Workshop "Current Problems of Neuroinformatics"
Neuroinformatics - 2018
Tuesday, October 9 14:00 – 16:00
Lecture-hall Алексеевский зал
Chair: Prof. EZHOV ALEXANDER
The Moscow Institute of Physics and Technology (State University)
Deep structural learning. A new look at reinforcement learning
We present a new approach to Deep Learning, including Deep Reinforcement Learning, which is orders of magnitude less computationally expensive than gradient descent learning of Deep Neural Networks. This makes it very attractive for use in future self-learning robotic operating systems.
2. E.D. SOROKOUMOV, A.L. PROSKURA, T.A. ZAPARA, A.S. RATUSHNYAK
1The Institute of Computational Technologies of SB RAS (ICT SB RAS), Novosibirsk
2
Physical basis of functioning and evolutionary origins of biological information systems
Attempts to create a theory of the work of the brain, due to their fragmentation, have not yet led to fundamentally significant results. To understand the principles and mechanisms of the operation of such a biological information system, it is probably necessary to proceed from the evolutionary origins and physical foundations of the existence of negentropic systems. With this approach, the level of complexity of the task becomes commensurate with existing theoretical and experimental capabilities. Given the structural and functional similarity of systems from the molecular level to the level of the whole brain, one can hope for the effectiveness of this approach. In this study, an attempt has been made to create a paradigm of the operation of biological information systems based on combining the existing data mosaic.
Tuesday, October 9 16:30 – 17:30
Lecture-hall Алексеевский зал
Chair: Prof. DUNIN-BARKOWSKI WITALI
1Scientific Research Institute for System Analysis, Moscow
2The Moscow Institute of Physics and Technology (State University)
Thursday, October 11 10:30 – 12:45
Lecture-hall Алексеевский зал
Chair: Prof. TIUMENTSEV YURY
Skobeltsyn Institute of Nuclear Physics Lomonosov Moscow State University
Fuzzy logics and adaptive neuro-fuzzy inference systems
This lecture is an elementary introduction into fuzzy logics and adaptive neuro-fuzzy inference systems. It covers the following topics: Membership functions. Shortcomings of binary logics and its paradoxicality. Continuous membership functions. Properties of fuzzy sets and operations with them. Fuzzy rules and fuzzy conclusions. Algorithms of fuzzy conclusions. Fuzzy logics and probability. Fuzzy logics of type II. Adaptive construction of fuzzy controllers. Fuzzy logics and neural networks. Adaptive Neuro-Fuzzy In-ference Systems (ANFIS). Applications of ANFIS, examples of their use. The lecture is illustrated by the results of applications of ANFIS for solving benchmark and real world problems.
5. N.G. MAKARENKO, I.S. KNYAZEVA
1The Central Astronomical Observatory of the Russian Academy of Sciences at Pulkovo, Saint-Petersburg
2Sanct-Petersburg State University
Topology and data
This work refers to several author's lectures given at the Neuroinformatics Conferences for the last few years and represents an elementary introduction to the methods of computational topology. This is a new area of applied mathematics which is based on the computer implementation of algebraic topology methods. It is called Topological Data Analysis (TDA). The article contains examples from the solar physics illustrating some theoretical concepts.
6. ALEXANDR A. EZHOV, ANDREI G. KHROMOV AND SVETLANA S. TERENTYEVA
State Research Center of Russian Federation "Troitsk institute for innovation & fusion research", Moscow
Quantumness and irrationality
Our interest to irrationality is mainly due to the agent modeling of unequal society and especially due to the study of critical phenomena in such models. Unfortunately, many similar models miss a point of agent irrational behavior. The last one dramatically change the model behavior. We argue that many current attempts to use quantum approach to describe irrational behavior deal in fact with the rational one. We also further elaborate the approach proposed by authors earlier and based on the definition of irrationality per se. It has been demonstrated that it is possible to define irrational action as the action which makes situation worser without any hope to improve it in future. It is shown that quantum approach is needed to describe this irrational behavior. Concretely, taking into account the analogy between decisions governed by classical implication function and simulated annealing it was shown that staying in classical domain it is impossible to receive decisions which are not rational. We also propose new different generalizations of the classical implication function to quantum domain which lead to the description of different kinds of irrationality including ones that suggest suicide-like behavior. These generalizations show analogy between modeling of irrational behavior and also of quantum simulated annealing.
Friday, October 12 10:30 – 13:00
Lecture-hall Алексеевский зал
Chair: Prof. SAMSONOVICH ALEXEI VLADIMIR
National Research Centre "Kurchatov Institute", Moscow
Grand design of cognitive organization: a look from the bottom, top, left and especially right
8. VITALY M. VERKHLYUTOV, VLADISLAV V. BALAEV, VADIM L. USHAKOV, BORIS M. VELICHKOVSKY
1Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow
2
3National Research Centre "Kurchatov Institute", Moscow
A novel methodology for simulation of EEG traveling waves on the folding surface of the human cerebral cortex
There is an ample evidence on the existence of traveling waves in the cortex of subhuman animals such as rats, ferrets, monkey, and even birds. These waves have been registered invasively by electrical and optical imaging techniques. Such methodology is not possible in healthy humans. Non-invasive EEG recordings show scalp waves propagation at rates two orders greater than the data obtained invasively in animal experiments. At the same time, it has recently been argued that the traveling waves of both local and global nature do exist in the human cortex. In this article, we report a novel methodology for simulation of EEG spatial dynamics as produced by depolarization waves with parameters taken from animal models. Our simulation of radially propagating waves takes into account the geometry of the surface of the gyri and sulci in the areas of the visual, motor, somatosensory and auditory cortex. The dynamics of the electrical field distribution on the scalp in our simulations is fully consistent with the experimental EEG data recorded in humans.
9. V.N. BONDAREV
Sevastopol State University
Digital signal processing with pulsed neural networks
This paper deals with the models of digital processing of signals represented by pulse trains. The input-output model of a pulsed neuron is considered. We analyze the conditions of application of this model for digital spectral analysis and filtration. It is proposed a generalized vector-matrix model of a multi-input pulse neuron, focused on solving digital signal processing problems. The scheme of adaptive filtering based on the multi-input pulse neuron model is considered. For this scheme, we derive supervised learning rules, which allow adapting the impulse responses of synaptic connections. To reduce the number of training parameters of a pulsed neuron we propose using a bank of orthogonal filters. Examples of adaptive signal reconstruction by pulse sequences, suppression of additive noise, adaptive synthesis of bandpass filters and filters with complex frequency response are demonstrated.
