The workshop will take place at the 4th Multidisciplinary Conference on Reinforcement Learning and Decision-Making (RLDM 2019) at McGill University in Montréal, QC, Canada, on July 10, 2019, from 1.00pm till 5.00pm (with a break between 3.10pm till 3.30pm). Each talk will last 25 min + 5 min questions.
[1.00-1.10pm] Introduction
Characterizing the origin of variability in learning and decision-making
The first part of the workshop aims at characterizing the different sources of variability in learning and decision-making. First, Anne will show how choice history actively shapes evidence accumulation during perceptual decision-making. Konstantinos will then explain how selective evidence integration can outperform perfect evidence accumulation in the presence of late noise. Valentin will present evidence that such late, "inference" noise is a significant feature of both perceptual and reward-guided decisions. Finally, Nicola will describe how autism-associated genotypes interact with sex differences in mice to explain strategic variability in reward-guided learning.
[1.10-1.40pm] Anne Urai, Cold Spring Harbor Laboratory, USA
[1.40-2.10pm] Konstantinos Tsetsos, University Medical Center Hamburg-Eppendorf, Germany
[2.10-2.40pm] Valentin Wyart, Inserm – Ecole Normale Supérieure, Paris, France
[2.40-3.10pm] Nicola Grissom, University of Minnesota, USA
Building resource-efficient theories of learning and decision-making
The second part of the workshop aims at deriving resource-efficient theories of learning and decision-making which can explain observed sources of variability and suboptimality (identified in the first part of the workshop) in terms of normative computations. First, Rahul will show how context dependence is adaptive from an "efficient-coding" perspective in the presence of internal noise. Rafael will then describe how optimal decision behavior in capacity-limited systems can be achieved via discrete, noisy samples which optimize the encoding of environmental regularities. Falk will close the workshop by presenting a computational theory for deriving rational heuristics from the finite time and limited cognitive resources available to decision-makers.
[3.30-4.00pm] Rahul Bhui, Harvard University, USA
[4.00-4.30pm] Rafael Polania, ETH Zürich, Switzerland
[4.30-5.00pm] Falk Lieder, Max Planck Institute for Intelligent Systems, Tübingen, Germany