Implementing Stochastic Hopfield-Network-based Linear Solvers on a Hardware-Constrained Neural Substrate
نویسندگان
چکیده
IBM’s TrueNorth neurosynaptic system provides an appealing platform for deploying numerical algorithms for ultra-low power, real time, and mobile applications. A recurrent Hopfield neural network is used to solve for the Moore-Penrose matrix pseudoinverse to solve a broad class of linear optimizations. The TrueNorth hardware platform is heavily constrained through weight quantization and severely limits range and precision of numerical representation and computation. We show that a flexible, robust, and realtime implementation of an optical flow algorithm can be deployed on TrueNorth with minimal resource allocation and high energy-efficiency. These results show promising potential for TrueNorth as an ultra-low power generalize matrix inverse calculator.
منابع مشابه
Computing Generalized Matrix Inverse on Spiking Neural Substrate
Emerging neural hardware substrates, such as IBM's TrueNorth Neurosynaptic System, can provide an appealing platform for deploying numerical algorithms. For example, a recurrent Hopfield neural network can be used to find the Moore-Penrose generalized inverse of a matrix, thus enabling a broad class of linear optimizations to be solved efficiently, at low energy cost. However, deploying numeric...
متن کاملA Modified Hopfield Neural Network Method for Equality Constrained State Estimation
Electric power system is a highly complex and non linear system. Its analysis and control in real time environment requires highly sophisticated computational skills. Computations are reaching a limit as far as conventional computer based algorithms are concerned. It is therefore required to find out newer methods which can be easily implemented on dedicated hardware. It is a very difficult tas...
متن کاملHardware Neural Network Implementation of Tracking System
A neural network (NN) filter/target-tracking system has been developed as reported in [6]. The design accepts and inputs signal data to a noise/target classifier which uses spectral estimation techniques to distinguish noise from real targets. In that design, the NN is used to calculate the coefficients of an auto regressive linear predictive filter. The current evolution of that design invokes...
متن کاملRecurrent Neural Networks for Sub-optimal Multiuser Detection
This paper explores the use of recurrent neural networks for sub-optimal detection in code division multiple access systems. Research has shown that detectors based on the Hopfield recurrent neural network suffer from localized optimization. The basic Hopfield model is reviewed and we illustrate its use as a multiuser receiver. We investigate the use of stochastic methods to achieve a global mi...
متن کاملAdaptive Control for a Class of Stochastic Passive Hopfield Networks
The paper presents passivity conditions for a class of stochastic Hopfield neural networks with state-dependent noise and with Markovian jumps. Our contributions are mainly based on the stability analysis of the class of stochastic neural networks considered, using infinitesimal generators of appropriate stochastic Lyapunov-type functions. The passivity conditions derived are expressed in terms...
متن کامل