Keynote Speakers

Neuroinformatics - 2018

Monday, October 8                    11:15 – 13:00
Lecture-hall Алексеевский зал


Institute of Higher Nervous Activity and Neurophysiology of RAS, Moscow
Experience in rehabilitation of post-stroke patients using an exoskeleton controlled by the brain-computer interface

National Research University "Higher School of Economics", Moscow
Brain-machine interfaces

Major advances have been made during the last two decades in the field of brain-machine interfaces (BMIs). BMIs link the brain to external devices, with an eventual goal of recovery of motor, sensory and cognitive functions to patients with neurological conditions. BMIs are also very effective for functional rehabilitation. BMIs have emerged from the early ideas that sounded like science fiction to the modern high-tech implementations. In particular, intracranial recordings using multichannel implants have enabled real-time control of artificial limbs by nonhuman primates and human subjects. BMIs can restore upper-limb and lower-limb functions. Furthermore, bidirectional BMIs can provide artificial sensory feedback, allowing users to perceive the movements of prosthetic limbs and their interaction with external objects. BMIs can also multitask, like simultaneously decoding orientation of spatial attention and motor goals. Recently, BMI approach was employed to build brain-nets that enable information exchange between individual brains and execution of cooperative tasks. Overall, BMIs appear to be an efficient approach to augmenting the brain function in both patients and healthy subjects, with limitless perspectives.

Monday, October 8                    14:00 – 16:00
Lecture-hall Алексеевский зал


London Institute for Mathematical Sciences, United Kingdom
Modern approaches to the construction of dialogue systems

Scientific Research Institute for System Analysis, Moscow
Planar Ising-spin models in probabilistic machine learning

One of the approaches in machine learning is probabilistic models that operate with such concepts as binary neurons, the energy of a system, the normalization constant (partition function), the entropy, etc. Most of them are migrated to machine learning from physics. In solid-state physics, the most interesting are the results obtained using asymptotic approximations for crystal lattices of infinite dimensions. One of such most investigated objects is the infi-nite two-dimensional lattice of Ising spin model. In machine learning, on the contrary, lattices of finite sizes are of the greatest interest. In this paper we describe algorithms for the exact calculation of the partition function and other statistical quantities for the planar Ising model of finite dimensions (from N=5x5 to N=1000x1000). A polynomial algorithm for finding the partition function is described in detail, based on the calculation of the determinant of the dual graph of a planar lattice. Also, the results obtained by the classical Metropolis algorithm are discussed. The results obtained with the help of these algorithms allow us to draw some conclusions about the behavior of the model of the two-dimensional Ising model in the finite-size case and compare them with the asymptotic formulas.

Tuesday, October 9                    11:00 – 13:00
Lecture-hall Алексеевский зал


National Research Nuclear University (MEPhI), Moscow
Intellectual agents based on a cognitive architecture supporting humanlike social emotionality and creativity

Human-friendly virtual and physical collaborative robots, or cobots, will work side-by-side with users as helping minds and hands in a variety of creative cognitive tasks, including design, invention, creation of art, or goal setting in unexpected situations in unpredictable environments. These tasks require autonomous reasoning and engage social-emotional attitudes, because the cobot needs to maintain mutual trust with the team or the user. In addition, the cobot needs to understand the global context to be able to determine its role and specific task in a joint mission. All this can be achieved based on a cognitive architecture, supporting social-emotional and narrative reasoning. A general concept of such architecture is presented here, together with an overview of evaluations of prototypes and potential practical applications.

Skolkovo Institute of Science and Technology (Skoltech)
Mathematics of neural networks. Open problem

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