Back to results
US 9,269,041 B2Granted
Hardware enhancements to radial basis function with restricted coulomb energy learning and/or K-nearest neighbor based neural network classifiers
Assigneein2H2
InventorsBruce K. McCormick, Bill H. Nagel, Chris J. McCormick, Mick Fandrich
Life of this patent
Estimating the term…
Abstract
This disclosure describes embodiments for a hardware based neural network integrated circuit classifier incorporating natively implemented Radial Basis functions, Restricted Coulomb Energy function, and/or kNN to make it more practical for handling a broader group of parallel algorithms.
Claims41
A nonlinear neuron classifier comprising:
- a neuron array that comprises a plurality of neurons that have variable component length and variable depth, and that processes an input vector of variable component length and variable component depth that is input into the classifier for comparison against vectors stored in the classifier;
- a system bus interface configured to detect and capture the input vector;
- and
- logic that is connected in parallel to individual neurons in the classifier and that broadcasts the input vector concurrently to neurons that store vectors of similar component length and depth, for vector distance calculations.
1 more element
A method comprising:
- storing a plurality of vectors of variable component length and variable depth in a neuron classifier comprising neurons of variable component length and variable depth;
- receiving an input vector of variable component length and variable depth by individual neurons of the neuron classifier;
- and
- processing the input vector by at least some of the individual neurons to calculate vector distances between the input vector and one of the stored first vectors.
1 more element
Dependencies are inferred from the claim text.