Thesis defense – Sze Ying Lam

Venue: Salle de réunion de l’INCIA and on Zoom:
https://u-bordeaux-fr.zoom.us/j/89717363335

Defense in english

Title: Empirical Validation of an Information Theoretic Model of Cognitive Effort

Abstract

The sensation of effort, from an evolutionary point of view, could be understood as a mechanism for signalling the expenditure of scarce resources and which allows their effi- cient allocation. Understanding the decision making processes that are involved in effort allocation is crucial if one is to gain insight into human behaviour. One type of effort that is observed and reported in humans, and is the central subject of this thesis, is cognitive effort. Although there is still no general consensus over the true nature of the resources that cognitive effort was developed to safeguard, its aversiveness and involvement in decision-making are widely agreed upon. The principle of least action, entailing the minimisation of effort, provides a rational account for seemingly sub-optimal behaviours. Nevertheless, there are major obstacles to overcome in studying cognitive effort, many of which are associated with complications and biases associated with the measurement of subjective experiences. In response to these limitations, some recent work has focused instead on the influence that these subjective experiences have over observable, free choices of engagement. Notably, a neuroeconomic approach was employed to establish preference functions that express cognitive effort costs and task rewards in a common currency. Following this line of research, an information theoretic model of cognitive effort is proposed in this thesis work. The motivation for such a model is three-fold. Firstly, the mathematical framework of information theory provides a natural common currency, that is information, for quantifying task difficulty, engagement and performance. This could provide a more direct interpretation of the relationship between task demand, effort expenditure and associated gains. Secondly, information theoretic measures derived from first principles set bounds on the information rate associated with automatic and controlled behaviours. Lastly, information theory provides the common framework in which the interpreta- tion of cognitive effort can be linked to well-established theories regarding computational efficiency in the brain such as efficient coding and/or predictive coding theorems. In this thesis work, a series of experiments were designed to validate the proposed model of cognitive effort. The main task used in these experiments is a continuous visual-motor tracking task with joystick control. In the first study, information theoretic measures representing information rate of the feed-back (controlled) and feed-forward (automatic) processing of the signal were derived from first principles and were validated through simulated tracking data from a linear quadratic regulator (LQR) model. These measures were subsequently applied to real tracking data to gain insight of their engage- ment in the task in terms of real-time information processing rate. The second study aims at investigating and comparing the effect that different task attributes, including signal speed, predictability and joystick delay have on feed-back and feed-forward information rate, as well as on performance. The third and fourth studies were dual-task experiments designed to investigate cross- task interactions in information rate and to infer global limits in the brain in terms of computational resources. Lastly, a model is built by modifying an intermittent controller to include an information bottleneck objective to provide a normative account of the cost/value trade-off in human tracking performance. This model is then applied to behavioural data to study the principles of allocation of information rate and the optimality of human motor control.

Keywords: cognitive effort, information theory, visuo-motor tracking, Dual-task

Publication

Lam S-Y, Zénon A. Information Rate in Humans during Visuomotor Tracking. Entropy. 2021; 23(2):228. https://doi.org/10.3390/e23020228

Jury

• Roshan Cools, Professor of Cognitive Neuropsychiatry, Radboud University Nijmegen
• Victoria Kostina, Professor of Electrical Engineering, California Institute of Technology
• Max Mulder, Professor Aerospace Human-Machine Systems, TU Delft
• Frederic Danion, CNRS Researcher (CeRCA/MSHS HDR), University of Poitiers
• Aymar De Rugy, CNRS Director of Research (DR2) HDR, University of Bordeaux