Symposia & Tutorials

Cognitive modeling in computer science and psychology: Bridging the gap (Symposium)

Rebecca Albrecht & Mikhail Spektor

Formal modeling approaches to human cognition is the cornerstone method of cognitive science. Its various sub-disciplines, including computer science, neuroscience, and psychology, rely on the computational perspective as a window to cognition. However, the methods and models that are being used as well as the goals of using cognitive modeling differ between them. For example, cognitive models in computer science rely on the assumption that cognitive processes are the result of general learning mechanisms that are able to find systematic patterns in unstructured data. These machine-learning methods span from low-level connectionist mechanisms to high-level logical representations. The former aim to describe basic cognitive and biological functions such as vision and motor action, whereas the latter describe higher level cognitive and executive functions, like reasoning and planning. In psychology, cognitive models are used in conjunction with behavioral data and fall on the continuum between the poles measurement models and process models. Process models explicitly formalize the assumed underlying cognitive processes such as attention, perception, or memory. These models are then evaluated relative to alternative models, rigorously selecting the processes that are essential and ruling out those that are not. In contrast, the mechanisms underlying measurement models are inherently agnostic with respect to the psychological processes they reflect. The associated model parameters gain psychological content through behavioral differences across experimental conditions. The question of how the different approaches to cognitive modeling may inform and benefit each other has been subject of discussion in the literature. The arguably most popular approach is the cognitive architecture ACT-R (Anderson, 2004, Psychol. Rev). ACT-R combines high-level rule representations with process assumptions about memory retrieval. However, ACT-R models suffer from various difficulties, including interpretation and falsifiability. The aim of the proposed symposium is to discuss different cognitive-modeling approaches from the various sub-disciplines of cognitive science, identify the overlaps between them, and critically reflect existing hybrid models. To do so, it comprises a total of five speakers from computer science and psychology, each representing a different aspect of the cognitive-modeling spectrum. The cornerstones of modeling techniques from computer science will be presented by Fabian Schrodt, introducing a neural-network model of social action understanding based on embodied simulation, and Ute Schmid, presenting a rule-learning framework with inductive programming as its basis. From psychology, Gidon Frischkorn will show the merits and limitations of using cognitive models as measurement tools and Mikhail Spektor will show an example of how process models are developed and evaluated. The symposium will conclude with a brief introduction about the ACT-R and will transition into an overarching discussion.

Speakers and talk titles:

  • Mikhail S. Spektor (University of Freiburg): Implementing value-based attentional capture in a computational process model of cognition
  • Gidon Frischkorn (University of Heidelberg): Using cognitive models as measurement tools: The appropriate representation of a person’s cognitive processes
  • Fabian Schrodt (University of Tübingen): A neurocomputational model of action understanding
  • Ute Schmid (University of Bamberg): Inductive programming: A generic approach to rule learning on the knowledge level
  • Rebecca Albrecht (University of Basel): Is ACT-R enough? The merits and flaws of hybrid cognitive architectures

Cognitive Technical Systems -- Towards Fluid Assistants? (Symposium)

Stefan Kopp & Ute Schmid

Modern Artificial Intelligence and Cognitive Systems are on the verge of penetrating the everyday life of human users and to free them of many tasks that are cumbersome, dangerous or exceed their abilities and resources. We are witnessing such systems being developed for, e.g., entertainment, healthcare, educational, or workplace settings, and they are being deployed by commercial players at an increasing pace. The goal is to assist users in their tasks and the systems attain a variety of roles and interaction paradigms in doing so, from responding to user instructions, to providing recommendations, engaging in negotiations, to carrying tasks or subtasks autonomously. In result, we expect to see a variety of integrated (“hybrid”) settings in which humans and technical systems come to collaborate in different ways in order to solve even time- or safety-critical tasks. However, and in spite of the apparent technology push, a number of crucial questions are far from being sufficiently understood or solved: How can cognitive systems recognize and represent the state of users adequately? How can assistants support their users in a non-distracting, unobtrusive way? How do systems need to adapt to the specific requirements of the user and the demands of a given situation and task? How can machine learning and computational cognitive science help to obtain deeper user models and policies for suitable assistance? How to evaluate and validate the acceptance and efficacy of such systems? How to ensure safety and reliability of human-machine systems in safety- critical environments? How can users understand and be kept aware of the abilities and limitations of an assistance system? Can systems themselves assess their current limitations and effects (supportive or harmful) and use this to choose a suitable assistive behavior? This symposium aims to bring together researchers from different disciplines, from Cognitive Science and Artificial Intelligence, Engineering and Control, to Psychology and Human Factors, to discuss the state-of-the-art in technical approaches to developing and applying cognitive systems for user assistance. It is thus closely related to KogWis 2018’s special focus on computational approaches to Cognitive Science. A special focus will be put on the notion of “fluid assistance” -- the ability of technical cognitive systems to uphold a deep understanding of the dynamically changing situation, task demands and the user’s internal states (cognitive, affective, or physiological) and to adapt flexibly and continuously the way in which to support the user. This vision, representing a next stage of user-adaptive assistive systems, combines and resolves the boundary between different roles, interaction modes and assistive effects a cognitive system can realize.

