[mephi] Научный семинар ИИКС НИЯУ МИФИ

Чт Мар 12 13:51:35 MSK 2020

Уважаемые коллеги,

1. Открыт прием статей на конференцию BICA*AI 2020: http://bica2020.bicasociety.org и https://bica2020.org
Приглашаем Вас принять дистанционное участие в виртуальной стендовой сессии с публикацией статьи при оплате сниженного оргвзноса. Все публикации индексируются в Web of Science и Scopus и выйдут в печать до 10 декабря 2020 г. Заявки принимаются только через EasyChair. Просьба выложить абстракт как можно раньше, до подачи статьи. Дедлайн будет продлен.

2. В понедельник, 16 марта, состоится очередное заседание научного семинара ИИКС НИЯУ МИФИ.
Начало в 17:00. Место проведения: НИЯУ МИФИ, аудитория К-716. 
Адрес: Каширское шоссе 31.

На повестке дня - доклад проф. Магнуса Джонсона (на английском языке; детали ниже). Приглашаются все. Просьба разослать объявление Вашим студентам и коллегам.
Тем, кому нужен пропуск в МИФИ, просьба обращаться заранее к Хапалову Виктору Игоревичу (VIKhapalov на mephi.ru): ему нужно прислать точные паспортные данные. Тем, кто не сможет приехать до закрытия бюро пропусков (до 16:45), необходимо сообщить об этом заранее для того, чтобы коллеги встретили их на проходной с готовым пропуском.

Информация о докладчике:

Magnus Johnsson is a cross-disciplinary scientist, senior lecturer and scientific editor from Sweden. He also does some consultancy work and pursue entrepreneurial ventures. He is currently at Malmö University in Sweden, but is also affiliated with Moscow Engineering Physics Institute and has held positions at Lund University in Sweden. He holds editorial positions at several international scientific journals and has served as a scientific editor for books and special journal issues. His research interests are wide and include, but are not limited to, autonomous systems, computer modeling, artificial neural networks, artificial intelligence, cognitive neuroscience, cognitive robotics, cognitive architectures, cognitive aids and the philosophy of mind. Dr. Johnsson has experience from working in the industry and he has a keen interest in the application of neural networks and artificial intelligence to fields like industry, finance, and medicine. Web page: www.magnusjohnsson.se E-mail: magnus на magnusjohnsson.se Recent invited talks: https://vimeo.com/384903411  https://vimeo.com/296707570

Информация о докладе:

Title: Bio-Inspired Machine Perception

Abstract:

I present some of my thoughts on how to build a cognitive architecture, in particular the perceptual parts, and present examples of implementations from my own research.

In principle you could try to identify theoretically possible principles for a cognitive architecture and then choose a well-integrated and mutually consistent subset (optimally selected with respect to existing technologies, such as computer architectures etc.) of these as a foundation for the design. A shortcut to find such a well-integrated and mutually consistent subset is to look at nature and the principles that have evolved in biological systems, and then take inspiration from these. This bio-inspired approach is what I adhere to.

As a starting point you could try to identify the minimal set of crucial systems that enables cognition in a biological system, such as a mammal, together with their most important mutual interactions. Then you can try to phase out what would constitute minimalist implementations of these systems and how they would be connected and interact. By implementing this you would produce a minimalist but holistic cognitive architecture that I believe would exhibit interesting behaviors.

On the other hand, you could focus on a specifically identified system - such as a sensory modality, say visual object recognition - and try maximize its performance, which is the approach of many researchers. My approach is somewhere in the middle.

Ultimately, you could try to identify and implement principles for how the architectural structure as such could self-organize in analogy with the developmental phases of the modeled biological system, but I will not go into that during this talk. Rather I will discuss some general principles for more composed cognitive architectures that I believe to be fruitful.

In my research I try to find such general principles by following a bio-inspired systems level modeling approach. Hence, my focus right now is not on whether the various components of a cognitive architecture is optimized regarding performance.

So far my focus have particularly been on finding principles for the sensory parts of a cognitive architecture. These employ self-organizing topographically ordered feature representations, reminiscent of what has been found in mammal brains, as fundamental building blocks.

These topographically ordered feature representations are to a greater or lesser degree hierarchically interconnected so that more complex feature representations are built up from less complex; laterally interconnected so that representations in various hierarchies are associated with each other; recurrently interconnected so that activity can be associated over time and so that activity in more complex feature representations can be associated with activity in feature representations of lower complexity.

I will discuss some principles that I believe are important for implementing perception in a cognitive architecture, that simultaneously enable various kinds of memory, expectations, and imagery. I will also discuss how these principles enable the representation of new concepts, as well as the notion of non-existent and perhaps impossible objects, and how they may enable the creation of an inner world and a first-person functional awareness of phenomenal information.

До встречи на семинаре. С уважением, --Алексей В. Самсонович, Ph.D. Professor, Department of Cybernetics, NRNU MEPhI