Igor Farkas

Igor Farkas
Comenius University Bratislava · Department of Applied Informatics

prof. Ing. Dr.

About

93
Publications
10,960
Reads
How we measure 'reads'
A 'read' is counted each time someone views a publication summary (such as the title, abstract, and list of authors), clicks on a figure, or views or downloads the full-text. Learn more
832
Citations
Citations since 2017
36 Research Items
327 Citations
2017201820192020202120222023020406080
2017201820192020202120222023020406080
2017201820192020202120222023020406080
2017201820192020202120222023020406080
Additional affiliations
September 2005 - February 2006
Universität des Saarlandes
Position
  • PostDoc Position
September 2000 - May 2003
University of Richmond
Position
  • PostDoc Position
September 1998 - December 1998
University of Texas at Austin
Position
  • PostDoc Position

Publications

Publications (93)
Chapter
Deep neural networks achieve remarkable performance in multiple fields. However, after proper training they suffer from an inherent vulnerability against adversarial examples (AEs). In this work we shed light on inner representations of the AEs by analysing their activations on the hidden layers. We test various types of AEs, each crafted using a s...
Preprint
Full-text available
Deep neural networks achieve remarkable performance in multiple fields. However, after proper training they suffer from an inherent vulnerability against adversarial examples (AEs). In this work we shed light on inner representations of the AEs by analysing their activations on the hidden layers. We test various types of AEs, each crafted using a s...
Article
Full-text available
Abstraction, one of the hallmarks of human cognition, continues to be the topic of a strong debate. The primary disagreement concerns whether or not abstract concepts can be accounted for within the scope of embodied cognition. In this paper, we introduce the embodied approach to conceptual knowledge and distinguish between embodiment and grounding...
Chapter
Full-text available
UBAL is a novel bidirectional neural network model with bio-inspired learning. It enhances contrastive Hebbian learning rule with an internal echo mechanism enabling self-supervised learning. UBAL approaches any problem as a bidirectional heteroassociation, which gives rise to emergent properties, such as generation of patterns while trained for cl...
Chapter
Hierarchical Reinforcement Learning (HRL) represents a viable approach to learning complex tasks, especially those with an inner hierarchical structure. The HRL methods decompose the problem into a typically two-layered hierarchy. At the lower level, individual skills are created to solve specific non-trivial subtasks, such as locomotion primitives...
Chapter
Full-text available
Intrinsic motivation (IM) research is a promising part of reinforcement learning which can push artificial agents to completely new frontiers. Namely, from agents with a simple action repertoire, driven by the human engineered reward, to more autonomous agents with their own goals and skill development, able to act successfully in the environments...
Book
The proceedings set LNCS 12891, LNCS 12892, LNCS 12893, LNCS 12894 and LNCS 12895 constitute the proceedings of the 30th International Conference on Artificial Neural Networks, ICANN 2021, held in Bratislava, Slovakia, in September 2021.* The total of 265 full papers presented in these proceedings was carefully reviewed and selected from 496 submis...
Chapter
Research on explainable artificial intelligence has progressed remarkably in the last years. In the subfield of deep learning, considerable effort has been invested to the understanding of deep classifiers that have proven successful in case of various benchmark datasets. Within the methods focusing on geometry-based understanding of the trained mo...
Book
The proceedings set LNCS 12891, LNCS 12892, LNCS 12893, LNCS 12894 and LNCS 12895 constitute the proceedings of the 30th International Conference on Artificial Neural Networks, ICANN 2021, held in Bratislava, Slovakia, in September 2021.* The total of 265 full papers presented in these proceedings was carefully reviewed and selected from 496 submis...
Book
The proceedings set LNCS 12891, LNCS 12892, LNCS 12893, LNCS 12894 and LNCS 12895 constitute the proceedings of the 30th International Conference on Artificial Neural Networks, ICANN 2021, held in Bratislava, Slovakia, in September 2021.* The total of 265 full papers presented in these proceedings was carefully reviewed and selected from 496 submis...
