecond. A powerful decomposition technique has already been devised and used to characterize the dynamic behavior of the system.<br/> Applications of the algorithm involve analog routing assessment or routing control and search problems in chip-level artificial intelligence. The decomposition method may be useful in analyzing the APD paradigm for other combinatorial problems on graphs.
