Workshop "Current Problems of Neuroinformatics"

Neuroinformatics - 2016


Tuesday, April 26                    11.00 – 13.00
Lecture-hall Алексеевский зал

Chair: Prof. SHUMSKIY SERGEY

1. BAZYAN A.S.
Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow
Molecular, cellular and system mechanisms of motivation and emotion, the implementation of goal-directed bahavior and emotional intense brain cognitive map

Molecular, cellular and systemic mechanisms of motivated goal-directed behavior are described basing on the process needs, motivations and emotions. Goal-directed behavior is impossible without the orientation in space and forming a cognitive map. This process implements the hippocampus, via the neocortical connections. The hippocampus is linked to the amygdala, which is involved in the implementation of emotional behavior and organizing emotionally intense cognitive map or context of the environment.

2. NEPOMNYASHCHIKH V.A., PANKOVA N.A., OSIPOVA E.A., IZVEKOV E.I., KRYLOV V.V.
I.D. Papanin Institute for Biology of Inland Waters, Russian Academy of Sciences, Borok, Yaroslavl region
Spontaneous organization of animal behavior in an unfamiliar environment

Animals employ a set of behavioral strategies when face an unfamiliar environment. Some strategies are common for very different organisms, from invertebrates to humans. Another behavioral property of different animals is that they alternate among several strategies instead of employing only one particular strategy. The alternation is spontaneous, and is observed even in the absence of any releasers. This spontaneous organization of behavior may help animals to explore a novel environment.

Thursday, April 28                    11.00 – 12.45
Lecture-hall Алексеевский зал

Chair: Prof. TIUMENTSEV YURY

3. KECMAN VOJISLAV
Virginia Commonwealth University, USA
Fast online algorithm for nonlinear support vector machines and other alike models

Paper presents a unique novel online learning algorithm (OLLA) for eight popular nonlinear, i.e., kernel, classifiers based on a classic stochastic gradient descent (SGD). In particular, the OLLA is derived for following classifiers: L1 and L2 support vector machines (SVMs) with both a quadratic regularizer and with the l_1 regularizer, Regularized huberized hinge loss, Regularized kernel logistic regression, Regularized exponential loss with l_1 regulizer (used in boosting) and Least squares (LS) SVM. The OLLA is aimed primarily for designing classifiers for large datasets. The novel learning model is very simple and comprised of few coding lines only. Comparisons of performances on few real datasets are shown.

4. NAGOEV Z.V.
Institute for Informatics and Control of Regional Problems KBNC Russian Academy of Sciences, Nal'chik
Neural-like multi-agent recursive cognitive architectures in the tasks of intelligent data processing and control

In many cases tasks of intelligent data processing and control are reduced to problems of function approximation. For example, for image recognition tasks it is necessary to approximate the discriminant function and for control tasks --- the control law. The brain copes well with such problems despite the difficult conditions of real environment, characterized by unstructuredness, stochasticity, dynamics, partial observability, uncertainty. At the heart of the ability of intelligence to approximate the required functions are adaptivity and self-learning, which at the neuromorphological level are provided by mechanism of neuroplasticity. Neural-like multi-agent recursive cognitive architectures is a concept and a formalism for describing hypothetic models of brain neuroplasticity, which is based on self-organization of rational agents. The lecture shows theoretical groundings of the metaphor of intelligent system design based on neural-like multi-agent recursive cognitive architectures, it lists a set of heuristics, built on this formalism, which is aimed to overcome barriers to computational complexity and labor, which constraints solution of applied problems, it gives examples of application of this approach for the solution of intelligent data processing and control tasks.

Friday, April 29                    10.00 – 11.30
Lecture-hall Алексеевский зал

Chair: Prof. TEREKHOV SERGE

5. DOROGOV A.YU.
Saint Petersburg Electrotechnical University "LETI"
Problems and challenges of Big Data in neuro-technology

Some peculiarities of neuro-technology applications for Big Data processing are considered. Typical Big Data characteristics are described. Problems of neuro-technology in Big Data processing are shown. Possible methods of decision the problem are offered. The methods include using of special type neural networks, reduction high dimensionality of application tasks and systemic decomposition for data.

6. MAKARENKO N.G.
The Central Astronomical Observatory of the Russian Academy of Sciences at Pulkovo, Saint-Petersburg
The topology from time series

Let’s consider a dynamical system in the phase space. If it has nice properties, a generic projection of its trajectory will be the observed time series. Takens’ algorithm allows us to reconstruct the phase portrait of the model using embedding of the series into Euclidean space of the appropriate dimension. The obtained model is equivalent to the original one up to continuous and differentiable transformations. This technique is referred to as "the geometry of the time series". Its natural generalization is the question of whether and how to restore the topology of the dynamical system from the time series? We are talking about reconstruction, which allow us to calculate the topological invariants such as the Betti numbers or theEuler characteristic. In particular cases we would like to distinguish the motion on a torus, sphere and a plane. Reconstruction is determined, of course, by the class of homologically equivalent spaces. The lecture provides a modern method to solve this problem and discusses some of the practical applications.

Российская академия наук Министерство образования и науки Российской Федерации МФТИ НИЯУ МИФИ НИИСИ РАН МАИ ТРИНИТИ