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-9 (pavilón informatiky) na FMFI UK.
Cognitive robotics aims at explaining human cognition by constructing artificial agents (physical or simulated) and equipping them with learning mechanisms to model cognition. I will introduce the paradigms of supervised, unsupervised and reinforcement learning in the context of artificial neural networks. We will illustrate their use in several examples in cognitive robotics. The ideas will be presented via selected tasks of motor learning, learning body schema and spatial cognition in a simulated humanoid robot. These examples will serve as motivation for potential research projects.
When speaking, meaning we want to convey is (somehow) rooted in our experience and (somehow) represented in brain. In this talk I will present several computational models of various aspects of memory, language and autonomous meaning construction.
The aim of the talk is to introduce selected problems within recent discussions on the nature of the Self. I intend to highlight inconsistencies on the concept or concepts of "self" ("I") as well as complexity of the phenomenon itself. I argue that recent findings from neurocognitive and clinical research support the idea of a physical nature of feeling the existence of the Self. I also aim to outline implications from research in cognitive linguistics on the nature of the concept of "self" and the way we conceptualise inner states of our experience.
o improve upper-limb neuro-rehabilitation in chronic stroke patients we apply new methods and tools of clinical training and machine learning for the design and development of an intelligent system allowing the users to go through the process of self-controlled training of impaired motor pathways. We combine the brain–computer interface (BCI) technology with a robotic arm system into a compact system that can be used as a robot-assisted neuro-rehabilitation tool: (1) We use mirror therapy not only to improve motor functions but also to identify subject’s “atoms”, i.e. specific EEG patterns associated with imagined or real-hand movements, using a parallel factor analysis. (2) We designed a BCI-based robotic system using motor imagery in a patient with an impaired right upper limb. The novelty of this approach lies in the control protocol which uses spatial and frequency weights of the estimated sensorimotor atoms during the MT sessions.
Accurate understanding of probability and risk is crucial for informed choices. However, decades of research on human decision making have provided evidence that people are prone to numerous biases in probabilistic reasoning. Large part of these deviations is related to external representation of data. For instance, people are sensitive to statistical format of numerical information and to verbal probability framing. The lecture will cover three main themes: i) a short review of experimental evidence on how simple changes in information wording affect processing of risks and probabilities, ii) practical applications in medical, environmental and financial domain, and iii) suggestions and challenges for further research.
The talk will outline the processes of neural development of human central nervous system and its functional units during prenatal and postnatal life. The hierarchy of central regulatory systems and processing in relation to emotions and social cognition will be introduced. I will deal with the hormonal influence, particularly testosterone, on particular anatomical structures involved in cognition and its consequences on brain development and nervous functions resulting in communication deficits and disturbances in social behaviour. The results of our own research in healthy human individuals and in patients with autism will be discussed.
Long-term synaptic plasticity is widely accepted to be the major mechanism involved in learning and memory. It remains unclear, however, which particular activity rules are utilized by neurons to induce synaptic plasticity in behaving animals and what is the role of dendrites in this context. In the first part of my talk, I will present computer simulations which indicate that the interplay of STDP-BCM plasticity rules and ongoing neuronal activity is able to account for experimentally observed synaptic plasticity in the hippocampus of awake rats. In the second part of my talk, I will describe anatomically detailed models of hippocampal neurons which predict that changes in dendritic morphology are able to selectively modulate local synaptic plasticity.
Humans recognize themselves in the mirror, but children under 18 months do not, thinking they are looking at a person behind the mirror. In animals, the mirror self-recognition is present only in exceptional cases, for instance in chimpanzees, but not in cats. Can a robot recognize itself in the mirror? And what is the origin of the self-recognition ability? Using the simulator of the humanoid robot iCub, and a camera capturing the area in the front of the monitor, we created a control system for examining how the mirror self-recognition emerges from basic mechanisms. These mechanisms include the body model (proprioception), mirroring, i.e. the ability to create an analogical model as a result of vision, imitation and social modelling. We follow the approach of Scassellati and Hart (Yale University) who suppose that a robot should recognize itself in the mirror due to perfect correlation between the body movement and the image seen in the mirror.
