Abstract
The Locally Competitive Algorithm (LCA) is a biologically plausible computational architecture for sparse coding, where a signal is represented as a linear combination of elements from an over-complete dictionary. In this paper we map the LCA algorithm on the brain-inspired, IBM TrueNorth Neurosynaptic System. We discuss data structures and representation as well as the architecture of functional processing units that perform non-linear threshold, vector-matrix multiplication. We also present the design of the micro-architectural units that facilitate the implementation of dynamical based iterative algorithms. Experimental results with the LCA algorithm using the limited precision, fixed-point arithmetic on TrueNorth compare favorably with results using floating-point computations on a general purpose computer. The scaling of the LCA algorithm within the constraints of the TrueNorth is also discussed.
Highlights
Physiological evidence exists of sparse coding being employed by biological systems to achieve processing efficiency (Olshausen and Field, 2004)
Experimental results of the Locally Competitive Algorithm (LCA) algorithm mapped to the TrueNorth, which offers limited precision, fixed-point arithmetic, compare favorably with results using floating-point computations on a general purpose computer
We demonstrate the success of the LCA corelet on TrueNorth for dictionaries containing randomly distributed values {–1,0,1} with up to 100 nodes, valuable for use in signal processing applications such as compressed sensing that use image patches as input data
Summary
Physiological evidence exists of sparse coding being employed by biological systems to achieve processing efficiency (Olshausen and Field, 2004). In more general terms the LCA algorithm is a non-linear dynamical system, that computes a sparse approximation of a signal iterating in time until the desired solution is stable. Addressing the challenges of implementing basic iterative linear algebra in neuromorphic hardware such as the TrueNorth, which is available for experimentation and supported by software programming environments (Amir et al, 2013), lays the foundation for further exploration of computational architectures inspired by the nervous system for a wide range of applications in sensory processing and cognitive computing These non-von Neumann combined hardware/software architectures are vital to improving computing capabilities as conventional CPU architectural improvements are plateauing.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
