Project Description

print

Cognitive psychology aims to understand how human beings sense, perceive, attend, remember, think, and solve problems. The research field has developed rapidly during the last five decades leading to several ground-breaking discoveries, some of which have recently received the Nobel prize [1]. Cognitive psychology has merged with the neurosciences forming the field of cognitive neuroscience, which focuses on how cognitive mechanisms are implemented in the central nervous system of living organisms [2]. Cognitive psychology has also joined forces with areas of the technical sciences such as signal and image processing, automatization, and ergonomics leading to the field of cognitive science which includes research areas such as computer vision, human-computer interaction, and artificial neural networks (ANN). Though the research groups at the Center for Visual Cognition (CVC), University of Copenhagen and the Intelligent Signal Processing group (ISP) at the department for Informatics and Mathematical Modeling (IMM) of the Danish Technical University (DTU) have collaborated within the field of cognitive neuroscience for several years [3, 4, 5, 6, 7] (cf Copenhagen Brain Research Center), the present proposal aims to open up an entirely new area of collaboration within the field of cognitive science. More specifically, the new research Center for Computational Cognitive Modeling will focus on describing and understanding cognitive systems and on developing artificial cognitive systems within a common mathematical framework.


Combining cognitive modeling and cognitive component analyzes will generate a better understanding of higher level cognitive systems

The simple fact that the physical world surrounding us is composed of separable objects has major implications for the structure of our cognitive system and the brain in which it is implemented. One of the most important consequences of this fact is that the unit of cognition is the object. We perceive, remember, attend, and think about objects – not the mosaic of impressions which they generate in our sense organs. At first, this may seem trivial; however the mechanisms which enable the cognitive system to extract knowledge about physical objects from the sense impressions are despite several decades of intense research still poorly understood. On the other hand, cognitive psychology has made significant progress in the understanding of how objects are represented and processed. One of the key findings has been that the capacity of the peripheral parts of our perceptual system is relatively large, enabling us to process information from several objects in parallel. However, we are severely limited in capacity when it comes to higher level cognitive processes. Thus when we have to consciously attend and think about several objects, we can only process about four objects at the same time! This latter constraint in capacity imposes a heavy pressure on our cognitive systems in both producing accurate and reliable object representations, and in attending and holding on to information from those objects which are most important for our current behavioral goals. The present project proposal aims to gain a better understanding of both 1) how efficient object representations are generated by the cognitive system using statistical methods such as independent component analysis (see COCA below) and 2) to describe and understand the cognitive mechanisms behind our ability to hold on to the information from the most pertinent objects in the outside physical world – our visual short-term memory.

Cognitive component analysis (COCA) is defined as the process of unsupervised grouping of data such that the ensuing group structure is well-aligned with that resulting from human cognitive activity [8]. In other words, COCA provides mathematical algorithms of how object or source segmentation may be achieved. In the present project, we wish to investigate the generality of the so-called independent component hypothesis: It is well documented that human perceptual systems can model complex multi-agent scenery. Human cognition uses a broad spectrum of cues for analyzing perceptual input and separates individual signal producing agents, such as speakers, gestures, affections, etc. – the units of cognition. Thus a key hypothesis in this proposal is that objects are defined in cognitive systems by their statistically independent behaviors. Unsupervised signal separation has also been achieved in computers using a variety of independent component analysis algorithms [9]. It is an intriguing fact that representations are found in human and animal perceptual systems which closely resembles the information theoretically optimal representations obtained by independent component analysis. We wish to go one step further and ask: Are such optimal representation rooted in independence also relevant in higher cognitive functions, such as thinking and problem-solving? Grouping of events or objects in categories is fundamental to human cognition. In machine learning, classification is a rather well-understood task when based on labeled examples [10]. Object clustering is a closely related unsupervised learning problem, in which we use general statistical rules to group objects, without a priori providing a set of labeled examples. It is a fascinating finding in many real world data sets that the label structure discovered by unsupervised learning closely coincides with labels obtained by letting a human or a group of humans perform classification, labels derived from human cognition. We speculate that the cognitive machinery developed for analyzing complex perceptual signals from multiagent environments may also be used in higher cognitive function, such as thinking, problem-solving, and social cognition, a hallmark of successful cognition. Thus the knowledge gained from COCA may potentially be very important in the understanding of how the input to the cognitive system is generated and of the architecture of higher level cognitive mechanisms.

