This note explains the scientific foundations that underpin the computational modeling work of the Contemplative Neurodynamics Program. Section 1 describes the physiological framework of autonomic regulation, allostasis, and interoception — the biological regulatory system the program seeks to model. Section 2 introduces the theoretical framework of active inference and the predictive brain, which provides the computational basis for that modeling. Section 3 reviews existing methodological approaches, identifies their limitations, and describes how the program addresses them.
1. Allostasis, Autonomic Regulation, and Interoception
A fundamental function of the nervous system is the regulation of the body’s internal physiological state while enabling adaptive interaction with the external environment. In modern physiology, this adaptive regulation is often described using the concept of allostasis, a process through which biological systems maintain stability through dynamic adjustment rather than fixed equilibrium. Originally proposed by Sterling and Eyer (1988) and later elaborated by McEwen (1998), allostasis emphasizes that physiological regulation is an active process in which the brain coordinates bodily systems in order to meet anticipated environmental demands.
Unlike classical homeostasis, which assumes that physiological variables remain near fixed set points, allostasis highlights the brain’s role in dynamically adjusting physiological processes in response to changing conditions (Sterling, 2012). Contemporary models further emphasize that this regulation is often predictive: drawing on frameworks such as predictive processing and active inference (Friston, 2010; Clark, 2013), these accounts propose that the brain continuously generates anticipatory adjustments to cardiovascular, metabolic, and respiratory systems based on expectations about internal and external demands, rather than simply reacting to deviations after they occur. The theoretical basis of these frameworks, and their relevance to the computational models developed in this program, is examined in Section 2.
These regulatory processes are mediated primarily through the autonomic nervous system (ANS), which coordinates physiological activity across the body, including cardiovascular function, respiratory dynamics, metabolic activity, and visceral organ function (Berntson et al., 1994). The ANS is commonly described as comprising two interacting branches. The sympathetic nervous system mobilizes physiological resources during conditions of challenge or stress, increasing heart rate, blood pressure, and metabolic activity. The parasympathetic nervous system, in contrast, supports recovery and restoration by promoting processes associated with rest, digestion, and physiological repair. Healthy physiological regulation depends not simply on the activity of either branch alone but on the flexible coordination between them. The ability to shift efficiently between states of sympathetic activation and parasympathetic recovery is a well-established indicator of physiological resilience and adaptive capacity, supported by an extensive body of research on heart rate variability and related measures (Thayer et al., 2012; Porges, 2007).
These autonomic processes are coordinated by an interconnected set of neural structures often referred to as the central autonomic network (Benarroch, 1993). This network includes cortical, subcortical, and brainstem regions that integrate sensory information from both the external environment and the body’s internal organs. Key components include the prefrontal cortex, which contributes to executive control and emotional regulation; the anterior cingulate cortex, which participates in monitoring internal state and guiding adaptive behavior; the insula, which plays a central role in representing internal bodily signals (Craig, 2009); and subcortical structures such as the amygdala, hypothalamus, and brainstem nuclei that coordinate autonomic responses (Critchley, 2005). Through the interaction of these structures, the brain continuously integrates cognitive, emotional, and physiological information in order to regulate bodily processes.
A critical component of this regulatory system is interoception, the sensing and representation of signals originating from the body’s internal organs (Craig, 2002). Interoceptive signals arise from multiple physiological systems, including cardiovascular activity, respiratory rhythms, visceral processes, and musculoskeletal feedback. These signals reach the brain through afferent pathways including the vagus nerve and spinal sensory pathways. The vagus nerve is of particular relevance: as the primary conduit for ascending visceral signals, it also serves as a key substrate through which voluntary practices such as controlled breathing can directly modulate autonomic and interoceptive processing (Porges, 2007; Berntson et al., 1994). Research in neuroscience has shown that interoceptive processing plays an important role in emotional experience, self-regulation, and decision-making (Seth, 2013; Critchley & Garfinkel, 2017; Damasio, 1994).
By integrating interoceptive information with cognitive and environmental inputs, the nervous system adjusts autonomic output and behavioral responses to maintain physiological balance. Critically, top-down input from prefrontal and cingulate cortices to the central autonomic network provides a mechanistic pathway through which voluntary cognitive processes — such as directed attention, controlled breathing, and deliberate bodily awareness — can influence autonomic and interoceptive regulation (Thayer & Lane, 2009). This top-down pathway is central to understanding how contemplative practices engage the brain–body regulatory system and, importantly, why it constitutes a tractable target for computational modeling — as elaborated in Section 2.
2. Active Inference and the Predictive Brain
Modern theoretical neuroscience increasingly characterizes the brain not as a passive processor of sensory information but as an active system that continuously generates predictions about the causes of its inputs. According to this view, the nervous system maintains internal models of both the external environment and the body’s internal physiological state, using these models to anticipate incoming sensory signals and guide behavior (Friston, 2010; Clark, 2013).
When sensory inputs differ from what the brain has predicted, the resulting discrepancy — known as a prediction error — serves as a signal for updating internal models or for initiating actions that bring sensory experience into closer alignment with predictions. This cycle of prediction, error, and updating is central to the framework of active inference, in which perception, action, and learning are understood as coordinated processes that minimize prediction errors across hierarchically organized neural systems (Friston et al., 2017).
Within this framework, the brain constructs generative models: probabilistic representations of the hidden causes that produce observable sensory signals. Because many of the conditions that matter most for survival — environmental threats, physiological needs, emotional states — cannot be directly observed, they must be inferred from available sensory data. Generative models allow the brain to perform this inference continuously and to revise its representations as new information arrives (Friston, 2010; Parr & Friston, 2019). This same principle — inferring hidden states from observable signals — is the foundation of the computational modeling approach used in this program.
As established in Section 1, this predictive logic extends explicitly to interoception. The brain generates predictions about internal physiological signals — including cardiovascular, respiratory, and visceral activity — and regulates autonomic processes in order to maintain expected bodily conditions (Seth & Friston, 2016; Barrett & Simmons, 2015). In this account, physiological regulation is not simply reactive but anticipatory: the nervous system adjusts internal states in advance of predicted demands rather than only correcting deviations after they occur. This view provides a formal computational basis for the allostatic framework described in Section 1, unifying both accounts under a single principle of predictive regulation.
A further implication of this framework is that internal physiological signals are not peripheral to cognition but actively shape perception, emotion, and behavior. Prediction errors arising from the body’s internal state influence the brain’s broader models of the environment and the self, contributing to the felt sense of emotional experience and influencing decision-making (Seth, 2013; Critchley & Garfinkel, 2017).
For the Contemplative Neurodynamics Program, the active inference framework serves a dual purpose. First, it provides a principled theoretical basis for understanding how contemplative practices interact with brain–body regulation: voluntary practices that modulate attention, breathing, or bodily awareness alter the sensory signals informing the brain’s internal models of physiological state, generating prediction errors that may drive changes in neural activity, autonomic output, and the generative models themselves (Pagnoni, 2019; Friston et al., 2017). Second, and equally important, it provides the mathematical and computational scaffolding for the program’s modeling work. Just as the brain infers hidden regulatory states from sensory signals, the program applies probabilistic generative models to multi-modal physiological recordings to infer latent system dynamics — producing computational representations of brain–body regulation that can be analyzed, predicted, and ultimately used to inform adaptive AI systems.
3. Methodological Approaches and Their Limitations
Developing reliable computational models of brain–body regulation requires physiological data that accurately reflects how regulatory dynamics unfold during naturalistic behavior. However, the methodological infrastructure of much current neuroscience research was developed for studying discrete cognitive tasks under controlled laboratory conditions. The resulting data are often poorly suited to training and validating the kinds of dynamical models the program aims to build. Several specific limitations are relevant.
Ecological Constraints in Neuroimaging
A significant portion of contemporary neuroscience research relies on functional magnetic resonance imaging (fMRI), which offers high spatial resolution for mapping brain activity. However, fMRI imposes substantial constraints on participant behavior: individuals must remain physically immobile inside the scanner, and studies are typically restricted to simplified tasks in highly controlled settings (Logothetis, 2008). These constraints make it difficult to observe the physiological processes — changes in breathing, bodily posture, and sustained attentional states — that are central to contemplative practice and to the regulatory dynamics the program seeks to model.
As a result, many studies of contemplative practices rely on measurements taken before and after practice sessions rather than recording neural and physiological activity during the practice itself (Tang et al., 2015). While such approaches can reveal broad changes in neural activation or physiological markers, they do not provide the continuous, time-resolved data needed to build and validate models of regulatory dynamics.
Limitations of Conventional Physiological Signal Analysis
Electroencephalography (EEG), heart rate monitoring, and respiratory measurement offer continuous, non-invasive recording of neural and physiological signals and are better suited to the study of ongoing regulation. However, the analytical methods commonly applied to these signals often rely on assumptions that limit their utility for the program’s computational objectives.
Many studies characterize physiological recordings using averaged spectral features — such as power within predefined EEG frequency bands — or summary statistics derived from heart rate variability (Malik et al., 1996; Sanei & Chambers, 2007). These approaches treat physiological signals as largely independent channels and assume stationary dynamics over time. As established in Sections 1 and 2, however, brain–body regulation emerges from nonlinear, cross-system interactions — between neural, cardiovascular, respiratory, and musculoskeletal processes — operating across multiple timescales. Static or averaged features cannot capture the transient coupling and evolving state dynamics that the program’s generative models are designed to infer.
Artifact Contamination in Naturalistic Recordings
A further challenge arises when recording electrophysiological signals during natural behavior. Contemplative practices that involve deliberate breathing regulation, postural adjustment, or subtle bodily movement generate muscular and motion-related artifacts that can substantially exceed the amplitude of the underlying neural signals (Urigüen & Garcia-Zapirain, 2015).
Standard signal-processing pipelines typically address this by removing contaminated data segments or applying artifact suppression filters. However, this introduces a difficult trade-off: the segments most affected by movement may be precisely those during which the most meaningful regulatory events are occurring. In practices that involve deliberate bodily engagement, artifact contamination and the physiological processes of interest are likely to co-occur — meaning that aggressive artifact removal may systematically exclude the data most relevant to model training.
Implications for the Contemplative Neurodynamics Program
These limitations collectively point to the need for a different methodological approach: one that prioritizes naturalistic, continuous data collection and applies analytical methods capable of handling the nonlinear, non-stationary dynamics of real physiological regulation.
Recent advances make this possible. Wearable physiological sensing technologies now enable synchronized multi-modal recording of neural, cardiovascular, respiratory, and motion signals during natural behavior, without the postural and behavioral constraints of laboratory neuroimaging (Lotte et al., 2018; Poh et al., 2010). In parallel, developments in computational neuroscience and machine learning provide tools capable of modeling complex dynamical systems from high-dimensional time-series data and inferring latent regulatory states from noisy, multi-modal observations (Pandarinath et al., 2018; Friston et al., 2017).
The Contemplative Neurodynamics Program integrates these developments by combining multi-modal wearable sensing with the probabilistic generative modeling framework described in Section 2. This approach yields the continuous, ecologically valid physiological data needed to develop computational models of adaptive regulation — models that are grounded in real-world dynamics and designed to generalize to broader applications in adaptive AI and neurotechnology.
4. References
Barrett, L. F., & Simmons, W. K. (2015). Interoceptive predictions in the brain. Nature Reviews Neuroscience, 16(7), 419–429.
Benarroch, E. E. (1993). The central autonomic network: Functional organization, dysfunction, and perspective. Mayo Clinic Proceedings, 68(10), 988–1001.
Berntson, G. G., Cacioppo, J. T., & Quigley, K. S. (1994). Autonomic cardiac control: I. Estimation and validation from pharmacological blockades. Psychophysiology, 31(6), 572–585.
Clark, A. (2013). Whatever next? Predictive brains, situated agents, and the future of cognitive science. Behavioral and Brain Sciences, 36(3), 181–204.
Craig, A. D. (2002). How do you feel? Interoception: The sense of the physiological condition of the body. Nature Reviews Neuroscience, 3(8), 655–666.
Craig, A. D. (2009). How do you feel — now? The anterior insula and human awareness. Nature Reviews Neuroscience, 10(1), 59–70.
Critchley, H. D. (2005). Neural mechanisms of autonomic, affective, and cognitive integration. Journal of Comparative Neurology, 493(1), 154–166.
Critchley, H. D., & Garfinkel, S. N. (2017). Interoception and emotion. Current Opinion in Psychology, 17, 7–14.
Damasio, A. R. (1994). Descartes’ Error: Emotion, Reason, and the Human Brain. Putnam.
Friston, K. (2010). The free-energy principle: A unified brain theory? Nature Reviews Neuroscience, 11(2), 127–138.
Friston, K., FitzGerald, T., Rigoli, F., Schwartenbeck, P., & Pezzulo, G. (2017). Active inference: A process theory. Neural Computation, 29(1), 1–49.
Logothetis, N. K. (2008). What we can and what we cannot do with fMRI. Nature, 453(7197), 869–878.
Lotte, F., Bougrain, L., Cichocki, A., Clerc, M., Congedo, M., Rakotomamonjy, A., & Yger, F. (2018). A review of classification algorithms for EEG-based brain–computer interfaces: A 10-year update. Journal of Neural Engineering, 15(3), 031005.
Malik, M., Bigger, J. T., Camm, A. J., Kleiger, R. E., Malliani, A., Moss, A. J., & Schwartz, P. J. (1996). Heart rate variability: Standards of measurement, physiological interpretation, and clinical use. European Heart Journal, 17(3), 354–381.
McEwen, B. S. (1998). Stress, adaptation, and disease: Allostasis and allostatic load. Annals of the New York Academy of Sciences, 840, 33–44.
Pagnoni, G. (2019). The contemplative exercise through the lens of predictive processing: A promise for stress reduction. Progress in Brain Research, 244, 199–215.
Pandarinath, C., O’Shea, D. J., Collins, J., Jozefowicz, R., Stavisky, S. D., Kao, J. C., … Shenoy, K. V. (2018). Inferring single-trial neural population dynamics using sequential auto-encoders. Nature Methods, 15(10), 805–815.
Parr, T., & Friston, K. J. (2019). Generalised free energy and active inference. Biological Cybernetics, 113(5–6), 495–513.
Poh, M. Z., Swenson, N. C., & Picard, R. W. (2010). A wearable sensor for unobtrusive, long-term assessment of electrodermal activity. IEEE Transactions on Biomedical Engineering, 57(5), 1243–1252.
Porges, S. W. (2007). The polyvagal perspective. Biological Psychology, 74(2), 116–143.
Sanei, S., & Chambers, J. A. (2007). EEG Signal Processing. Wiley.
Seth, A. K. (2013). Interoceptive inference, emotion, and the embodied self. Trends in Cognitive Sciences, 17(11), 565–573.
Seth, A. K., & Friston, K. J. (2016). Active interoceptive inference and the emotional brain. Philosophical Transactions of the Royal Society B, 371(1708), 20160007.
Sterling, P. (2012). Allostasis: A model of predictive regulation. Physiology & Behavior, 106(1), 5–15.
Sterling, P., & Eyer, J. (1988). Allostasis: A new paradigm to explain arousal pathology. In S. Fisher & J. Reason (Eds.), Handbook of Life Stress, Cognition and Health (pp. 629–649). Wiley.
Tang, Y. Y., Hölzel, B. K., & Posner, M. I. (2015). The neuroscience of mindfulness meditation. Nature Reviews Neuroscience, 16(4), 213–225.
Thayer, J. F., Åhs, F., Fredrikson, M., Sollers, J. J., & Wager, T. D. (2012). A meta-analysis of heart rate variability and neuroimaging studies: Implications for heart rate variability as a marker of stress and health. Neuroscience & Biobehavioral Reviews, 36(2), 747–756.
Thayer, J. F., & Lane, R. D. (2009). Claude Bernard and the heart–brain connection: Further elaboration of a model of neurovisceral integration. Neuroscience & Biobehavioral Reviews, 33(2), 81–88.
Urigüen, J. A., & Garcia-Zapirain, B. (2015). EEG artifact removal — state-of-the-art and guidelines. Journal of Neural Engineering, 12(3), 031001.