NeuroCode is the main component of the suite SIMCOG. It is an artificial spiking neural network editor created to be embed in complete AI agents. The user-friendly icons approach is a strong issue of the software, building a relative complex neural architecture in just few clicks. Once a user has built a "brain controller", it could be test on either a rich virtual 3D-world (NeuroSim) or several physical platforms (Khepera III, K-Team co., LegoMindstorm NXT and EV3), thanks to the compilation mode versatility. A particularity of NeuroCode stands on its middle strategy for detail neural properties modeling combine with actual robotic facilities. A non-exhaustive list for neural features are :


Neuron components:

Several different curves signatures for the basis of the membrane potential variation.

Many sets of standard variables: initial membrane potential, refractory period, after-hyperpolarization potential, resting potential, leak functions, burst functions and noise functions.

Synapse components:

Possibilities of single to multiple, auto-associative or hetero-associative links between neurons.

Inhibitory or excitatory types of synapse.

Time-delay value for spike transmission.

Dynamical synaptic weights variation with scalar value from 0 to 100%.

Individual post-potential sequence for the signal expected to receive on the target. EPSP or IPSP (excitatory or inhibitory post-synaptic potential) are based on an alpha function. The temporal pattern depends on the current states of the synaptic weight value, the parameter of the amplitude and the length of the signal as well as the neural phase state of the target receiver.

Targets of synapses could be neurons (interneurons or sensory-neurons, motoneurons) or effectors (motors, artificial muscles).

Few learning rules could be attached to a synapse such as STDP (spike-timing dependent-plasticity), habituation or sensitization, dynamically modulating the synaptic weights.

Learning functions
©2024 All rights reserved.