Convolutional Neural Networks (CNN)
This type of artificial neural network is used to recognize objects in photographs, pictures, or videos. Convolutional Neural Networks are based on the same basic working principles as the sight mechanism in living organisms. These principles first came to light in the work of Nobel laureates David Hubel and Torsten Wiesel, who discovered that nerve cells in the visual cortex respond to stimuli based on different angles of inclination (orientation). The first layers of convolutional networks are able to react to lines with a certain angle of inclination, while the subsequent layers respond to distinctive combinations of these initial lines. The more layers of neural network an image passes through, the more complex the object can be recognized by this network. At the end of the network there are filters (artificial neurons) that are activated when presented with certain facial feature markers, gender markers, age markers, animals, cars, streets or anything else which the neural network has learned to recognize in large collections of images (datasets).
Corticomorphic Neural Networks (CorNN)
Corticomorphic - cerebral cortex-like artificial neural networks have been developed by OCAS to provide rapid learning without the need to present a neural network with an infinite number of other neural networks to learn from. Corticomorphic networks are constructed as laminar (layered) structures consisting of vertically oriented columns of interconnected artificial neurons with distinct functional roles for each of the layers and nuclei. Hierarchically interconnected columns of neurons form functional fields that provide data processing of various levels of complexity in an artificial cortex. Within the framework of corticomorphic neural networks is an innate ability to remember certain objects. The networks ability to recall is a selective process that involves it considering the initial associative basis (logic), independent ongoing association, the branching of associations, error, disregarding the inessential, various types of systemic reconsolidation of memory and other cognitive functions. An important element of corticomorphic neural networks are the afferent-invariant neurons which encode various semantic objects (models of pyramidal cells of the cortex, they are sometimes called "grandmother neurons") capable of being activated by a specific structure of the inputs of their receptive field, as well as novelty neurons and other types of neurons.
Recurrent Neural Networks (RNN)
Recurrent networks are designed to analyze and predict signal sequences, such as natural language texts. The response of recurrent networks depends on both the current input stimuli and the past network conditions. Some types of recurrent neural networks, such as LSTM networks of long short-term memory, are specially adapted for analyzing tasks in which the analyzed events, for example, certain words or phrases, are separated by different time intervals. For the analysis of texts in natural languages, network ensembles are created, including multilayer recurrent networks with bidirectional coding, concatenation (the gluing of multidirectional vectors) and attentional mechanisms.
Models of Neurons with Multiphase Memory
The memory trace forms in the nerve cell, passing through a series of stages, from the appearance of electrical potentials to the formation of new protein receptors and structural rearrangement as a result of the inclusion of genes in the cell nucleus. In the artificial neurons developed by OCAS, the memory trace consolidation process is modelled, including the short, medium and long-term memory formation phases, with the shorter phases serving as the basis for the longer-term ones. The OCAS neurons use parameters such as the frequency of action potentials, axonal facilitation, the release of a mediator, the events of a mediator-receptor connection, postsynaptic potentials and the synthesis of new receptors of the mediator during learning. Different cognitive abilities of neural networks are associated with each of the phases of memory formation in artificial neurons.

To manage the growth and development of large artificial neural networks, OCAS has developed a cybergenomics technology — software algorithms that provide for the formation and modification of the network structure by simulating the effects of genetic expression in living organisms. Elements of artificial neurons (dendrites, soma, axons, and also synapses) are created as sets of software cybergens ensuring their deployment and structural changes. If necessary, signalling events in a neural network can cause its structural reorganization, realizing the effects of epigenic regulation.
Software Development of Biomorphic Neural Networks
In addition to using deep learning libraries such as Tensorflow, Caffe, Theano, and PyTorch, OKAS created two of its own Neural IDE software frameworks, OCAS Cortiphica and OCAS Cognetica. These environments allow you to create various models of neurons, to carry out the visualized development of neural network architectures with various types of neurons and connections between them, to ensure the functioning of the created neural networks, and also to implement technologies for the growth and degradation of neural networks (cybergenomics). Developed by OCAS, Neural IDE is designed as parallel software systems for supercomputers, allows you to create neural networks with tens of millions of soms (bodies), dendrites and axons (processes) of neurons and provide the ability to scale software capacity by increasing the number of cluster computing nodes.
Intellectual Property

Certificate of State Registration of Computer Programs №2018662805 -
Tool for the Development of Neural Networks "OCAS Cognetica"

Certificate of State Registration of Computer Programs №2018662806 -
Integrated Environment for the Development of Neural Networks and Cybergenomics "OCAS Cortiphica"

Certificate of State Registration of Computer Programs №2018662807 -
Interactive System "OCAS Assistant"

Certificate of State Registration of Computer Programs №2018662899 -
Image Recognition System "OCAS Vision"