Cerebrovascular diseases including stroke are the second most common cause of death and disability worldwide. One of the most devastating consequences of stroke is a cognitive impairment, which can significantly affect activities of daily living eventually leading to dementia. Dementia due to stroke or other cerebrovascular disease is the most common type of dementia after Alzheimer’s disease. However, even more patients after stroke have milder forms of vascular cognitive impairment which do not fulfill criteria for dementia but do diminish their quality of life. Cognitive syndrome in patients with cerebrovascular diseases is characterized by memory deficit consisting of impaired recall but relative preservation of recognition, dysexecutive syndrome, slowed information processing and mood changes. Our research also showed characteristic changes in the EEG alpha frequency band which correlate with cognitive impairment. Interestingly, vascular dementia and Alzheimer's disease share the same, well-known vascular risk factors: hypertension, diabetes, elevated cholesterol, smoking etc. These data suggest interaction between vascular changes and amyloid pathology and led some researchers to formulate vascular hypotheses of Alzheimer's disease. Controlling vascular risk factors a significant number of dementia cases can be prevented and the burden of this disease can be reduced.
Cognitive impairment is a common consequence of acute cerebrovascular accident (stroke). In the talk, we will more specifically focus on three particular aspects of cognition, namely: attention, working memory and motor skills. We implemented computerized versions of standard psychological tests (lateralized attention network test, digit span and fine-motor redrawing task) adjusted for our experimental group of elderly people after stroke. To assess the effect of stroke itself, we administered the tests to an age-matched control group of healthy senior volunteers. This research is a part of a larger cooperative project which aims to find a link between daily cognitive performance of stroke patients and the previous night sleep profiles (sleep quality). At the end of the talk, preliminary results of the main project will be briefly summarized.
Research in neuroscience over the past few decades has shed new light on glial cells which were always considered as purely passive supportive cells. New data provides evidence that astrocytes, a group of glial cells, possess important physiological functions that distinguishes them from passive cells. It is now known that astrocytes are actively involved in neuronal communication regulation and synaptic transmission. Similar to neurons, astrocytes form glial syncytium that enables them to communicate with one another over long distances using Ca2+ signals. Since this is a relatively new area of research in neuroscience, computational models (biophysical and connectionist) are still missing. In my talk I briefly introduce glial cells (mostly their physiology), focus on existing connectionist models and also present my preliminary results.
In computer vision, recent and rapid advances in convolutional deep neural networks (DNNs) have resulted in image-based computational models of object recognition which, for the first time, rival human performance. This talk will focus on practical use of such supervised models in our company for real-time detection and classification. We will also discuss some of the techniques how we achieve real-time performance and how we generate our datasets just as we will discuss some prevailing problems and unrealistic expectations.
In the talk, it will be explained why automatic model building for time-series has become attractive for energy industry and why they decided to develop an automatic model building engine TIM (Tangent Information Modeller). The talk will include mathematical challenges of automation in model building and some real-world deployment scenarios will be described where TIM was used to build large-scale forecasting systems. Information Criteria and Information Geometry topics will also be discussed, as advances in this field helped to enable automatic model building of high-quality models with excellent generalization capabilities.
We present our improved algorithm for finding the clique number of simple undirected graph based on Ostergaard's algorithm applied to functional brain networks. The clique number of a graph is a size of its maximum clique. Finding that clique is a NP-hard problem. Our algorithm implements several pruning techniques which greatly restricts depth-first search branching using the original method. The resulting algorithm works faster on arbitrary simple undirected graphs, but the best performance is on the graphs with a scale-free property. We have used this algorithm to find and analyse clique numbers of 40 functional brain networks for three groups of subjects: elderly patients suffering from Alzheimer disease, elderly people and young healthy individuals.
Global trend towards urbanization calls for new mobility concepts. The UP-Drive consortium is convinced that automated driving technology is the key component enabling more comfort and safety, reduction of congestion and more efficient use of resources. Yet, today's automated driving technology is not mature enough to handle the complexity of urban traffic. The main goal of UP-Drive is to push forward the perception, localization and reasoning abilities of autonomous vehicles. In the course of the project, we are building a prototype car systems capable of driverless operation in complex urban environments. Our focus is placed on residential areas and speeds up to 30 km/h.