Book
The proceedings set LNCS 12891, LNCS 12892, LNCS 12893, LNCS 12894 and LNCS 12895 constitute the proceedings of the 30th International Conference on Artificial Neural Networks, ICANN 2021, held in Bratislava, Slovakia, in September 2021.* The total of 265 full papers presented in these proceedings was carefully reviewed and selected from 496 submis...
Book
The proceedings set LNCS 12891, LNCS 12892, LNCS 12893, LNCS 12894 and LNCS 12895 constitute the proceedings of the 30th International Conference on Artificial Neural Networks, ICANN 2021, held in Bratislava, Slovakia, in September 2021.* The total of 265 full papers presented in these proceedings was carefully reviewed and selected from 496 submis...
Book
The proceedings set LNCS 12396 and 12397 constitute the proceedings of the 29th International Conference on Artificial Neural Networks, ICANN 2020, held in Bratislava, Slovakia, in September 2020.* The total of 139 full papers presented in these proceedings was carefully reviewed and selected from 249 submissions. They were organized in 2 volumes f...
Book
The proceedings set LNCS 12396 and 12397 constitute the proceedings of the 29th International Conference on Artificial Neural Networks, ICANN 2020, held in Bratislava, Slovakia, in September 2020.* The total of 139 full papers presented in these proceedings was carefully reviewed and selected from 249 submissions. They were organized in 2 volumes f...
Chapter
Reinforcement learning has become an established class of powerful machine learning methods operating online on sequential tasks by direct interaction with an environment instead of processing precollected training datasets. At the same time, the nature of many tasks with an inner hierarchical structure has evoked interest in hierarchical RL approa...
Chapter
Neural networks, especially deep architectures, have proven excellent tools in solving various tasks, including classification. However, they are susceptible to adversarial inputs, which are similar to original ones, but yield incorrect classifications, often with high confidence. This reveals the lack of robustness in these models. In this paper,...
Chapter
Full-text available
In recent years, visuospatial cognitive functions, which play a crucial role in human cognition, have sparked interest among psychologists and neuroscientists, focusing on assessment, training and restoration of these functions. Virtual reality, recognized as a modern technology, addressing the real-life aspects of visuospatial processing, provides...
Conference Paper
Full-text available
Artificial neural networks, in particular the deep end-to-end architectures trained by error backpropagation (BP), are currently the topmost used learning systems. However, learning in such systems is only loosely inspired by the actual neural mechanisms. Algorithms based on local activation differences were designed as a biologically plausible alt...
Chapter
In classification tasks, the set of training examples for each class can be viewed as a limited sampling from an ideal infinite manifold of all sensible representants of this class. A layered artificial neural network model trained for such a task can then be interpreted as a stack of continuous transformations which gradually mold these complex ma...
Conference Paper
A major role attributed to mirror neurons, according to the direct matching hypothesis, is to mediate the link between an observed action and agent's own motor repertoire, to provide understanding "from inside". The mirror neurons gave rise to various models but one of the issues not tackled by them is the perspective in/variance. Neurons in STS vi...
Chapter
For the last few decades, the neuroscientific research has highlighted the importance of astrocytes, a type of glial cells, in the information processing capabilities. By dynamic bidirectional communication with neurons, astrocytes regulate their excitability through a variety of mechanisms. Traditional artificial neural networks (ANNs) are connect...
Conference Paper
Full-text available
Current research in neuroscience has begun to shift perspective from neurons as sole information processors to including the astro-cytes as equal and cooperating units in this function. Recent evidence sheds new light on astrocytes and presents them as important regulators of neuronal activity and synaptic plasticity. In this paper, we present a mu...
Article
Full-text available
We provide direct electrophysiological evidence that mirror therapy (MT) can change brain activity and aid in the recovery of motor function after stroke. In this longitudinal single-case study, the subject was a 58-yr-old man with right-hand hemiplegia due to ischemic stroke. Over a 9-mo period we treated him with MT twice a week and measured elec...
