Brain-inspired global-local hybrid learning towards human-like intelligence

by   Yujie Wu, et al.

The combination of neuroscience-oriented and computer-science-oriented approaches is the most promising method to develop artificial general intelligence (AGI) that can learn general tasks similar to humans. Currently, two main routes of learning exist, including neuroscience-inspired methods, represented by local synaptic plasticity, and machine-learning methods, represented by backpropagation. Both have advantages and complement each other, but neither can solve all learning problems well. Integrating these two methods into one network may provide better learning abilities for general tasks. Here, we report a hybrid spiking neural network model that integrates the two approaches by introducing a meta-local module and a two-phase causality modelling method. The model can not only optimize local plasticity rules, but also receive top-down supervision information. In addition to flexibly supporting multiple spike-based coding schemes, we demonstrate that this model facilitates learning of many general tasks, including fault-tolerance learning, few-shot learning and multiple-task learning, and show its efficiency on the Tianjic neuromorphic platform. This work provides a new route for brain-inspired computing and facilitates AGI development.



There are no comments yet.


page 13


Increasing Liquid State Machine Performance with Edge-of-Chaos Dynamics Organized by Astrocyte-modulated Plasticity

The liquid state machine (LSM) combines low training complexity and biol...

Recent Advances and New Frontiers in Spiking Neural Networks

In recent years, spiking neural networks (SNNs) have received extensive ...

Evolving-to-Learn Reinforcement Learning Tasks with Spiking Neural Networks

Inspired by the natural nervous system, synaptic plasticity rules are ap...

A Survey of Neuromorphic Computing and Neural Networks in Hardware

Neuromorphic computing has come to refer to a variety of brain-inspired ...

Wearable uBrain: Fabric Based-Spiking Neural Network

On garment intelligence influenced by artificial neural networks and neu...

Dendritic cortical microcircuits approximate the backpropagation algorithm

Deep learning has seen remarkable developments over the last years, many...

Neuromorphic Architecture Optimization for Task-Specific Dynamic Learning

The ability to learn and adapt in real time is a central feature of biol...
This week in AI

Get the week's most popular data science and artificial intelligence research sent straight to your inbox every Saturday.