Sich selbst entwickelndes, adaptives Verhalten in künstlichen kognitiven Lernsystemen basierend auf selbstorganisierenden, sensomotorischen Körperwelten
Zusammenfassung der Projektergebnisse
This project had the goal to model the ideomotor principle of learning from psychology by means of neuro-cognitive, self-organizing learning architectures. First, conceptual and theoretical work on anticipatory behavior has been conducted, which shows why, when, where, and in what way anticipatory mechanisms shape and optimize brain development. Second, several computational models have been developed, including SURE REACH (Sensorimotor Unsupervised REdundancy REsolving ArCHitecture) for flexible reaching, the XCSF classifier system for flexible directional arm control, and MMF (Modular Modality-Frame Model) for distributed bodily representations. Moreover, the TGNG (time-growing neural gas) algorithm was developed and combined with motivational drives based on principles of homeostasis, resulting in the autonomous generation of goal-directed behavior. All systems essentially develop sensorimotor-grounded, spatial encodings, and spatial mappings suitable for the maintenance of robust bodily representations and/or for flexible, adaptive, redundancy-resolving behavioral control. Meanwhile, close relations to predictive coding and Bayesian information processing have been revealed. Third, also dynamic neural systems were investigated, which are currently being integrated into the sensorimotor-grounded spatial encodings. Fourth, a large set of cognitive psychological experiments was conducted. These experiments showed that multisensory sources of information indeed interact highly flexibly and across different, bodily-grounded frames of reference. Moreover, the experiments showed that our behavior is highly goal-oriented typically taking even subsequent goals into consideration. However, the behavior is not necessarily fully optimal seeing that it can be influenced by habitual routines and even simple sensory biases. Finally, the psychological experiments have confirmed that the brain indeed strives to maintain an overall consistent body image, as was hypothesized by the MMF model. Fifth, during the project lifetime I could establish myself and my group in the cognitive science community, as indicated, for example, by the successful organization of the biannual German Cognitive Science meeting in Tübingen in September 2014. Many of the implications gained in this project fit very well and have partially contributed to the current developments of predictive encoding, the free energy principle, and the enactive, pragmatic turn in the cognitive sciences. It is my strong believe that these insights will help to understood the structure of concepts in our brains, including their development. Once we have reached this understanding, it will be inevitably possible to develop the next generation of intelligent systems, opening up currently unimaginable options for really smart devices, including self-organizing factories, adaptive production lines, innovative software tools, and intelligent, adaptive, and supportive agents in hard- and software.
Projektbezogene Publikationen (Auswahl)
- (2008). Function approximation with XCS: Hyperellipsoidal conditions, recursive least squares, and compaction. IEEE Transactions on Evolutionary Computation, 12, 355-376
Butz, M. V., Lanzi, P. L., and Wilson, S. W.
- (2008). How and why the brain lays the foundations for a conscious self. Constructivist Foundations, 4, 1-42
Butz, M. V.
- (2010). Planning and control of hand orientation in grasping movements. Experimental Brain Research, 202, 867-878
Herbort, O. and Butz, M. V.
- (2010). Remapping motion across modalities: Tactile rotations influence visual motion judgments. Experimental Brain Research, 207, 1-11
Butz, M. V., Thomaschke, R., Linhardt, M. J., and Herbort, O.
- (2010). Self-organizing sensorimotor maps plus internal motivations yield animal-like behavior. Adaptive Behavior, 18, 315-337
Butz, M. V., Shirinov, E., and Reif, K. L.
- (2011). Habitual and goal-directed factors in (everyday) object handling. Experimental Brain Research, 213, 371-382
Herbort, O. and Butz, M. V.
- (2012). Balanced echo state networks. Neural Networks, 36, 35-45
Koryakin, D., Lohmann, J., and Butz, M. V.
- (2012). Resource management and scalability of the XCSF learning classifier system. Theoretical Computer Science, 425, 126-141
Stalph, P. O., Llorà, X., Goldberg, D. E., and Butz, M. V.
(Siehe online unter https://doi.org/10.1016/j.tcs.2010.07.007) - (2013). Modeling the temporal dynamics of visual working memory. Cognitive Systems Research, 24, 80-86
Lohmann, J., Herbort, O., and Butz, M. V.
- (2013). Modular neuron-based body estimation: Maintaining consistency over different limbs, modalities, and frames of reference. Frontiers in Computational Neuroscience, 7, 148
Ehrenfeld, S., Herbort, O., and Butz, M. V.
(Siehe online unter https://doi.org/10.3389/fncom.2013.00148) - (2013). Separating goals from behavioral control: Implications from learning predictive modularizations. New Ideas in Psychology, 31, 302-312
Butz, M. V.
- (2013). The modular modality frame model: continuous body state estimation and plausibility-weighted information fusion. Biological Cybernetics, 107, 61-82
Ehrenfeld, S., Herbort, O., and Butz, M. V.
(Siehe online unter https://doi.org/10.1007/s00422-012-0526-2) - (2014). Filtering sensory information with XCSF: Improving learning robustness and robot arm control performance. Evolutionary Computation, 22, 139-158
Kneissler, J., Stalph, P., Drugowitsch, J., and Butz, M. V.
(Siehe online unter https://doi.org/10.1162/EVCO_a_00108) - (2014). Rubber hand illusion affects joint angle perception. PLoS ONE, 9, e92854
Butz, M. V., Kutter, E. F., and Lorenz, C.
(Siehe online unter https://doi.org/10.1371/journal.pone.0092854)