Chapter
Full-text available
Current research in neuroscience has begun to shift perspective from neurons as sole information processors to including the astrocytes as equal and cooperating units in this function. Recent evidence sheds new light on astrocytes and presents them as important regulators of neuronal activity and synaptic plasticity. In this paper, we present a mul...
Chapter
Full-text available
Since the standard error backpropagation algorithm for supervised learning was shown biologically implausible, alternative models of training that use only local activation variables have been proposed. In this paper we present a novel algorithm called UBAL, inspired by the GeneRec model. We shortly describe the model and show the performance of th...
Conference Paper
Visuospatial functions play a crucial role in human cognition, which has elicited over years a great deal of research focusing on their assessment, training and restoration. Interestingly , although our visuospatial capacities allow us to understand and infer relationships of 3D objects in space, these 3D aspects of visuospatial processing are prof...
Article
Full-text available
From the onset of cognitive revolution, the concept of mental imagery has been given different, many times opposing, theoretical accounts. Mental imagery appears to be a ubiquitous, yet wholly individual, easy to explain experience on the one hand, being hard to deal with scientifically on the other hand. The focus of this research is on an enactiv...
Conference Paper
Full-text available
To 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) tech...
Conference Paper
Full-text available
Recently, we systematically investigated short-term memory of an echo state network fed with a scalar random input, using computational simulations. We studied the effect of proper reservoir initialization and its subsequent orthogonalization, using two similar gradient descent iterative procedures. It was shown that the measure defined by Jaeger a...
Article
Full-text available
Using the iCub humanoid robot with an artificial pressure-sensitive skin, we investigate how representations of the whole skin surface resembling those found in primate primary somatosensory cortex can be formed from local tactile stimulations traversing the body of the physical robot. We employ the well-known self-organizing map (SOM) algorithm an...
Article
Reservoir computing became very popular due to its potential for efficient design of recurrent neural networks, exploiting the computational properties of the reservoir structure. Various approaches, ranging from appropriate reservoir initialization to its optimization by training have been proposed. In this paper, we extend our previous work and f...
Article
Full-text available
Objective: Mirror therapy (MT) is an approach of neurorehabilita- tion improving motor functions after stroke. MT represents a mental process by which an individual rehearses a given motor action by reflecting movements of the non-paretic side in a mirror as if it were the affected side. Although a number of small-scale research studies have shown...
Conference Paper
Full-text available
ion is a core concept in cognitive science, representing a challenge for all theories of cognition. Conceptualization of abstraction is also complicated by the fact that it is an entity with several potential meanings and involved mechanisms. Abstraction occupies the agenda of many disciplines, including psychology, linguistics , artificial intelli...
Conference Paper
Full-text available
Analogical reasoning, along with inductive, deductive or abduc-tive reasoning, belongs to the fundamental human mechanisms for the environment exploration, learning, or problem solving. Modeling this ability using computer simulations is important, as it might offer mechanistic explanation of these phenomena. In this work, we focus on the part–whol...
Conference Paper
Full-text available
We use computational simulations to analyse the behavior of the recently proposed Bidirectional Activation-based Learning algorithm (BAL) which was inspired by the Generalized Recirculation algorithm (GeneRec). Both algorithms avoid biologically implausible backpropagation of the error signal, and instead use propagation of neuron activations, whic...
Conference Paper
Reservoir computing provides a promising approach to efficient training of recurrent neural networks, by exploiting the computational properties of the reservoir structure. Various approaches, ranging from suitable initialization to reservoir optimization by training have been proposed. In this paper we take a closer look at short-term memory capac...
Conference Paper
Full-text available
The brain encodes the space in various reference frames. The key role in spatial transformations is played by the posterior parietal cortex where neurons combine retinal location of vi-sual stimulus with gaze direction to encode spatial informa-tion. This nonlinear dependence of neuronal responses, gain modulation, is considered a fundamental compu...