List of speakers:

  • Ute Schmid (Cognitive Systems Group, University of Bamberg): Explaining Classifier Decisions in an Interactive Learning Environment
  • Meike Jipp (Human Factors and Testing, Institute of Transportation Systems, DLR): User-state recognition as challenge for empathic assistants and automization
  • Andreas Wendemuth (Cognitive Systems, Institute for Information Technology and Communication, University of Magdeburg): Intelligent driver assistants
  • Dirk Söffker (Dynamics and Control, Fac. Of Engineering, University of Duisburg-Essen): Intention and option: Modelling and Recognition of human driver behavior
  • Stefan Kopp (Social Cognitive Systems, Center of Excellence Cognitive Interaction Technology, Bielefeld University): Cognitive Systems for deep and fluid assistance and collaboration

Multinodal Processing in the Visual System (Symposium)

Ralf Galuske

The aim of this symposium is to elucidate the functional and dynamical interactions between different centers in the mammalian visual system and to identify their relevance for information processing in the central nervous system.

  • Julien Vezoli (Ernst Strüngmann-Institute, Frankfurt/M, Germany): The Relation between Anatomical Connection Strength and Inter-areal Functional Connectivity through Rhythmic Synchronization
  • Ralf Galuske (Centre for Cognitive Science, TU Darmstadt, Darmstadt, Germany): Functional Topography of Cortical Feedback Connections in the Visual System
  • Miriam Müller (Ernst Strüngmann-Institute, Frankfurt/M, Germany): Revising the Interhemispheric Imbalance Model of Neglect
  • Ricardo Kienitz (Dept. of Neurology, Frankfurt University Medical School, Frankfurt/M, Germany): Theta Rhythmic Spiking and Attentional Sampling Arising from Cortical Receptive Field Interactions
  • Michael Wibral (Dept. of Psychiatry, Frankfurt University Medical School, Frankfurt/M, Germany): Neural Information Dynamics from Cells to Systems

Neural and Probabilistic Deep Learning (Tutorial)

Kristian Kersting

Our minds make inferences that appear to go far beyond standard machineclearning. Whereas people can learn flexible representations and use them for a wider range of learning tasks, traditional machine learning algorithms have been mainly employed in a rigid way, constructing a single function from a table of training examples. In this tutorial, I shall review deep learning approaches, a more flexible function approximation. Specifically, I will touch upon function approximators like convolutional neural networks that are robust and allow for real-time inference. However, they requiring fixed inputs and outputs and do not provide probabilities. Therefore I will also touch up upon Sum-product networks (SPNs). They are deep models that are suitable for both function approximation and probability estimation. Overall, I will review generative and discriminative deep learning approaches, both in a neural and (explicit) probabilistic fashion.

Bayesian Modeling (Tutorial)

Constantin Rothkopf & Frank Jäkel

The first part of the course is a basic introduction to probability theory from a Bayesian perspective. We will also discuss how Bayesian inference differs from frequentist inference. In the second part of the course we will discuss why Bayesian Decision Theory provides a good starting point for probabilistic models of perception and cognition. The focus here will be on Rational Analysis and Ideal Observer models that provide an analysis of the task, the environment, the background assumptions and the limitations of the cognitive system under study. We will go through several examples from signal detection to categorization to illustrate the approach.