Abstract
In this report, a distributed neural network of coupled oscillators is applied to an industrial pattern recognition problem. The network stems from the study of the neurophysiology of the olfactory system. It is shown that the network serves as an associative memory, which possesses chaotic dynamics. The problem addressed is machine recognition of industrial screws, bolts, etc. in simulated real time in accordance with tolerated deviations from manufacturing specifications. After preprocessing, inputs are represented as 1 × 64 binary vectors. We show that our chaotic neural network can accomplish this pattern recognition task better than a standard Bayesian statistical method, a neural network binary autoassociator, a three-layer feedforward network under back propagation learning, and our earlier olfactory bulb model that relies on a Hopf bifurcation from equilibrium to limit cycle. The existence of the chaotic dynamics provides the network with its capability to suppress noise and irrelevant information with respect to the recognition task. The collective effectiveness of the “cell-assemblies” and the threshold function of each individual channel enhance the quality of the network as an associative memory. The network classifies an uninterrupted sequence of objects at 200 ms of simulated real time for each object. It reliably distinguishes the unacceptable objects (i.e., 100% correct classification), which is a crucial requirement for this specific application. The effectiveness of the chaotic dynamics may depend on the broad spectrum of the oscillations, which may force classification by spatial rather than temporal characteristics of the operation. Further study of this biologically derived model is needed to determine whether its chaotic dynamics rather than other as yet unidentified attributes is responsible for the superior performance, and, if so, how it contributes to that end.
Highlights
This is a continuation of an exploration of pattern recognition capabilities of the olfactory bulb model
The function that is minimized is not an energy level: it is best described as a difference in pattern between an on-going event and a collection of prototypicat spatial patterns, which we describe as lobes or wings of a global chaotic attractor (Yao & Freeman 1990)
Except that we introduce a chaotic associative memory into our pattern recognition system, the system shares similar features with other systems based on neural networks
Summary
This is a continuation of an exploration of pattern recognition capabilities of the olfactory bulb model Biological pattern recognition systems do not go to equilibrium and do not minimize an energy function Instead, they maintain continuing oscillatory activity, sometimes nearly periodic but most commonly chaotic. Our work has been directed toward simulating the pattern recognition capabilities of oscillatory systems, first using limit cycle attractors (Freeman et al, 1988) and using chaotic attractors (Yao & Freeman, 1989, 1990). The approach is closely related to the analogy by Haken (1987) between pattern recognition and nonequilibrium phase transitions in fluids and lasers According to his "'slaving principle", when a chaotic system is raised in energy and brought close to a separatrix, a small fluctuation spreads rapidly and entrains the entire system into a coherent spatial pattern. We compare the efficac~ of our model using chaotic attractors with our previous model using limit cycle attractors, with a binary autoassociator using point attractors, with standard statistical approaches, and with a three-layer back propagation network
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
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 callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