Conference Paper
Full-text available
We present a model of a bidirectional three-layer neural network with sigmoidal units, which can be trained to learn arbitrary mappings. We introduce a bidirectional activation-based learning algo-rithm (BAL), inspired by O'Reilly's supervised Generalized Recircula-tion (GeneRec) algorithm that has been designed as a biologically plau-sible alterna...
Conference Paper
Full-text available
We present a model of a bidirectional three-layer neural network with sigmoidal units, which can be trained to learn arbitrary mappings. We introduce a bidirectional activation-based learning algorithm (BAL), inspired by O’Reilly’s supervised Generalized Recirculation (GeneRec) algorithm that has been designed as a biologically plausible alternativ...
Conference Paper
Full-text available
Action understanding undoubtedly involves visual representations. However, linking the observed action with the respective motor category might facilitate processing and provide us with the mechanism to “step into the shoes” of the observed agent. Such principle might be very useful also for a cognitive robot allowing it to link the observed action...
Article
Current connectionist models of bilingual language processing have been largely restricted to localist stationary models. To fully capture the dynamics of bilingual processing, we present SOMBIP, a self-organizing model of bilingual processing that has learning characteristics. SOMBIP consists of two interconnected self-organizing neural networks,...
Conference Paper
Full-text available
Abstrakt V príspevku predstavujeme model systému zrkadlia-cich neurónov vyvinutý v rámci kognitívnej robo-tiky použitím simulátora robota iCub. Pri formovaní modulárnej architektúry nášho modelu sme navrhli algoritmus obojsmerného učenia odvodený od bio-logicky plauzibilného modelu GeneRec. V ˇ clánku prezentujeme výsledky testov zameraných na vlas...
Article
Full-text available
We propose a bio-inspired unsupervised connectionist architecture and apply it to grounding the spatial phrases. The two-layer architecture combines by concatenation the information from the visual and the phonological inputs. In the first layer, the visual pathway employs separate ‘what’ and ‘where’ subsystems that represent the identity and spati...
Article
Full-text available
The concept of computation remains a frequently discussed topic in cognitive science, but there is no consensus about its meaning and the role in this field. I discuss this concept in wider sense, also including nonclassical computation, in the light of Marr’s three levels of analysis and their relevance for main modeling frameworks pursued in cogn...
Conference Paper
Full-text available
V aktuálnom výskume vzniku komunikácie je základnou lokálnou interakciou dyadická koordinačná hra. Tieto interakcie však nemodelujú komunikačné situácie, v ktorých by agenty vytvárali navzájom súperiace koalície a boj o postavenie v rámci skupiny. Našou hypotézou je, že práve tieto druhy motivácií, najmä potreba spoločne klamať, viedli ku vzniku tý...
Article
Full-text available
The recent outburst of interest in cognitive developmental robotics is fueled by the ambition to propose ecologically plausible mechanisms of how, among other things, a learning agent/robot could ground linguistic meanings in its sensorimotor behavior. Along this stream, we propose a model that allows the simulated iCub robot to learn the meanings...
Article
Full-text available
Evidence from behavioral studies demonstrates that spoken language guides attention in a related visual scene and that attended scene information can influence the comprehension process. Here we model sentence comprehension within visual contexts. A recurrent neural network is trained to associate the linguistic input with the visual scene and to p...
Conference Paper
Full-text available
We present the results of an ongoing research in the area of symbol grounding. We develop a biologically inspired model for grounding the spatial terms that employs separate visual what and where subsystems that are integrated with the symbolic linguistic subsystem in the simplified neural model. The model grounds color, shape and spatial relations...
Conference Paper
Full-text available
Abstrakt  Jednou z dôležitých základných kognitívnych schopností človeka (a niektorých druhov zvierat) je porozumenie pozorovaným motorickým akciám. Súčasné teórie porozumenia akciám možno rozdeliť do dvoch skupín, ktoré sa líšia pohľadom na úlohu zrkadliacich neurónov, objavených v mozgu opíc i človeka. Vizuálna hypotéza predpokladá porozumenie a...