Computer based cognitive models of short-term memory systems have become increasingly important in cognitive psychology over the last 10 years. Specifically, computer based models have made significant contributions to the testing and development of computational theories of perception, memory, and human problem-solving. The CVC has been a lead player in the field, resulting in the development of formal mathematical models of pattern recognition and visual attention [11, 12, 13, 14]. We intend to extend the research in formal models, working on models of visual short-term memory (VSTM), one of the central mechanisms of higher cognitive functions in humans. VSTM is defined as our immediate memory of ongoing visual events or objects in the environment. This type of memory, compared to visual long-term memory, is characterized by being flexible, yet severely limited in capacity (it only holds up to about four objects or events), only stable for a short period of time (usually seconds), and dependent on conscious control. In other words, visual short-term memory is the mechanism behind what we in daily life speak of as our conscious visual experience. We have already developed a preliminary artificial neural network model of VSTM [15,16] which will be further developed and investigated. In the present research program, we intend to extend our research effort into VSTM by developing a general model of short-term memory including the other senses apart from vision and short-term memory involved in higher cognitive processes such as thinking and problem-solving. Because the content of the short-term memory systems is the “raw data” for higher cognitive processes in general, the general capacity limit found in the short-term memory systems [17] also determines our capabilities in thinking and problem-solving [18, 19]. The present research program will shed light on these very limitations and on the architecture of the cognitive systems involved.


New experimental paradigms and application of advanced statistical methods such as ICA will yield better and more reliable models of higher cognitive functions

The separation of source signals from unknown noisy mixtures is an active signal processing research field referred to as blind source separation (BSS). Independent Component Analysis (ICA) is a BSS technique based on assumed statistical independence of the source signals. The technique of ICA is a relatively new invention. It was introduced for the first time in the early 1980s in the context of artificial neural network (ANN) modeling. In the mid-1990s new algorithms were introduced by several research groups, demonstrating impressive results on problems like the cocktail-party effect, where the waveforms of individual speech sources are found from their mixtures. Lars Kai Hansen will contribute with expertise in the field of statistical modeling such as ICA and ANN modeling where he is an international leading expert. He is head of the ISP, comprising three DTU-staff members, three postdocs, and sixteen PhD-Students. ISP researchers have specialized in design, evaluation and visualization of adaptive information systems, with bio-medical, multimedia, and industrial applications.
To investigate the structure of VSTM we will use well established behavioral paradigms within experimental cognitive psychology. These will include whole report, partial report, visual search and change detection. We will use the paradigms in combination with formal mathematical probabilistic modeling which will enable us to estimate model parameters from the observed data of the experiments [14]. Further, we wish to develop new behavioral paradigms and compare the reliability of model estimates computed on the bases of the new and the traditional paradigms. This method will enable us to optimize paradigm selection, thus increasing the efficiency of data acquisition. Søren Kyllingsbæk will contribute with expertise in modeling of cognitive processes in humans and development of behavioral experimental 4 paradigms. He been affiliated for serveral years to the CVC which has an international reputation for both its rigorous experimental work as well as being a leading research unit within mathematical modeling of cognitive processes. Combining the expertise in experimental psychology and modeling of cognitive processes with the advanced statistical methods such as ICA and ANN will provide both a deeper understanding of higher level cognitive processes in human cognition as well as the knowledge from which artificial cognitive systems may be developed.


Integrated cognitive models will be of major importance for cognitive science in Denmark and abroad as well as for the IT industry and society in general

The development of integrated computational models of higher cognitive functions involved in thinking and problem-solving will be of major importance to the research field of cognitive science both in Denmark as well as abroad. The research groups behind this proposal participate in research networks within the fields of computer vision, neuroinformatics, signal processing, and intelligent media research, both nationally and internationally,. The development of new cognitive models of human cognition and artificial cognitive models will be a major catalyst for research within the already established research networks. Moreover, we intend to transform the scientific innovations developed in the Center for Computational Cognitive Modeling into industrial applications. The knowledge of higher level cognitive processes could for example be used in the development of new and more efficient multimedia internet search engines, multimedia object segmentation algorithms, better software for human computer interfaces. The IPS already hosts a number of application oriented PhD projects with Danish Hospitals and companies such as Oticon, a/s, GN-resound, Nokia, MAN B&W, among others. Lars Kai Hansen’s part of the research is already funded by the Danish Research Council funded project “intelligent sound”.


