FIELD: information systems.
SUBSTANCE: group of inventions relates to neural systems and could be used for setting local rules of competitive training, which results in a sparse connectivity among network processing units. Method comprises the steps of: calculating an output signal processing unit in the computer network at least partially based on at least one existing weight and changing at least one weight of a processing unit using a local learning rules, wherein local learning rule creates a sparse connection between the processing units of a computer network by limiting to a predetermined value the weight vector having the weights associated with connections between the processing unit.
EFFECT: technical result is to increase training effectiveness of the area network.
60 cl, 11 dwg
| Title | Year | Author | Number |
|---|---|---|---|
| METHOD AND DEVICE FOR NEURAL TIME CODING, TRAINING AND RECOGNITION | 2012 |
|
RU2597504C2 |
| DETECTING THE ONSET OF SOMNOLENCE | 2016 |
|
RU2734339C2 |
| SELF-CONTROLLED OBJECT DETECTION SYSTEM | 2020 |
|
RU2742803C1 |
| METHOD FOR PROVIDING COMPUTER VISION | 2022 |
|
RU2791587C1 |
| CALCULATION AND/OR ADAPTATION OF MODELS OF MANUFACTURING AND/OR TECHNICAL PROCESSES | 2020 |
|
RU2816861C2 |
| CLUSTERING OF DOCUMENTS | 2020 |
|
RU2768209C1 |
| ELIMINATING IMAGE BLUR | 2020 |
|
RU2742346C1 |
| RELEVANCY ASSESSMENT FOR ARTIFICIAL NEURAL NETWORKS | 2015 |
|
RU2703343C2 |
| SEGMENTATION OF HUMAN TISSUES IN COMPUTER IMAGE | 2017 |
|
RU2654199C1 |
| TRAINING CONVOLUTIONAL NEURAL NETWORK ON GRAPHICS PROCESSING UNITS | 2006 |
|
RU2424561C2 |