Article
Acknowledgements Asada M., K. Hosoda, Y. Kuniyoshi, H. Ishiguro, T. Inui, Y. Yoshikawa, M. Ogino, and C. Yoshida, Cognitive developmental robotics: a survey, IEEE Transactions on Autonomous Mental Development, vol.1, no.1, 12-34, 2009. Lallee S., C. Madden, M. Hoen, and P. Dominey, Linking language with embodied and teleological representations of...
Article
Full-text available
Self-organizing neural network models have recently been extended to more general data structures, such as sequences or trees. We empirically compare three recursive models of the self-organizing map—SOMSD, MSOM and RecSOM—using three different tree data sets with the increasing level of complexity: binary syntactic trees, ternary linguistic propos...
Article
Full-text available
Mental causation is a philosophical concept attempting to describe the causal effect of the immaterial mind on subject behavior. Various types of causal-ity have different interpretations in the literature. I propose and explain this con-cept within the framework of the reciprocal causality operating in the brain bidi-rectionally between local and...
Article
Full-text available
A software system Gel Analysis System for Epo (GASepo) has been developed within an international WADA project. As recent WADA criteria of rEpo positivity are based on identification of each relevant object (band) in Epo images, development of suitable methods of image segmentation and object classification were needed for the GASepo system. In the...
Conference Paper
Full-text available
Processing structured data is a continuing challenge for connectionist models that aim at becoming a plausible explanation of human cognition. The recently proposed linear Recursive Auto-Associative Memory (RAAM) model was shown to have a much higher encoding capacity and not to be subject to overtraining compared to classical RAAM. We assess the e...
Conference Paper
Full-text available
During the last decade, self-organizing neural maps have been extended to more general data structures, such as sequences or trees. To gain insight into how these models learn the tree data, we empirically compare three recursive versions of the self-organizing map - SOMSD, MSOM and RecSOM - using two data sets with the different levels of complexi...
Article
Full-text available
As potential candidates for explaining human cognition, connectionist models of sentence processing must demonstrate their ability to behave systematically, gen- eralizing from a small training set. It has recently been shown that simple recur- rent networks and, to a greater extent, echo-state networks possess some ability to generalize in artific...
Conference Paper
Full-text available
As potential candidates for human cognition, connection- ist models of sentence processing must learn to behave systematically by generalizing from a small traning set. It was recently shown that Elman networks and, to a greater extent, echo state networks (ESN) possess lim- ited ability to generalize in artificial language learning tasks. We study...
Article
Full-text available
Abstrakt. Tento článok ponúka náčrt výpočtového modelu vzťahu myseľ-mozog, ktorý sa opiera o poznatky z kognitívnej neurovedy, teórie nelineárnych dynamických systémov a samoorganizácie, pričom sa kladie dôraz na vysvetlenie kauzálnej interakcie medzi mozgom a mysľou. Budem argumentovať, že na lepšie pochopenie tohto problému je nutné uvažovať tri...
Article
Full-text available
Recently there has been an outburst of interest in extending topographic maps of vectorial data to more general data structures, such as sequences or trees. However, there is no general consensus as to how best to process sequences using topographic maps, and this topic remains an active focus of neurocomputational research. The representational ca...
Conference Paper
Full-text available
Recently, there has been an outburst of interest in extend- ing topographic maps of vectorial data to more general data structures, such as sequences or trees. The representational capabilities and inter- nal representations of the models are not well understood. We concen- trate on a generalization of the Self-Organizing Map (SOM) for pro- cessing...
Conference Paper
Full-text available
Recently, there has been a considerable research activity in extending topographic maps of vectorial data to more general data structures, such as sequences or trees. However, the representational capabilities and internal representations of the models are not well understood. We rigorously analyze a generalization of the Self-Organizing Map (SOM)...
