Abstract:There is a lack. of general-purpose in-place learning networks that model feature layers in the cortex Currently. It means that a general yet adaptive high-dimensional function approximator is lacked. Here, in-place learning is a biological concept rooted in the genomic equivalence principle. In another word, in-place learning means that each neuron is fully responsible for its own learning in its environment and there is no need for an external learner. In this chapter, we present the multilayer In-place Learning Network (MILN) for this ambitious goal. Computationally, in-place learning provides unusually efficient learning algorithms whose simplicity, low computational complexity, and generality are set apart from typical conventional learning algorithms. Based on the neuroscience literature, and in the 6-layer laminar cortex, we model the layer 4 and layer 2/3 as the feature layers, where layer 4 uses unsupervised learning and layer 2/3 uses supervised learning. As a requirement for autonomous mental development, MILN generates invariant neurons in different layers, with increasing invariance from earlier to later layers and the total invariance in the last motor layer. Such self-generated invariant representation is enabled mainly by descending (top-down) connections. The self-generated invariant representation is used as intermediate representation for learning later tasks in open-ended development. Also, this in-place learning engine can incrementally learn multiple tasks. It Is important for tasks of robotics Such as auto-navigation task and object recognition task through complex motor and sensory interactions. Experimental results demonstrate the effectiveness and efficacy of the proposed modal.

A lateral inhibition neural network based on neocortex topology

Adaptive Deep Spiking Neural Network with Global-Local Learning Via Balanced Excitatory and Inhibitory Mechanism

The effect of lateral inhibitory connections in spatial architecture neural network

Research on Artificial Neural Networks with Spatial Architecture Based on Span Connection and Lateral Inhibition Mechanism.

Biological lateral inhibition and digital cellular neural network applied on associations memory

Local interconnection neural networks

A New Neural Network Model And Its Application

Local Interconnection Neural Network and Its Optical Implementation

A Neuron-Significance-Index-Based Self-Organization Pruning Algorithm For S-Linn

Biologically Inspired Structure Learning with Reverse Knowledge Distillation for Spiking Neural Networks

Local lateral connectivity is sufficient for replicating cortex-like topographical organization in deep neural networks

Lateral Inhibition-Inspired Convolutional Neural Network for Visual Attention and Saliency Detection.

A Lateral Inhibitory Spiking Neural Network for Sparse Representation in Visual Cortex.

Multilayer In-Place Learning Networks with Adaptive Lateral Connections: Models and Simulations

Multilayer In-Place Learning Networks for Modeling Functional Layers in the Laminar Cortex.

A Novel Neuron Connection Model Mimicking Human Beings.

Intralayer-Connected Spiking Neural Network with Hybrid Training Using Backpropagation and Probabilistic Spike-Timing Dependent Plasticity

Sparse Code Formation with Linear Inhibition

Super Interaction Neural Network.

A New Neural Network Model Based on the Recent Discovery of Brain Microenvironment

Cortex Inspired In-Place Learning Networks For General Invariance And Multi-Tasks