Complex adaptive systems across physics, biology, and cognitive science constantly organize chaotic fluctuations into coherent, goal-directed behaviors. Attractor dynamics provides the mathematical and conceptual architecture necessary to explain how such multidimensional systems settle into stable, self-sustaining states over time. By modeling behavior as trajectories moving through a high-dimensional state space toward preferred regions, attractor dynamics bridges the gap between microscopic physical interactions and macroscopic functional order.
Attractor Dynamics
1. Concise Definition
Attractor dynamics refers to the mathematical and theoretical framework describing how the state of a dynamical systems theory evolves over time toward a specific state, set of states, or geometric manifold known as an attractor. Once a system’s trajectory enters the basin of attraction governing such an attractor, it asymptotically converges toward it and resists small perturbations, maintaining stable functional configurations despite ongoing noise.
In cognitive science, computational neuroscience, and developmental psychology, attractor dynamics characterizes how recurrent interactions among components—such as interconnected neural populations or coordinated musculoskeletal assemblies—produce persistent memories, stable perceptual categorizations, stereotypic motor synergies, and discrete emotional states. Rather than conceptualizing behavior as a linear chain of input-output processing, attractor dynamics frames cognitive and biological operations as continuous movements across an energetic or topological landscape defined by peaks, valleys, and transitions.
2. Etymology & Linguistic Origin
The term is a compound of the English noun “attractor” and the plural noun “dynamics.” “Attractor” derives from the classical Latin attrahere (compounded from ad-, meaning “to” or “toward,” and trahere, meaning “to draw” or “to pull”). The suffix -or denotes an agent or instrument that performs the pulling. While used colloquially in early physics to denote gravitational or magnetic pull, the formal mathematical use of “attractor” emerged in the late 19th and mid-20th centuries within qualitative dynamics and topology.
“Dynamics” stems from the ancient Greek dynamikos (δυναμικός), meaning “powerful” or “pertaining to force,” which itself derives from dynamis (δύναμις, “power” or “ability”). The term was introduced into early modern natural philosophy by Gottfried Wilhelm Leibniz in 1695 to formalize the study of bodies in motion and the forces governing them. The synthesis of both terms into “attractor dynamics” crystallised during the mid-20th century, particularly within qualitative differential equations, non-equilibrium thermodynamics, and cybernetics, before being imported into computational biology by figures such as René Thom, Hermann Haken, and John Hopfield.
3. Pronunciation & Grammatical Form
Pronunciation: Attractor dynamics is phonetically transcribed in the International Phonetic Alphabet (IPA) as /əˈtræk.tər daɪˈnæm.ɪks/ (American English: [əˈtræk.tɚ daɪˈnæm.ɪks]).
Grammatical Form: The term functions as a compound noun phrase. “Attractor” acts as a noun adjunct modifying the plural noun “dynamics.” While “dynamics” takes a plural form morphology, it frequently governs a singular verb agreement when referring to the scientific paradigm or mathematical field (e.g., “Attractor dynamics provides a framework…”), but takes plural agreement when referring to the specific behavioral interactions within a system (e.g., “The attractor dynamics of this network are characterized by high multistability”). Derivative adjectival forms include “attractor-based” and “attractor-driven.”
4. Detailed Conceptual Explanation
To understand attractor dynamics, one must first operationalize the construct of phase space (or state space). A phase space is an abstract mathematical manifold in which all possible instantaneous configurations of a system are represented as distinct points. Each dimension or coordinate of this space corresponds to an independent state variable of the system, such as the membrane potentials of individual neurons, the angles and velocities of limb joints, or the concentrations of interacting neurochemicals. As time advances, the state of the system traces a continuous trajectory through this high-dimensional coordinate space, governed by deterministic or stochastic equations of motion.
An attractor represents a bounded subset of this phase space toward which neighboring trajectories asymptotically converge as time approaches infinity. Crucially, an attractor is invariant under the system’s dynamics: once a trajectory lands upon the attractor, it remains bound to that subset unless an external perturbation or parameter change displaces it. The geometric region of phase space containing all initial conditions that eventually evolve into a given attractor is defined as its “basin of attraction.” Multiple attractors can coexist within the same state space, producing a multistable regime where the ultimate fate of the system depends on its initial starting conditions and the intervening perturbations.
The concept is often visualized metaphorically via an “energy landscape” or potential surface. Imagine a topography of undulating hills and valleys across which a sphere rolls under the influence of gravity and friction. The valleys or depressions represent attractors: any marble placed on an adjacent slope will roll downward and settle at the lowest local point (the local minimum). The ridges separating adjacent valleys act as “separatrices,” delineating the boundaries of distinct basins of attraction. In non-conservative, dissipative biological systems, trajectories lose energy (or reduce computational error), compelling the high-dimensional degrees of freedom to collapse into lower-dimensional, highly organized manifolds.
In cognitive and biological systems, this mechanism confers stability and fault tolerance. In biological neural circuits, noisy sensory inputs or partial cues initiate trajectories that fall somewhere within an attractor basin. The intrinsic recurrent connectivity of the network pulls the trajectory toward the nearest attractor state, effectively filtering out noise, completing degraded patterns, and establishing a persistent mental representation. Consequently, attractor dynamics provides an intuitive physical substrate for phenomena that require sustained persistence over time, such as working memory, attentional focus, and categorical decision boundaries.
5. Historical Development
The origin of attractor dynamics traces back to the qualitative theory of differential equations pioneered by Henri Poincaré in the late 19th century. Poincaré shifted mathematical focus away from seeking exact, analytical solutions for complex multi-body problems (such as the three-body problem in celestial mechanics) toward analyzing the global geometric properties of trajectories in phase space. His work revealed that non-linear equations could generate stable closed orbits (limit cycles) and complex, non-repeating trajectories.
In the 1920s and 1930s, electrical engineer Balthasar van der Pol and mathematician Aleksandr Andronov expanded Poincaré’s qualitative methods to study relaxation oscillations in triode circuits and mechanical systems. Andronov formally introduced the concept of limit cycles into control theory, laying the empirical foundation for self-sustaining biological rhythms. Concurrently, Conrad Hal Waddington formulated his famous “epigenetic landscape” metaphor in the 1940s, positing that cellular differentiation operates akin to a ball rolling down branching developmental valleys toward terminal cell fates—a conceptual precursor to biological attractor dynamics.
During the 1960s and 1970s, meteorologist Edward Lorenz discovered chaotic attractors while modeling atmospheric convection, demonstrating that simple deterministic systems could produce bounded, non-periodic, fractal attractors sensitive to initial conditions. Around the same time, mathematician René Thom introduced catastrophe theory, which formalized how continuous shifts in control parameters could trigger sudden, discontinuous topological jumps between competing attractors. Hermann Haken unified these insights under the banner of “synergetics,” analyzing self-organization and phase transitions in lasers and physical media via order parameters.
The paradigm entered cognitive science and computational neuroscience during the 1980s. John Hopfield published his landmark 1982 paper introducing the Hopfield network, proving that symmetrically coupled recurrent artificial neural networks possess an energy function (Lyapunov function) that guarantees convergence to discrete point attractors, thereby implementing associative content-addressable memory. Simultaneously, J.A. Scott Kelso, Hermann Haken, and Gregor Schöner developed the Haken-Kelso-Bunz (HKB) model, applying non-linear attractor dynamics to human motor coordination and revealing bifurcations during phase transitions in rhythmic finger movements. By the 1990s, Esther Thelen and Linda Smith extended this framework to infant motor development, solidifying attractor dynamics as a cornerstone of embodied and situated cognitive science.
6. Theoretical Foundations
Attractor dynamics is rooted in several interconnected theoretical frameworks, the most fundamental of which is classical Non-linear Dynamical Systems Theory (DST). DST asserts that natural phenomena are best understood through sets of coupled differential or difference equations where outputs are non-proportional to inputs. Because living systems are non-linear, small changes in context or physiological parameters can induce profound shifts in behavioral output, known as bifurcations, wherein attractors appear, disappear, or alter their topological stability.
The framework also relies heavily on Non-equilibrium Thermodynamics, as developed by Ilya Prigogine. Living organisms are open dissipative systems that continuously exchange matter and energy with their environments. To maintain internal order and stave off entropic decay, biological systems dissipate energy, which naturally contracts their phase space volume. This mathematical contraction of phase space volume is the formal physical condition that enables the formation of attractors, translating thermodynamic dissipation into stable macroscopic biological patterns.
From the perspective of Connectionism and Recurrent Neural Network (RNN) theory, attractor dynamics formalizes cognitive representation. Traditional classical computational models treated mental representations as static, symbolic tokens manipulated via algorithmic rules. Connectionism, grounded in attractor concepts, conceptualizes a representation as a dynamic state of collective neural activity. Synaptic weights, sculpted by learning mechanisms such as Hebbian plasticity, shape the recurrent connectivity matrix, carving valleys into the network’s energy landscape. Perception becomes the process of a stimulus nudging the network into a basin of attraction; memory recall is the autonomous descent toward the attractor minimum.
Finally, Ecological Psychology—originated by James J. Gibson and elaborated by Edward Reed and Michael Turvey—converges with attractor dynamics to reject computational mentalism. In the ecological framework, perception and action form a continuous, closed-loop cycle. Rather than maintaining internal, detached models of the world, animals and their environments constitute a coupled dynamical system where behavioral affordances manifest as temporary attractors in the action-perception manifold.
7. Key Components, Types & Dimensions
Attractors in biological and cognitive systems are categorized by their geometric topology and behavioral consequences in phase space:
- Fixed-Point Attractors (Point Attractors): The simplest form of attractor, consisting of a single equilibrium point in state space. Trajectories entering its basin settle into a static, unchanging configuration. In neuroscience, point attractors model the stable maintenance of discrete items in working memory, categorical choices in perceptual decision-making, and static postural positions in motor control.
- Limit Cycles (Periodic Attractors): Closed, continuous orbital paths in state space that repeat periodically over time. Perturbations displacing the trajectory away from the orbit are corrected, returning the system to its cyclical trajectory. Limit cycles underpin autonomous, rhythmic biological behaviors such as cardiac pacemaking, respiratory rhythms, neural oscillations (e.g., theta and gamma bands), and locomotor gaits (walking, running).
- Torus Attractors (Quasi-Periodic Attractors): Geometric structures shaped like a doughnut surface, produced when a system exhibits two or more incommensurate (linearly independent) periodic frequencies. The trajectory wraps continuously around the surface of the torus without ever precisely intersecting its own path, representing complex multi-frequency biological coordination.
- Strange (Chaotic) Attractors: Complex, fractal geometric objects characterized by deterministic chaos. While trajectories remain bounded within a finite region of phase space, they exhibit extreme sensitivity to initial conditions (positive Lyapunov exponents). Neighboring paths diverge exponentially over time, rendering long-term prediction impossible while maintaining a globally constrained, deterministic structure (e.g., the Lorenz attractor). Chaotic dynamics have been observed in olfactory sensory processing, cardiac variability, and electroencephalographic (EEG) background fluctuations.
- Continuous (Line and Ring) Attractors: Manifolds of infinite, continuously connected fixed points that share identical energy levels, forming a neutral line, plane, or ring in phase space. The system can freely drift along the attractor manifold in response to internal integration signals while remaining intensely resistant to perturbations orthogonal to the manifold. Ring attractors represent the canonical neural architecture underlying head-direction cells, spatial navigation grids, and continuous parametric working memory (such as remembering a specific color shade or spatial angle).
- Basin of Attraction and Separatrix: The basin defines the full catchment domain of an attractor. The separatrix represents the mathematical boundary or ridge separating two or more distinct basins of attraction. The height and steepness of the separatrix govern the stability, resilience, and transition probability between alternative states.
8. Examples & Illustrative Cases
A classic empirical demonstration of attractor dynamics in human behavior is the Haken-Kelso-Bunz (HKB) finger-coordination experiment. When human participants are instructed to oscillate their index fingers rhythmically, two stable attractor states exist at low cycling frequencies: parallel (in-phase) motion and symmetrical (anti-phase) motion. As an experimenter increases the oscillation pacing via a metronome, the anti-phase attractor becomes unstable. At a critical threshold frequency, the system undergoes a sudden bifurcation (a non-equilibrium phase transition), involuntarily snapping the participant’s fingers into the highly stable in-phase attractor pattern. This switch occurs without conscious executive intervention, illustrating how macroscopic motor coordination self-organizes according to attractor stability landscapes.
In computational psychiatry, major depressive disorder has been modeled as an abnormally deep, hyper-stable fixed-point attractor. In an individual suffering from clinical depression, recurrent negative cognitive patterns and self-referential rumination correspond to a steep potential well in neural activity space. Healthy individuals possess a flexible, dynamic landscape where transient environmental stressors push neural trajectories into negative emotional states, but endogenous regulatory mechanisms easily bounce the state back out into neutral or positive basins. In chronic depression, the basin of the negative state is so profound and steep that standard environmental rewards or cognitive reappraisals lack the energetic momentum needed to cross the separatrix, trapping the patient in persistent anhedonia.
Another illustrative case is observed in rodent spatial navigation via head-direction (HD) cells. HD cells, located in regions like the postsubiculum and anterodorsal thalamic nucleus, fire maximally when an animal’s head points in a specific direction. Neurologically, these neurons are arranged in a ring attractor architecture. Excitatory connections couple neurons with similar preferred directions, while global inhibitory interneurons suppress distant neurons. This recurrent connectivity creates a sustained “bump” of activity on the ring manifold. As the animal rotates its head, vestibular and visual inputs continuously shift the location of this activity bump along the ring, preserving a continuous, drift-resistant internal compass.
9. Measurement & Assessment
Quantifying attractor dynamics in empirical, noisy biological data requires specialized mathematical techniques from non-linear time series analysis:
- Phase Space Reconstruction (Delay Embedding): In biological experiments, researchers rarely possess the capacity to measure every relevant state variable simultaneously. According to Takens’ embedding theorem, the multi-dimensional geometric structure of an underlying attractor can be topologically reconstructed from a single observed scalar time series (such as an EEG channel or movement velocity trace) by creating time-lagged copies of the signal: [x(t), x(t + τ), x(t + 2τ), …, x(t + (m-1)τ)], where τ is the time delay and m is the embedding dimension.
- Recurrence Quantification Analysis (RQA): RQA is an analytical method designed to identify recurring patterns, laminarity, and structural determinism in dynamic time series. It maps a signal onto a two-dimensional recurrence plot, where points represent times at which the system revisits approximately identical coordinates in phase space. RQA yields quantitative metrics such as percent determinism, recurrence rate, and entropy, revealing how rigidly an empirical system adheres to an attractor.
- Largest Lyapunov Exponent (LLE): The LLE measures the average exponential rate of divergence or convergence of infinitesimally close trajectories in phase space. A negative LLE indicates convergence toward a stable fixed-point or limit-cycle attractor; an LLE of zero indicates a neutral limit cycle or continuous manifold; a positive LLE provides definitive mathematical evidence of a chaotic attractor, signaling local unpredictability alongside global boundedness.
- Neural Latent Space Dimensionality Reduction: In contemporary systems neuroscience, multielectrode arrays record hundreds or thousands of single units simultaneously. Dimensionality reduction algorithms—such as Principal Component Analysis (PCA), Gaussian Process Factor Analysis (GPFA), and autoencoders—project massive neuronal firing vectors onto low-dimensional latent manifolds. Researchers then fit dynamical systems equations to these latent trajectories to directly extract vector fields, fixed points, saddles, and energy barriers.
10. Applications & Practical Significance
The applications of attractor dynamics span a vast interdisciplinary spectrum. In clinical neuroscience and computational psychiatry, attractor models explain drug addiction, obsessive-compulsive disorder (OCD), and epilepsy. Addiction is conceptualized as the pathological deepening of an incentive-salience attractor basin: drug cues reliably channel neural activity into an inescapable craving state. In epilepsy, seizures represent a catastrophic transition from normal, complex, high-dimensional neural chaos into an abnormally rigid, hyper-synchronized limit-cycle attractor. Therapeutic interventions such as Deep Brain Stimulation (DBS) or transcranial magnetic stimulation (TMS) are designed to provide targeted energetic “kicks” that disrupt pathological attractors and nudge neural trajectories back into healthy, flexible regimes.
In robotics and embodied artificial intelligence, attractor dynamics replaces brittle, pre-programmed computational look-up tables with Dynamic Movement Primitives (DMPs). By embedding movement goals as point or limit-cycle attractors in a robotic control architecture, robots can execute smooth reaching, grasping, and bipedal walking motions that automatically adjust to unexpected environmental obstacles or mechanical knocks without halting execution, naturally settling back into the intended trajectory.
In ecological science and environmental management, attractor dynamics underpins the study of ecosystem resilience, regime shifts, and tipping points. Coral reefs, shallow lakes, and rainforests exhibit alternative stable states. When ecological degradation exceeds a critical threshold, a catastrophic bifurcation occurs: a lush, biodiverse lake abruptly collapses into a turbid, eutrophic dead zone. Understanding the topology of these ecological basins allows conservation scientists to measure early-warning signals (such as “critical slowing down”) before an irreversible transition occurs.
11. Research & Empirical Evidence
Empirical evidence validating attractor dynamics in biological nervous systems has accelerated dramatically over the past two decades. In a seminal study on perceptual decision-making, Christopher Wong and Xiao-Jing Wang (2006) developed a biophysically realistic recurrent attractor network that precisely modeled primate behavior and spiking activity in the lateral intraparietal (LIP) area during random-dot motion tasks. Their empirical and computational findings confirmed that perceptual choices emerge via competitive attractor dynamics, where accumulating sensory evidence tilts the energy landscape until neural firing rates settle definitively into one of two discrete decision attractors.
Direct physiological confirmation of continuous ring attractors was achieved in work on the Drosophila melanogaster central complex. Using two-photon calcium imaging in behaving fruit flies, Sung Soo Kim and colleagues (2017) observed a localized bump of neural activity within an anatomically ring-shaped neuropil (the ellipsoid body). As the fly navigated or was passively rotated in darkness, this activity bump rotated smoothly around the physical ring of neurons, maintaining its spatial orientation precisely as predicted by continuous attractor network theories first postulated decades prior.
Furthermore, motor systems research by Mark Churchland, John Cunningham, Krishna Shenoy, and colleagues (2012) at Stanford University demonstrated that motor cortex dynamics during reaching movements do not encode static muscle parameters. Instead, multielectrode recordings in non-human primates revealed that populations of motor cortical neurons execute low-dimensional, quasi-periodic rotational trajectories in state space. This finding demonstrated that motor preparation sets up initial states within an attractor landscape, and execution unfolds as an autonomous, self-sustaining dynamical cascade.
12. Cultural & Cross-Cultural Considerations
While the mathematical formulations of attractor dynamics are universal, cultural, environmental, and linguistic factors heavily sculpt the specific configurations of cognitive attractor landscapes. Language acquisition is an exemplary case. Every natural language carves phonemic space into distinct basins of attraction. An infant is born capable of distinguishing nearly all human speech sounds. However, prolonged exposure to a native language deepens specific phonemic attractors while flattening others.
For instance, native speakers of Japanese often struggle to distinguish between the English phonemes /r/ and /l/. In an English speaker’s linguistic phase space, these two sounds occupy distinct, steep attractor basins separated by a clear categorical boundary. In a native Japanese speaker, the auditory system has organized a single broad attractor basin representing the Japanese liquid consonant /ɾ/. Incoming auditory signals for both /r/ and /l/ fall inevitably into this single well, rendering categorical differentiation difficult without targeted retraining that restructures the perceptual energy landscape.
At a broader epistemological level, cultural cognitive styles influence how human minds navigate problem-solving state spaces. Cross-cultural psychological research (e.g., Richard Nisbett’s work on holistic versus analytic cognition) suggests that Western, highly individualistic educational environments cultivate mental representations with sharp, isolated attractor basins—facilitating categorical segregation and linear causal attribution. Conversely, East Asian, holistic cultural contexts tend to reinforce cognitive landscapes with flatter, highly interconnected basins, supporting relational attention, context-dependent judgments, and dynamic tolerance for dialectical contradiction.
13. Criticisms, Debates & Limitations
Despite its theoretical elegance, attractor dynamics faces several methodological, conceptual, and empirical challenges within cognitive science and neuroscience:
- The Challenge of Non-Stationarity: Classical attractor dynamics assumes that the underlying system parameters remain constant over time, allowing the phase space geometry to remain fixed. However, biological brains are intensely non-stationary. Neuromodulatory tone (e.g., surges of dopamine, acetylcholine, or norepinephrine), ongoing synaptic plasticity, and shifting attention constantly morph the energy landscape itself. Critics argue that treating neural operations as static attractors oversimplifies an architecture that rarely stays still long enough for a trajectory to settle into true mathematical equilibrium.
- Transient Dynamics vs. Asymptotic Equilibrium: True mathematical attractors are asymptotic states reached as time goes to infinity ($t o \infty$). Biological computations, by contrast, are rapid, flexible, and transient. An animal executing an escape maneuver cannot afford to wait for a neural network to settle into a fixed point. Many neuroscientists (such as Timothy Busman and Gyorgy Buzsáki) emphasize that neural circuits operate predominantly in non-equilibrium transient regimes—riding sequences of unstable saddle points rather than lingering in deep attractor wells.
- Metaphorical Overextension: In psychological and social sciences, “attractor dynamics” is frequently co-opted as an untestable metaphor rather than an empirically verified model. Describing an individual’s persistent habit or a social group’s ideological polarization as an “attractor” often adds linguistic flair without delivering formal mathematical formulations, quantitative phase space parameters, or falsifiable predictions.
- Energy Landscape Assumptions: In symmetric physical systems, an energy (Lyapunov) function guarantees that trajectories flow steadily downhill toward local minima. However, biological neural networks are asymmetric; synaptic connections between neurons are directed and rarely reciprocal with equal weight ($W_{ij}
eq W_{ji}$). In the absence of symmetry, conservative energy landscapes break down, leading to non-gradient vector fields that can produce complex limit cycles, chaotic wanders, or non-convergent wandering that traditional attractor models struggle to characterize.
14. Related Terms & Distinctions
To avoid conceptual conflation, attractor dynamics must be carefully distinguished from adjacent dynamical constructs:
- Repellor: The topological inverse of an attractor. A repellor is an unstable fixed point or invariant manifold in phase space from which all neighboring trajectories diverge over time. While an attractor pulls states inward, a repellor actively drives trajectories away.
- Saddle Point: A critical point in phase space that behaves as an attractor along certain geometric dimensions (stable manifolds) but as a repellor along others (unstable manifolds). Trajectories approach the saddle, slow down briefly, and are then deflected outward along the unstable trajectory. Saddle points govern sequential, transient biological computations, such as the switching between distinct behavioral syllables in birdsong.
- Bifurcation: A qualitative topological restructuring of a dynamical system’s phase portrait resulting from a smooth, continuous change in an underlying control parameter. While attractor dynamics describes motion within a fixed landscape, a bifurcation describes the transformation or destruction of the landscape itself (e.g., the sudden splitting of one attractor into two).
- Homeostasis: A classical physiological concept denoting the active maintenance of internal physiological variables (such as body temperature or blood glucose) around a static set point via negative feedback loops. Attractor dynamics encompasses homeostasis (as a simple point attractor) but extends far beyond it to accommodate multistability, oscillations, chaos, and continuous behavioral landscapes.
- Homeorhesis: The stabilization not of a static point, but of an entire developmental or temporal trajectory (a “chreod” in Waddington’s terminology). If an organism’s development is perturbed, homeorhetic mechanisms guide the developmental path back toward the canalized trajectory, operating as an extended spatio-temporal attractor.
15. Summary / Key Takeaways
Attractor dynamics constitutes a transformative mathematical framework that reframes cognition, behavior, and physiological regulation as geometric trajectories navigating a non-linear state space. Rather than relying on rigid, pre-programmed computational instructions, complex biological systems self-organize. Through recurrent interactions, their high-dimensional degrees of freedom collapse into low-dimensional manifolds, manifesting as point attractors, limit cycles, continuous manifolds, or chaotic trajectories.
Across diverse scales of biological organization—from the coordinated firing of cortical ensembles and the maintenance of spatial orientation to motor synergies, psychiatric states, and ecological resilience—attractor landscapes provide the fundamental mechanism for maintaining structural stability in the presence of noise while retaining the capacity for adaptive, state-dependent transitions. Despite valid critiques regarding biological non-stationarity and transient processing, attractor dynamics remains an indispensable paradigm for understanding how coherent order arises spontaneously from physical complexity.
References
- Churchland, M. M., Cunningham, J. P., Kaufman, M. T., Foster, J. D., Nuyujukian, P., Ryu, S. I., & Shenoy, K. V. (2012). Neural population dynamics during reaching. Nature, 487(7405), 51–56. https://doi.org/10.1038/nature11129
- Haken, H., Kelso, J. A. S., & Bunz, H. (1985). A theoretical model of phase transitions in human hand movements. Biological Cybernetics, 51(5), 347–356. https://doi.org/10.1007/BF00336922
- Hopfield, J. J. (1982). Neural networks and physical systems with emergent collective computational abilities. Proceedings of the National Academy of Sciences, 79(8), 2554–2558. https://doi.org/10.1073/pnas.79.8.2554
- Kim, S. S., Rouault, H., Druckmann, S., & Jayaraman, V. (2017). Ring attractor dynamics in the Drosophila central complex. Science, 356(6340), 849–853. https://doi.org/10.1126/science.aal4837
- Thelen, E., & Smith, L. B. (1994). A dynamic systems approach to the development of cognition and action. MIT Press. https://mitpress.mit.edu/9780262700597/
- Wong, K. F., & Wang, X. J. (2006). A recurrent network mechanism of time integration in perceptual decisions. Journal of Neuroscience, 26(4), 1314–1328. https://doi.org/10.1523/JNEUROSCI.3733-05.2006