Article
Full-text available
Recently, there has been an outburst of interest in extending topographic maps of vectorial data to more general data structures, such as sequences or trees. The representational capabilities and internal representations of the models are not well understood. We concentrate on a generalization of the Self-Organizing Map (SOM) for processing sequent...
Article
Full-text available
In this paper we present a self-organizing neural network model of early lexical development called DevLex. The network consists of two self-organizing maps (a growing semantic map and a growing phonological map) that are connected via associative links trained by Hebbian learning. The model captures a number of important phenomena that occur in ea...
Article
Full-text available
In this paper, we describe a self-organizing neural network model that addresses the process of early lexical acquisition in young children. The growing lexicon is modeled by combined semantic word representations based on distributional statistics of words and on grounded semantic features of words. Changing semantic word representations are assum...
Conference Paper
Full-text available
In this paper we present the DevLex model of language acquisition. DevLex consists of two self-organizing maps (a growing semantic map and a phonological map) that are connected via associative links. It simulates the early stages of lexical development in children, in particular, word confusion as evidenced in naming errors. The simulation results...
Article
Full-text available
rms (Dale & Fenson, 1996). At every iteration during simulation, the semantic and phonological representations of words are simultaneously presented to both maps. Through self-organization (using Kohonen's algorithm), the network forms an activity on the phonological map in response to the phonological input, and an activity on the semantic map in...
Article
We present a self-organizing neural network model that can acquire an incremental lexicon. The model allows the acquisition of new words without disrupting learned structure. The model consists of three major components. First, the word co-occurrence detector computes word transition probabilities and represents word meanings in terms of context ve...
Article
Full-text available
In this paper we present a self-organizing connectionist model of the acquisition of word meaning. Our model consists of two neural networks and builds on the basic concepts of Hebbian learning and self-organization. One network learns to approximate word transition probabilities, which are used for lexical representation, and the other network, a...
Article
We propose a novel neural network model for representing data structures. The model consists of a hierarchy of Self-Organizing Maps (SOMs) equipped with leaky integrating units. Each of the maps is thus designed to represent sequences of data in a fashion resembling Barnsley’s iterated function system. Each data structure is decomposed into a hiera...
Article
One of the important issues of the Public administration reform strategy proposal in the Slovak Republic is the new concept of management of the incoming and outcoming financial flows within the regions. The goal of the strategy has been to set up a natural and automatic scheme for the management of these flows in order to provide a sufficient posi...
Article
The main goals of the state debt management are recognised in three areas: 1. direct minimization of the (long-term) costs, 2. risk minimization, especially related to the fluctuations of interest rates, and 3. establishing various conditions which could improve the state debt management process. Main aims for the future are discussed regarding the...
Article
Exploratory analysis of financial and economic data is being accepted as a very valuable and important stage of solving of majority of financial and economic tasks. We introduce a short overview of neural network approaches and applications in the financial and economic domain. Self-organizing map (SOM) used provides a unique and integrating neural...
Article
Exploratory analysis of financial and economic data is being accepted as a very valuable and important stage of solving of majority of financial and economic tasks. We introduce a short overview of neural network approaches and applications in the financial and economic domain. Self-organizing map (SOM) used provides a unique and integrating neural...
Article
Full-text available
. One of the main problems associated with artificial neural networks online learning methods is the estimation of model order. In this paper, we report about a new approach to constructing a resource-allocating radial basis function network exploiting weights adaptation using recursive least-squares technique based on Givens QR decomposition. Furt...
Conference Paper
Full-text available
A model is proposed to demonstrate how neurons in the primary visual cortex could self-organize to represent the direction of motion. The model is based on a temporal extension of the self-organizing map where neurons act as leaky integrators. The map is trained with moving Gaussian inputs, and it develops a retinotopic map with orientation columns...
Article
Full-text available
We present a time-series prediction method based on the combination of an unsupervised growing neural network - Dynamic Cell Structures (DCSs) and local linear models (LLMs). DCSs represent the attractor of the underlying dynamic system in the form of directed graph and thus provides the proper quantization of the state space data. Whereas such a m...