References

1 Kahneman, D. (2002). Maps of bounded rationality: a perspective on intuitive judgment and choice. Nobel Price Lecture.
2 Gazzaniga, M. S. (2000). Cognitive Neuroscience (2nd ed). Cambridge, MA: MIT Press.
3 Hugdahl, K., Brønnick, K.,Kyllingsbæk, S.,Law, I., Gade, A., & Paulson, O. B. (1999). Brain activation during dichotic presentations of consonant-vowel and musical instrument stimuli: a 15O-PET study. Neuropsychologia, 37, 431-440.
4 Hugdahl, K., Law, I., Kyllingsbæk, S., Brønnick, K., Gade, A., & Paulson, O. B. (2000) Effects of attention on dichotic listening: An 15O-PET study. Human Brain Mapping, 10(2), 87-97.
5 Bundesen, C., Larsen, A., Kyllingsbæk, S., Paulson, O. B., & Law, I. (2002). Attentional effects in the visual pathways: a whole-brain PET study. Experimental Brain Research, 147, 394-406.
6 Humphreys, G. W., Kyllingsbæk, S., Watson, D. G., Olivers, C. N. L., Law, I., & Paulson, O. B. (2004). Parieto-occipital areas involved in efficient filtering in search: A time course analysis of visual marking using behavioural and functional imaging procedures. Quarterly Journal of Experimental Psychology: Section A, 57, 610-635.
7 Larsen, A., Kyllingsbæk, S., Law, I., & Bundesen, C. (2005). Activation in the MT-complex during visual perception of apparent motion and temporal succession. Neuropsychologia, 43, 1060-1071.
8 Hansen, L., K, Ahrendt, P., & Larsen, J. (2005). Towards Cognitive Component Analyzis. AKRR'05 - International and Interdisciplinary Conference on Adaptive Knowledge Representation and Reasoning, Pattern Recognition Society of Finland, Finnish Artificial Intelligence Society, Finnish Cognitive Linguistics Society.
9 Hyvarinen, A., Karhunen, J., & Oja, E. (2001) Independent Component Analysis. John Wiley & Sons.
10 Bishop, C., M. (1995). Neural Networks for Pattern Recognition. Oxford University Press, Oxford.
11 Larsen, A. & Bundesen, C.(1996). A Template Matching Pandemonium Recognizes Handwritten Characters With High Accuray. Memory & Cognition, 24, 136-143.
12 Bundesen, C. (1990). A theory of visual attention. Psychological Review, 97, 523-547.
13 Bundesen, C., Habekost, T., & Kyllingsbæk, S (2005). A neural theory of visual attention. Bridging cognition and neurophysiology. Psychological Review, 112, 291-328.
14 Kyllingsbæk, S. (in press). Modeling visual attention. Behavior Research Methods.
15 Kyllingsbæk, S. (2003). Variations in visual short-term memory capacity within individuals [Abstract]. In T. Bajo & J. Lupiánez (Eds.), XIII Conference of the European Society for Cognitive Psychology : book of abstracts (p. 259), Granada: Editorial Universidad de Granada.
16 Usher, M., & Cohen, J. D. (1999). Short term memory and selection processes in a frontal-lobe model. In D. Heinke, G. W. Humphreys, & A. Olson (Eds.), Connectionist models in cognitive neuroscience (pp. 78–91). Berlin, Germany: Springer-Verlag.
17 Cowan, N. (2001). The magical number 4 in short-term memory: a reconsideration of mental storage capacity. Behavioral and Brain Sciences, 24, 87-114.
18 Logan, G. D. (2002). An instance theory of attention and memory. Psychological Review, 109, 376-400.
19 Logan, G. D., & Gordon, R. D. (2001). Executive control of visual attention in dual-task situations. Psychological Review, 108, 393-434.