Abstract:The purpose of this work is study of the neurodynamic foundations of the creative activity of the brain. Modern AI systems using deep neural network training require large amounts of input data, high computational costs and long training times. On the contrary, the brain can learn from small datasets in no time and, crucially, it is fundamentally creative. Methods. The study was carried out through computational experiments with neural networks containing 5 and 7 oscillatory layers (circuits) trained to represent abstract concepts of a certain class of animals. The scheme of neural networks with even cyclic inhibition (ECI networks) contains only bilateral inhibitory connections and consists of two subnets: a reference noncoding network, which is an analogue of the default brain mode neural network, and the main information network that receives time sequences of environmental signals and contextual inputs. After training, the reading of the population phase codes was performed with a simple linear decoder. Results. Conceptual learning of the network leads to the generation of a number of spatial abstract images that are distinguished by the most pronounced features of the relevant line of animals. In computational experiments, a wide set of isomorphic representations of concepts was obtained through: a) transformations of image spaces in a wide range of time scales of the training input signal flow, b) internal regulation of the time scales of mental representations of concepts, c) confirmation on the model of the dependence of psychological proximity of concepts on semantic distance; d) calling from memory (decoding) distributed groups of neurons of animal concepts, which the network has not been trained in. Conclusion. This paper shows for the first time how, using a small set of event input data (a sequence of 4 CCW and 2 CW signals) and very limited computational resources, ECI networks exhibit creative cognitions based on relational relationships, conceptual learning and generalization of knowledge.

Effect in the spectra of eigenvalues and dynamics of RNNs trained with Excitatory-Inhibitory constraint

Different eigenvalue distributions encode the same temporal tasks in recurrent neural networks

Stimulus-Driven and Spontaneous Dynamics in Excitatory-Inhibitory Recurrent Neural Networks for Sequence Representation

Exploring weight initialization, diversity of solutions, and degradation in recurrent neural networks trained for temporal and decision-making tasks

Training dynamically balanced excitatory-inhibitory networks

The effects of inhibitory and excitatory neurons on the dynamics and control of avalanching neural networks

Local Homeostatic Regulation of the Spectral Radius of Echo-State Networks

Approximating Nonlinear Functions With Latent Boundaries in Low-Rank Excitatory-Inhibitory Spiking Networks

Interpreting recurrent neural networks behaviour via excitable network attractors

Consequences of Dale's law on the stability-complexity relationship of random neural networks

Modeling and Contractivity of Neural-Synaptic Networks with Hebbian Learning

Recurrent Neural Network Modeling and Low-Dimensional Analysis of Parkinson's Dynamics

Vibrational Resonance in a Randomly Connected Neural Network

Learning recurrent dynamics in spiking networks

Aligned and oblique dynamics in recurrent neural networks

Neurodynamic model for creative cognition of relational networks with even cyclic inhibition

Excitation-inhibition balance controls information encoding in neural populations

Identifying the impact of local connectivity patterns on dynamics in excitatory-inhibitory networks

Excitatory/Inhibitory Balance Emerges as a Key Factor for RBN Performance, Overriding Attractor Dynamics

Model-free prediction of spatiotemporal dynamical systems with recurrent neural networks: Role of network spectral radius