Title:
BRAIN-LIKE VISUAL NEURAL NETWORK HAVING FORWARD LEARNING AND META-LEARNING FUNCTIONS
Document Type and Number:
WIPO Patent Application WO/2021/233179
Kind Code:
A1
Abstract:
A brain-like visual neural network having forward learning and meta-learning functions, comprising primary feature encoding modules (1) and composite feature encoding modules (2), comprising active and automatic attention mechanisms, comprising a neural circuit that explicitly encodes position information of visual features, comprising forward and backward nerve pathways, and supporting upward and downward information processing processes. By using multiple plasticity processes having biological rationality, forward learning can be performed, visual characterization information in an input image or a video stream is quickly encoded into memory information, information abstraction and information component modulation processes are performed to obtain common feature information and difference feature information among objects, an information channel having multiple information dimensions and information abstraction levels is formed, detailed information is retained while generalization capability is improved, reinforcement learning, supervised learning, and a novelty signal modulation process are also supported, an end-to-end training paradigm having the disadvantages of error backpropagation and gradient descent is not depended on, and a basis is provided for a neural mimetic chip.
More Like This:
Inventors:
REN HUALONG (CN)
Application Number:
PCT/CN2021/093354
Publication Date:
November 25, 2021
Filing Date:
May 12, 2021
Export Citation:
Assignee:
NEUROCEAN TECH INC (CN)
International Classes:
G06N3/063
Foreign References:
| CN110569886A | 2019-12-13 | |||
| CN108333941A | 2018-07-27 | |||
| CN107092959A | 2017-08-25 | |||
| CN110210563A | 2019-09-06 | |||
| US9864933B1 | 2018-01-09 |
Attorney, Agent or Firm:
SHENZHEN ZHONGYI UNION INTELLECTUAL PROPERTY AGENCY CO.,LTD. (CN)
Download PDF: