Abstract
This chapter introduces an explanatory middleman, the concept of an attractor landscape. It is used to show what is an invariant of behavior: an abstract representation of a functional mechanism. With an experiment in the evolution of active tracking, recurrent neural networks are evolved that enable agents to track a moving object in a simulated environment. The invariant behavior, i.e., the functional mechanism implemented by agents, is a two-dimensional version of the well-known – from cybernetics – negative feedback, which subsumes an ample range of both simple and complex organismic functions. I will (1) show how, despite extreme variability in network structures, constancy of behavior reflects invariant features of the attractor landscapes, (2) show that behavioral function only exists as a potential, until it is evoked, and (3) show that even networks with radically different attractors may implement the same embodied function, as long as these networks possess certain invariant features of the attractor landscape. This chapter also addresses constancy arising from level crossing, convergence, as variable activity patterns have their dimensions effectively reduced at the level of actuators. I will show how attractors are made equivalent through convergence. How the ongoingness of behavior appears as the attractor landscape is explored through organism–environment interaction. This will lead to the concept of a “metatransient,” with respect to which we show effectively what it means for an organism, to implicitly represent the coupling between the organism and its environment.
This is a preview of subscription content, log in via an institution.
Buying options
Tax calculation will be finalised at checkout
Purchases are for personal use only
Learn about institutional subscriptionsNotes
- 1.
Parameterizations are meant in the manner that has become usual in the dynamical systems literature, in which parameters are variables that change slowly with respect to the dynamics of the network.
- 2.
It is helpful to be aware that “qualitatively distinct dynamic behavior” might lead to qualitatively similar agent behavior, which might be a source of ambiguity (see Sect. 8.4.6.2).
- 3.
The idea that readout units sample different aspects of the same attractor is analogous to the approach of liquid state machines or echo state networks (dynamic reservoir networks) [19], with the difference that their attractors are generated randomly to satisfy certain requirements for complex dynamics, whereas our attractors are incrementally evolved. This difference results from the role of the attractors, in which we think that there are attractors that are more apt to solve some kinds of problems, and that artificial evolution is a good method to beget them.
- 4.
This does not mean, however, that all the agent may reach every coexisting attractor during behavior, since the possible states are also bounded by the possible history of interactions.
- 5.
Consider when the ball is subject to gravity, with the velocity on the vertical axis, even if the head locks its position to the ball, thereby keeping a constant input pattern, it must nevertheless accelerate downward.
- 6.
The internal states of the hidden layer are inherited across parameterizations.
- 7.
These input patterns were obtained with the same network operating under 500-Hz update frequency, so the differences between steps would be tiny. Note that this test would not have been possible if the network had not been robust to different update frequencies.
- 8.
Note that this reflexive depiction of behavior does not need to mean it is stereotypical. In The Structure of Behavior, Merleau-Ponty [22] makes a case for the inadequacy of a physiological theory of behavior based on reflexes, since a quantitative change of the stimulus induces a qualitative change in behavior. But as we have seen in our case, even a small quantitative difference in stimulus may invoke a qualitatively different reaction, not bijective with the stimulus.
References
Arbib MA (1972) The metaphorical brain, an introduction to cybernetics and brain theory. MIT, Cambridge, CA
Ashby W (1960) Design for a brain: The origin of adaptive behavior, 2nd edn. Chapman & Hall, London
Banerjee A (2001) The roles played by external input and synaptic modulations in the dynamics of neuronal systems. Behav Brain Sci 24(5):811–812
Barandiaran X, Moreno A (2006) On what makes certain dynamical systems cognitive: A minimally cognitive organization program. Adaptive Behavior 14(2):171–185. DOI 10.1177/105971230601400208, URL http://adb.sagepub.com/cgi/content/abstract/14/2/171, http://adb.sagepub.com/cgi/reprint/14/2/171.pdf
Beer R (2009) Beyond control: The dynamics of brain-body-environment interaction in motor systems. In: Sternad D (ed) Progress in motor control V: A multidisciplinary perspective. Springer, New York
Beer R, Gallagher J (1992) Evolving dynamical neural networks for adaptive behavior. Adapt Behav 1(1):91–122
Beer RD (1995) A dynamical systems perspective on agent-environment interaction. Artif Intell (72):173–215
Berry H, Quoy M (2006) Structure and dynamics of random recurrent neural networks. Adapt Behav 14(2):129–137. DOI 10.1177/105971230601400204, URL http://adb.sagepub.com/cgi/content/abstract/14/2/129, http://adb.sagepub.com/cgi/reprint/14/2/129.pdf
Boeddeker N, Egelhaaf M (2005) A single control system for smooth and saccade-like pursuit in blowflies. J Exp Biol (208):1563–1572
Edelman GM (1989) The remembered present. Basic Books, New York
Freeman W (2000) Mesoscopic neurodynamics: From neuron to brain. J Physiol Paris 94(5–6):303–322
Harvey I, Paolo ED, Wood R, Quinn M, Tuci E (2005) Evolutionary robotics: A new scientific tool for studying cognition. Artif Life 11(1–2):79–98. URL http://www.mitpressjournals.org/doi/abs/10.1162/1064546053278991
von Holst VE, Mittelstaedt H (1950) Das Reafferenzprinzip. Die Naturwissenschaften 37(20):464–476
Homberg U, Paech A (2002) Ultrastructure and orientation of ommatidia in the dorsal rim area of the locust compound eye. Arthropod Struct Dev 30(4):271–280
Hülse M (2006) Multifunktionalität rekurrenter neuronaler netze – synthese und analyse nichtlinearer kontrolle autonomer roboter. PhD thesis, Universität Osnabrück
Hülse M, Ghazi-Zahedi K, Pasemann F (2002) Dynamical neural schmitt trigger for robot control. In: Dorronsoro JR (ed) ICANN, Springer, vol ICANN 2002, LNCS 2415, pp 783–788
Ijspeert AJ, Nakanishi J, Schaal S (2003) Learning attractor landscapes for learning motor primitives. In: Advances in neural information processing systems, MIT, Cambridge, CA
Ikegami T, Tani J (2002) Chaotic itinerancy needs embodied cognition to explain memory dynamics. Behav Brain Sci 24(05):818–819
Jaegger H, Maas W, Markram H (2007) special issue: Echo state networks and liquid state machines. Neural Netw 20(3):290–297
Kaneko K (1990) Clustering, coding, switching, hierarchical ordering, and control in network of chaotic elements. Physica D 41(37)
Kaneko K, Tsuda I (2003) Chaotic itinerancy. Chaos 13(3):926–936
Merleau-Ponty M (1963 (translation), 1942) The structure of behavior. Duquesne University Press, Philadelphia, PA
Molter C, Salihoglu U, Bersini H (2007) The road to chaos by time-asymmetric hebbian learning in recurrent neural networks. Neural Comput 19:80–110
Pasemann F (1993) Discrete dynamics of two neuron networks. Open Syst Inf Dyn 2(1):49–66
Philipona D, O’Regan J, Nadal J, Coenen OM (2004) Perception of the structure of the physical world using unknown multimodal sensors and effectors. Adv Neural Inf Process Syst 16:945–952
Rossel S, Corlija J, Schuster S (2002) Predicting three-dimensional target motion: how archer fish determine where to catch their dislodged prey. J Exp Biol 205(21):3321–3326. URL http://jeb.biologists.org/cgi/content/abstract/205/21/3321, http://jeb.biologists.org/cgi/reprint/205/21/3321.pdf
Sterelny K (2005) Thought in a hostile world. MIT, Cambridge, CA
Tani J (1998) An interpretation of the ‘self’ from the dynamical systems perspective: A constructivist approach. J Conscious Stud 5(5–6):516–542
Taylor C (1999) The atomists leucippus and democritus: Fragments: A text and translation. University of Toronto Press, Toronto
Tsuda I (1991) Chaotic itinerancy as a dynamical basis of hermeneutics in brain and mind. In: Microcomputers and attention. Manchester University Press, Manchester
Tsuda I (2001) Toward an interpretation of dynamic neural activity in terms of chaotic dynamical systems. Behav Brain Sci 24:793–847
von Uexküll J (1934) Bedeutungslehre / Streifzüge durch die Umwelten von Tieren und Menschen, 1956th edn. Rowohlt Hamburg
Varela F (1979) Principles of biological autonomy. North Holland, New York
Varela F, Maturana H, Uribe R (1974) Autopoiesis: The organization of living systems, its characterization and a model. Curr Model Biol 5(4):187–96
Von Foerster H (2003) Understanding understanding: Essays on cybernetics and cognition. Springer, New York
Author information
Authors and Affiliations
Corresponding author
Rights and permissions
Copyright information
© 2011 Springer Science+Business Media, LLC
About this chapter
Cite this chapter
Negrello, M. (2011). Attractor Landscapes and the Invariants of Behavior. In: Invariants of Behavior. Springer Series in Cognitive and Neural Systems, vol 1. Springer, New York, NY. https://doi.org/10.1007/978-1-4419-8804-1_8
Download citation
DOI: https://doi.org/10.1007/978-1-4419-8804-1_8
Published:
Publisher Name: Springer, New York, NY
Print ISBN: 978-1-4419-8803-4
Online ISBN: 978-1-4419-8804-1
eBook Packages: Biomedical and Life SciencesBiomedical and Life Sciences (R0)