Computational and Biological Principles of Sensorimotor Learning
Zusammenfassung der Projektergebnisse
In this project, we have pursued a three-pronged approach: (A) test and develop quantitative models of structural motor learning, (B) test neuro-economic principles and models for sensorimotor interactions and learning, and (C) develop neuro-economic principles and models for learning and decisionmaking. Refining our previous results on structural learning, we found that structural motor learning in humans can be induced by gradual task variation and that the selection between structures can be described by Bayesian model selection and in particular Bayesian Occam’s Razor. In particular, we found that hierarchical Bayesian model that learn on different time scales can capture learning of statistical abstractions across motor tasks. More generally, we found that structural learning might be understood as a process of abstraction arising due to the fundamental trade-off between task utility and information, for example in the face of limited computational resources. Due to the analogy between utility and energy and computation cost and entropy, thermodynamics can be viewed as another manifestation of such a general theory of decision making with information processing costs. Known models of ambiguity and risk-sensitivity can be considered as special cases of this information-theoretic bounded rationality approach. Applications for synthetic learning systems include entropic regularization for generalization, specialization and division of labor, and the coupling between action and perception modules. When testing the information-utility trade-off in sensorimotor tasks with human subjects, we found that the previously reported Bayesian optimality of sensorimotor integration in humans is risksensitive and that cooperation in sensorimotor interactions is sensitive to ambiguity in line with our model predictions. As randomness plays a central role in entropic cost models, we found that the sensorimotor system can be trained to generate varying levels of randomness by training in pursuitevasion games. Finally, in line with previous results from behavioural economics and in line with our model predictions, we found that ambiguity in sensorimotor tasks is processed differently to known risk uncertainties and that sensorimotor interactions with private information can be understood within the framework of signalling games.
Projektbezogene Publikationen (Auswahl)
-
(2012) A sensorimotor paradigm for Bayesian model selection, Frontiers in Human Neuroscience 6:291
Genewein T, Braun DA
-
(2012) Risk-Sensitivity in Bayesian Sensorimotor Integration, PLoS Computational Biology 8(9) 1-7
Grau-Moya J, Ortega PA, Braun DA
-
(2013) The effect of model uncertainty on cooperation in sensorimotor interactions, Journal of the Royal Society Interface 10(87) 1-11
Grau-Moya J, Hez E, Pezzulo G, Braun DA
-
(2013) Thermodynamics as a theory of decision-making with information processing costs, Proceedings of the Royal Society A 469: 20120683
Ortega PA, Braun DA
-
(2014) Assessing randomness and complexity in human motion trajectories through analysis of symbolic sequences, Frontiers in Human Neuroscience 8(168) 1-13
Peng Z, Genewein T, Braun DA
-
(2014) Occam’s Razor in Sensorimotor Learning, Proceedings of the Royal Society B 281(1783) 1-7
Genewein T, Braun DA
-
(2015) A reward-maximizing spiking neuron as a bounded rational decision maker, Neural computation 27 (8), 1686-1720
Leibfried F, Braun DA
-
(2015) Bounded rationality, abstraction and hierarchical decision-making: an information-theoretic optimality principle, Frontiers in Robotics and AI 2, 27
Genewein T, Leibfried F, Grau-Moya J, Braun DA
-
(2015) Signaling equilibria in sensorimotor interactions, Cognition 141, 73-86
Leibfried F, Grau-Moya J, Braun DA
-
(2016) Decision-Making under Ambiguity Is Modulated by Visual Framing, but Not by Motor vs. Non-Motor Context. Experiments and an Information-Theoretic Ambiguity Model, PloS ONE 11 (4)
Grau-Moya J, Ortega PA, Braun DA