The human brain is a complex network of interacting computational units. Decades of experimental and theoretical research have aimed at understanding the nature of these interactions and how they underlie behavior and cognition. Arguably, neuroscience already possesses a diverse set of conceptual, mathematical, and experimental tools to tackle the question of ‘who is doing what in the brain?’ However, such a mechanistic understanding is still not reached, as the debate over the essence of ‘causation’ and ‘mechanism’ in brain-behavior relationships is ongoing.

In this one-day hybrid symposium, we will bring together philosophers as well as computational and experimental neuroscientists to stimulate this debate. We will begin with philosophical ‘frameworks’ to derive a clearer picture of the conceptual foundations of causal explanations. We then turn to ‘applications’ to grasp how computational neuroscience models can provide mechanistic explanations. Finally, presentations by experimental neuroscientists will outline the status, challenges, and outlook of causal mechanisms provided by current experimental work and data.

The planned presentations aim to provide a broad view of how different approaches in neuroscience - and science in general - understand causal neural mechanisms and how they uncover causal relationships among brain regions and between the brain and behavior.

The symposium will take place on Fri, 2nd Dec 2022, 3 pm-8 pm CET at the Institute of Computational Neuroscience (ICNS), University Medical Center Hamburg-Eppendorf, Germany, in a hybrid format. We will happily accommodate up to 30 participants on-site at the ICNS (first come, first served), while online speakers and audiences can join via Zoom. In both cases, please register below:

Confirmed Speakers

Matthew Braham

Professor of Practical Philosophy, University of Hamburg, Hamburg, Germany

Konrad P. Kording

Penn Integrated Knowledge Professor, University of Pennsylvania, Pennsylvania, United States

Beate Krickel

Professor for Philosophy of Cognition, Institute of History and Philosophy of Science, Technology, and Literature, Technical University Berlin, Berlin, Germany

Stefano Panzeri

Director of the Department of Excellence for Neural Information Processing, University Medical Center Hamburg, Hamburg, Germany

Daniele Marinazzo

Professor of Neuroimaging Data Analysis, Ghent University, Ghent, Belgium

Jan Gläscher

Principal Investigator of the Valuation and Social Decision-Making group, Institute for Systems Neuroscience, University Medical Center Hamburg, Hamburg, Germany

Preliminary Program

15:00-15:30 Konrad P. Kording: Causal vs. Non-causal Approaches to Brain Function.

15:30-15:45 Q&A

15:45-16:15 Matthew Braham: What Philosophers Have to Say About Causation.

16:15-16:30 Q&A

16:30-17:00 Beate Krickel: Can Mechanistic Explanations of Cognition be Computational or Representational?

17:00-17:15 Q&A

17:15-17:45 Jan Gläscher: Learning from and Communicating with Expectancy Violations.

17:45-18:00 Q&A

18:00-18:30 Daniele Marinazzo: Disentangling Mechanisms and Behavior in Complex Systems.

18:30-18:45 Q&A

18:45-19:15 Stefano Panzeri: Contributions of Neuron-to-neuron Interactions to Perceptual Decision-making.

19:15-19:30 Q&A

19:30-20:00 Wrapping up and plenary discussion

(all times are in Central European Time (CET))

Available Abstracts

Can Mechanistic Explanations of Cognition be Computational or Representational? By Beate Krickel

Cognitive neuroscientists offer different types of explanations of cognition. This talk focuses on (1) mechanistic, (2) computational, and (3) representational explanations of cognition. Mechanistic explanations describe how neurons, networks of neurons, or brain regions interact to produce a particular cognitive or behavioral phenomenon. Computational explanations rely on abstract mathematical models of cognitive processes that specify the rules by which neural processing (may) operate. Representational explanations detail which neural activity patterns represent which items in the world. Often, these types of explanation appear in one and the same explanatory text (e.g., in research publications). This raises the question of how these types of explanation are related. According to proponents of the new mechanistic approach to explanation, the three types of explanation are not really different. Rather, computational and representational explanations are in fact special versions of mechanistic explanations. I will present a challenge for this view—the Compatibility Challenge—which seems to show that mechanistic explanations cannot be computational or representational. I will discuss strategies for dealing with the challenge and potential consequences for cognitive neuroscience.

Learning from and Communicating with Expectancy Violations By Jan Gläscher

Expectancy violations have long been recognized as the driving force behind learning and its computational equivalent, the prediction error (PE), represents the teaching signal in most contemporary learning theories. These theories continuously adjust expectancies in order to improve predictions about future events thus driving down the prediction error in the long run. An alternative view of expectancy violation revolves around their role in communicating salient information. The ensuing surprise marks important states and recruits additional processing resources. These general principles of expectancy violations can thus be exploited during communication in novel context, when a common dictionary has not be established yet. I will present two recent studies from my lab exemplifying these principles. Using a novel Markov decision task that elicits orthogonal reward and state prediction errors, we are able to show how these PEs predict multivariate pattern changes in distinct brain regions that in turn explain changes in behavioral policy due to learning. I will then present data form a study using the Tacit Communication Game, in which the Sender designs spatial trajectories on a grid-like game board to communicate a goal position the Receiver. Building on earlier modeling efforts we developed a novel computational model that rests on the idea that the Sender deliberately uses expectancy violations from the Receiver’s perspective to communicate the goal position. Taken together, these divergent principles of minimization and maximization of expectancy violations have different implications for the organization of cognition and behavior.

The Venue

Institute of Computational Neuroscience, University Medical Center Hamburg-Eppendorf (UKE), Hamburg, Germany

Address:

Building W36

Martinistraße 52

20246 Hamburg

Organized by:

Kayson Fakhar (k.fakhar[at]uke.de)

Mariia Popova (m.popova[at]uke.de)

Claus C. Hilgetag (c.hilgetag[at]uke.de)