Workshop "Current Problems of Neuroinformatics"

Neuroinformatics - 2015


Wednesday, January 21                    10:30 – 12:00
Lecture-hall 406

Chair: Prof. TIUMENTSEV YURY

1. V.L. USHAKOV

Large-scale networks in human brain

In the talk it will be considered the modern neurobiological methods of human cognitive functions studying and the large-scale neuronal networks visualization on the basis of the EEG, fMRI and eye tracking methods.

2. S.A. SHUMSKY
The Moscow Institute of Physics and Technology (State University)
Reverse-engineering the brain architecture: role and co-operation of main subsystems

We study the role and co-operation of the main subsystems of mammalian brain. Namely: (i) unsupervised learning of typical patterns in afferent and efferent signals in neocortex; (ii) reinforcement learning of biologically relevant behavior in basal ganglia; and (iii) supervised learning of routine behavior in cerebellum. Parallel organization of such co-operation allows for enhancing brain capabilities by increasing the number of modules in course of mammalian evolution. The same principle of Deep Control may be used in constructing adaptive controllers with scalable architecture.

Thursday, January 22                    10:30 – 13:00
Lecture-hall 406

Chair: Prof. YAKHNO VLADIMIR

3. M.V. KISELEV
The Chuvash state university named after I. N. Ulyanov
Computer simulation of spiking neural networks

The lecture is devoted to a branch of neurscience targeted at computer simulation of spiking neural networks, the most biologically plausible class of neural networks. Varoius models of spiking neurons and networks are discussed. General purpose and specialized neuromorphic computers used for their simulation, methodology for this simulation are described. The final part of the lecture is devoted to practical applications of spiking neural networks and possible perspectives of this research field.

4. YA.B. KAZANOVICH
Institute of Mathematical Problems of Biology RAS
How animals orient themselves in space? Experimental evidence and modeling

The Nobel Prize in Physiology or Medicine 2014 was awarded with one half to John O'Keefe and the other half jointly to May-Britt Moser and Edvard I. Moser "for their discoveries of cells that constitute a positioning system in the brain". The report will cover the main experimental results obtained by the prize-winners and some neural network models which are based on these results.

5. A. A. FROLOV, A. V. ALEXANDROV, P.D. BOBROV, E.V. BIRYUKOVA
Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow
Biologically adequate principles of movement control of human extremities exoskeletons

Post-stroke movement recovery is an important task of neurorehabilitation. To solve it, exoskeletons of human extremities controlled by brain-computer interface are utilized. However the development of exoskeletons is restricted by not sufficient using the principles of human movement control by the central nervous system. Here, two such principles are suggested: 1) the creation of motor synergies, i.e. coordination of joint angles and joint torques during the movement of multi-joint human extremities; 2) joint torque control by a feedback from joint angles with the time delay.

Friday, January 23                    10:30 – 12:00
Lecture-hall 406

Chair: Prof. LITINSKII LEONID

6. V.A. GOLOVKO
Brest State Technical University, Belarus
From multilayers perceptrons to deep belief neural networks: training paradigms and application

This lecture discusses and analyzes the basic paradigm of learning perceptron neural networks: from single layer perceptron to multilayer deep belief neural networks, which are considered now as a revolution in the field of data mining. It is shown the inconsistency of some of the myths about the possibilities of perceptron neural networks and is substantiated the transition to deep belief neural networks. The basic models of deep belief neural networks training are examined, which are based on the restricted Boltzmann machine (RBM) and auto-associative approach. A new method for training of RBM is proposed and is shown that traditional approach for restricted Boltzmann machine training is particular case of proposed technique, which is based on minimization of reconstruction square error. It is proved the equivalence of maximizing the probability distribution of the data in a restricted Boltzmann machine and the minimization of the total reconstruction squared error in layers of RBM. The application of deep belief neural network for compression, visualization and data recognition is considered.

7. N.G. MAKARENKO
The Central Astronomical Observatory of the Russian Academy of Sciences at Pulkovo, Saint-Petersburg
Statistical topology, persistence landscapes and random fields

The lecture is an introduction to modern methods of topological analysis of experimental data. Scalar fields are divided into set of intervals, for each of them we build a special algebraic structure namely a simplicial complex. Number of connected components and holes for such a complex is measured by so-called Betti numbers. The distribution of these invariants by level sets or "lifetimes" allow to recognize patterns of random fields. This analysis becomes meaningful after we introduce satistic methods: specific techniques for point clouds averaging and comparison. As an application, in this lecture we consider problems of "learning manifolds" in a noisy points sample up to a persistent homology. Examples of applications are taken from medicine, astronomy and space monitoring tasks.

Российская нейросетевая ассоциация Российская академия наук Министерство образования и науки Российской Федерации МФТИ НИЯУ МИФИ НИИСИ РАН МАИ Институт перспективных исследований мозга МГУ
AIRI iLabs Приоритет 2030