New Neural Network Architecture Simulates the Human Brain

Rappoport Ari, HUJI, School of Computer Science and Engineering, Computer Science


  • Machine learning is the most important technological development of the century, with countless applications.
  • Modern machine learning algorithms utilize neural network architectures (deep learning).
  • Today's neural networks do not operate like biological brains, and do not provide sufficient performance.  

Our Innovation

  • A new innovative neural network architecture based on a new scientific theory of brain function, and simulates how real biological brains work.
  • Each basic computational unit has subunits corresponding to cortical layers, allowing simultaneous flow in all directions, including bottom-up and top-down (`predictions', feedback).
  • Has potential to show human-like intelligence


  • The network finds a pairing between system inputs (digital sensory inputs - e.g. pictures, sounds, data of any kind) and system outputs (e.g., object identification, robot motor commands, etc.).
  • The architecture organizes neurons into units called nodes and organizes inter-neuron connections into different types of networks.
  • Its simulation utilizes a process (the R process) that has several different stages (R modes) and combines nodes, networks, R modes, agents and particles in novel ways.
  • It allows neurons to directly modulate sensory inputs and presents new learning (training) algorithms. It uses auxiliary and specific structures to assist the simulation.


  • The RPNN has the potential to be utilized in a large number of applications world-wide.
  • Specific areas include computer vision, natural language understanding, motor control, autonomous vehicles and robots of all kinds.


Patent Status

Published US-2020-0410346-A1

Contact for more information:

Anna Pellivert
Contact ME: