Seminár z kognitívnej vedy a umelej inteligencie je (od zimného semestra 2015-2016) pokračovaním spoločného seminára z umelej inteligencie, ktorý organizovali Ústav aplikovanej informatiky FIIT STU (prof. Kvasnička) a Katedra aplikovanej informatiky FMFI UK (prof. Farkaš, a pred ním doc. Šefránek). V zimnom semestri je seminár určený hlavne pre študentov kognitívnej vedy, aby mali možnosť zorientovať sa v existujúcom výskume v našom okolí, v letnom semestri má rozšírený záber aj na umelú inteligenciu. Po celý rok sú však vítaní všetci záujemcovia, ktorí sú chcú vypočuť ponúkané prednášky.
Organizátorom seminára je prof. Igor Farkaš.
Čas a miesto konania seminára: utorok 16:30-17:45 v miestnosti I-23 (pavilón informatiky) na FMFI UK.
Humans are perfect at object handling. For a very long time, industrial robots were doing only simple, hard-coded manipulation. High-quality 3D scanning enabled us to localize objects precisely, and plan their picking and movement to avoid collisions. Localization and picking are tasks that we are currently solving with both analytical and machine learning approach. In the talk, we discuss both approaches and compare their advantages and disadvantages.
Relations stand for the links between entities (is unlocked by (COMPUTER PASSWORD)). On the one hand, the same relation can be held between different entities (“is an instrument of” works for "artist brush", "tailor needle", "hairdresser scissors", and "fisherman fishing", etc.) and in turn, may be accessed from different instances (Popov and Hristova, 2015), even when it is irrelevant and can disrupt the task at hand (Hristova, 2009). Furthermore, an active relation may mislead memory as suggested by the Relational luring effect (RLE) (Popov, Hristova, Anders, 2017): 1) word pairs were falsely recognised as studied if they were instances of the learned relations ("snail shell" instead of "bear den") and 2) the correct rejection RTs significantly slowed down with the number of the trials since previous instances of the same relation. On the other hand, since every two entities can be linked through different relations ("artist brush" can be meaningfully related via:”is an instrument of”, “broke”, “hide”, “toss” etc.), the winning ones may be the most contextually or goal relevant, but also the most typical ones. This encoding priority may indicate that some relations are more contextually relevant (Hristova, 2009), but also more typical for the perceived entities. Indeed, it turned out that the typicality of the instance, but not of the role-fillers predicts the RLE (Popov, Hristova, Pavlova, in prep). Hence, the relational, rather than the semantic or the role similarity may explain the RLE and respectively, the heightened readiness of the Long-Term relational representation it witnesses for.
Speech data mining (that will be represented by transcription in this talk) is an important discipline of machine learning. In addition to classical ML and computer science components, it also sources other sciences such as physiology, lexicography, and phonetics, making it a funny inter-disciplinary domain. Similarly to other ML sub-fields, it has been turned upside-down in the recent years by the massive use of neural networks. Although their use in speech dates back to 2000, it is only around 2010 they took the ground and started to dominate the field. Brno speech group has been through all these changes, sometimes following the others, sometimes defining the history. The talk will cover these developments as well as some new research trends.
Customers expect a lot from an online store: that it won't waste their time, that they will find products that fit them, that the experience will be personalized for them, etc. This is exactly what Artificial Intelligence can achieve if it has relevant data that is often hard to get. Fashion retail is in special position because most relevant data is contained in images of products. During the talk we will look at few technologies based on neural networks that can help online fashion retailers to meet customer expectations.
The past decade has witnessed enormous growth in research and applications of machine learning techniques. Modern deep learning techniques have beaten humans in several important benchmarks, yet it is clear that intelligence available in today's systems lags behind even the simplest animals. Part of the lag is theoretical, since there is little agreement on how to build connectionist systems. We would like to draw inspiration from biology, but there is still only fragmentary understanding of learning mechanisms in animal nervous systems. The other critical hurdle that needs to be cleared is what hardware should be used to implement future artificial intelligence systems. In our talk we will present our work on memristors, a promising nanotechnology device that may serve to emulate synapses in the next generation of artificial neural networks. Our results will be framed within the context of competition between deep learning and neuromorphic approaches to artificial intelligence.
In a narrow sense, religion is a unique feature of the human species. According to archaeological findings as well as research in anthropology, this ability is one of the oldest cultural features of a human kind. It has stood up with the rise of gene-cultural coevolution. Religion, in the meaning of religious thinking and religious behaviour, began to form a competitive advantage for human groups and individuals quickly. For groups, it was an additional strengthening mechanism, affecting closer cooperation, higher levels of intra-group altruism, better division of labour, and greater efficiency in resource gathering. Even it is not possible to name with certainty, which attributes of early human affective thinking emerge into religious, reflection of both, cultural and genetic evolution, may show us a way of evolution of religion and cognitive mechanisms, which allowed it and in a feedback loop, stimulating it further.