摘要:
A dynamically stable associative learning neural network system include a plurality of synapses (122,22-28), a non-linear function circuit (30) and an adaptive weight circuit (150) for adjusting the weight of each synapse based upon the present signal and the prior history of signals applied to the input of the particular synapse and the present signal and the prior history of signals applied to the input of a predetermined set of other synapses. An embodiment of a conditional-signal neuron circuit (100) receives input signals from conditional stimuli and an unconditional-signal neuron circuit (110) receives input signals from unconditional stimuli. A neural network (200) is formed by a set of conditional-signal and unconditional-signal neuron circuits connected by flow-through synapses to form separate paths between each input (215) and a corresponding output (245). In one embodiment, the neural network (200) is initialized by varying the weight of the input signals from conditional stimuli, until a dynamic equilibrium is reached.
摘要:
A dynamically stable associative learning neural network system include a plurality of synapses (122,22-28), a non-linear function circuit (30) and an adaptive weight circuit (150) for adjusting the weight of each synapse based upon the present signal and the prior history of signals applied to the input of the particular synapse and the present signal and the prior history of signals applied to the input of a predetermined set of other collateral synapses. A flow-through neuron circuit (1110) embodiment includes a flow-through synapse (122) having a predetermined fixed weight. A neural network is formed by a set of flow-through neuron circuits connected by flow-through synapses to form separate paths between each input (215) and a corresponding output (245). In one embodiment (200), the neuron network is initialized by setting the adjustable synapses at some value near the minimum weight and setting the flow-through neuron circuits at some arbitrarily high weight. The neural network embodiments are taught by successively application of sets of inputs signals to the input terminals until a dynamic equilibrium is reached.
摘要:
An image recognition and classification system includes a preprocessor in which a "top-down" method is used to extract features from an image; an associative learning neural network system, which groups the features into patterns and classifies the patterns: and a feedback mechanism which improves system performance by tuning preprocessor scale, feature detection, and feature selection.
摘要:
A dynamically stable associative learning neural system includes a plurality of neural network architectural units. A neural network architectural unit has as input both condition stimuli and unconditioned stimulus, an output neuron for accepting the input, and patch elements interposed between each input and the output neuron. The patches in the architectural unit can be modified and added. A neural network can be formed from a single unit, a layer of units, or multiple layers of units.
摘要:
A dynamically stable associative learning neural network system includes, in its basic architectural unit, at least one each of a conditioned signal input, an unconditioned signal input and an output. Interposed between input and output elements are "patches," or storage areas of dynamic interaction between conditioned and unconditioned signals which process information to achieve associative learning locally under rules designed for application-related goals of the system. Patches may be fixed or variable in size. Adjustments to a patch radius may be by "pruning" or "budding." The neural network is taught by successive application of training sets of input signals to the input terminals until a dynamic equilibrium is reached. Enhancements and expansions of the basic unit result in multilayered (multi-subnetworked) systems having increased capabilities for complex pattern classification and feature recognition.
摘要:
A dynamically stable associative learning neural network system include a plurality of synapses and a non-linear function circuit and includes an adaptive weight circuit for adjusting the weight of each synapse based upon the present signal and the prior history of signals applied to the input of the particular synapse and the present signal and the prior history of signals applied to the input of a predetermined set of other collateral synapses. A flow-through neuron circuit embodiment includes a flow-through synapse having a predetermined fixed weight. A neural network is formed employing neuron circuits of both the above types. A set of flow-through neuron circuits are connected by flow-through synapses to form separate paths between each input terminal and a corresponding output terminal. Other neuron circuits having only adjustable weight synapses are included within the network. This neuron network is initialized by setting the adjustable synapses at some value near the minimum weight. The neural network is taught by successively application of sets of inputs signals to the input terminals until a dynamic equilibrium is reached.