Neuromorphic architectures are characterized by a network of asynchronous computation units that resemble the behavior of neurons in the brain. With much interest devoted to how these architectures can be utilized in learning and intelligent systems, we note that the massive parallelism of neuromorphic architectures can also be leveraged to solve some NP problems more efficiently than on a von Neumann architecture. In this work, we demonstrate how a neuromorphic processor can be used to solve the classic vertex cover problem via an Ising spin model. Mapping the Ising model to a neural architecture itself requires solving two NP-hard problems at initialization, for which we use approximate solutions. The maximum fan-in and fan-out constraint, common on many neuromorphic processors, requires a graph partitioning, and solving the time-dependency constraints in updating the graph amounts to a graph coloring problem. As a result, space and time efficiency is decreased only by a constant factor without degrading solution quality.

This work integrates deep learning and symbolic programming paradigms into a unified method for deploying applications to a neuromorphic system. The approach removes the need for coordination among disjoint co-processors by embedding both types entirely on a neuromorphic processor. This integration provides a flexible approach for using each technique where it performs best. A single neuromorphic solution can seamlessly deploy neural networks for classifying sensor-driven noisy data obtained from the environment alongside programmed symbolic logic to processes the input from the networks. We present a concrete implementation of the proposed framework using the TrueNorth neuromorphic processor to play blackjack using a pre-programmed optimal strategy algorithm combined with a neural network trained to classify card images as input. Future extensions of this approach will develop a symbolic neuromorphic compiler for automatically creating networks from a symbolic programming language.

The bound to factor large integers is dominated by the computational effort to discover numbers that are smooth, typically performed by sieving a polynomial sequence. On a von Neumann architecture, sieving has log-log amortized time complexity to check each value for smoothness. This work presents a neuromorphic sieve that achieves a constant time check for smoothness by exploiting two characteristic properties of neuromorphic architectures: constant time synaptic integration and massively parallel computation. The approach is validated by modifying msieve, one of the fastest publicly available integer factorization implementations, to use the IBM Neurosynaptic System (NS1e) as a coprocessor for the sieving stage.

Abstract not provided