Research projects and other activities

ArNI – Artificial NeuroIntelligence

Project Goal

The ArNI project is aimed at development of the wide range of AI technologies based on spiking neural networks (SNN) – the most biologically plausible models of artificial neural networks. It includes several goals:

  • Development of the SNN theory – methods for analysis and prognosis of SNN behavior, their learning (especially – reinforcement learning), implementation of various memory mechanisms in SNNs, creation of standard neural structures for SNN solving typical problems, optimization of network hyperparameters.
  • Development of neuroprocessor architectures for simulation of large SNNs, creation of algorithmic foundations for practical tasks from robotics, intelligent sensors, IoT, security etc.
  • Development of methodology for creation of very large SNNs targeted at building the intelligent systems comparable with the human intelligence or outclassing it, simulating human brain’s cognitive functions, creation of a constructive theory of higher mental functions.

The project content.

The project is based on computer simulation of large SNNs.

The current stage of the project includes the following tasks:

  1. Development of the ArNI-X SNN simulator for CPU and GPU.
  2. Design of novel models of spiking neurons and synaptic plasticity optimized for various classes of problems.
  3. Design of novel SNN learning algorithms (unsupervised learning, supervised learning, reinforcement learning).
  4. Study of memory mechanism of various levels in SNNs.
  5. Development of algorithms for SNN hyperparameter optimization.
  6. Analysis of efficiency of the SNN models developed from point of view of implementing them on existing and future neurochips (Loihi, AltAI). Creation of hardware-friendly models of neurons. Determination of functional specifications for neuroprocessors for implementing the necessary SNN features.
  7. Determination of practical application areas where usage of software/hardware SNN simulators would be most efficient.

Membership in research communities

I am a Principal Investigator in the Intel Neuromorphic Research Community (INRC). I am a member of the Russian Neural Network Society, and of Program Comittees of several conferences on neural networks (IJCNN, NEUROINFORMATICS).