pith. sign in

arxiv: 2605.15862 · v1 · pith:MF234KB4new · submitted 2026-05-15 · 💻 cs.LG · q-bio.NC

From Observed Viability to Internal Predictive Approximation: A Single-Subject Latent-Space Analysis of Gait Dynamics Under Occlusal Constraint

Jacques Raynal , Pierre Slangen , Elsa Raynal , Jacques Margerit This is my paper

Pith reviewed 2026-05-20 20:39 UTC · model grok-4.3

classification 💻 cs.LG q-bio.NC
keywords single-subject analysisgait dynamicslatent spaceocclusal constraintprincipal component analysisneural networklongitudinal transformationParkinsonian gait
0
0 comments X

The pith

A neural network can internally approximate observed longitudinal changes in gait latent space from a single Parkinsonian subject.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

This paper tests whether a longitudinal shift in gait organization, recorded eleven weeks apart under varying jaw positions, can be captured inside a predictive model built from the same limited dataset. Instrumented insole measurements from one Parkinsonian participant are reduced via principal component analysis to a PC1-PC2 plane, then a feed-forward neural network is trained to map the first session's patterns onto the second. The resulting approximation reproduces the displacement ordering among three centric-relation conditions and largely preserves the overall pattern when all six occlusal probes are included. The work stays strictly within this single-subject scope and makes no claims about clinical prediction or causal effects of occlusion.

Core claim

Observed longitudinal latent transformations of gait dynamics can be internally approximated within this single-subject dataset: a PCA-derived PC1-PC2 representation combined with a supervised feed-forward neural network trained to map session M1 to M2 preserves the ordering OC3 < ONL < OC2.5 for centric-relation conditions and maintains the global structure of exploratory displacement patterns across the full set of six occlusal probes, with held-out and leave-condition-out checks showing condition-dependent variability.

What carries the argument

PCA-constructed PC1-PC2 latent space paired with a feed-forward neural network trained to approximate the M1-to-M2 session transformation in gait data collected under six occlusal probes.

If this is right

  • The relative ordering of displacement magnitudes for centric-relation occlusal conditions is preserved by the trained approximation.
  • The global structure of exploratory displacement patterns across all six probes remains intact under the M1-to-M2 mapping.
  • Condition-dependent variability appears in held-out M2 and leave-condition-out tests, indicating that approximation quality is not uniform across probes.
  • The approach supplies a restricted methodological bridge for constructing internal predictive viability models from longitudinal biomechanical recordings.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The single-subject internal approximation could be tested for scaling to small multi-subject cohorts to separate true longitudinal signals from individual artifacts.
  • Similar latent-space mappings might be applied to other constrained biomechanical tasks, such as posture or balance under different sensory conditions.
  • If the preserved structure reflects stable organization rather than overfitting, the same pipeline could be used to quantify how much session-to-session change is predictable from within-subject data alone.
  • The condition-dependent variability observed in leave-condition-out tests suggests a natural next step of identifying which probes contribute most to the latent transformation.

Load-bearing premise

The two-dimensional PCA space and the neural network trained on this small single-subject dataset capture a meaningful approximation of the underlying gait transformation instead of noise or session-specific artifacts.

What would settle it

A replication study on the same or additional subjects that fails to reproduce the OC3 < ONL < OC2.5 displacement ordering or the global six-probe structure in the approximated latent space would falsify the internal-approximation claim.

Figures

Figures reproduced from arXiv: 2605.15862 by Elsa Raynal, Jacques Margerit, Jacques Raynal, Pierre Slangen.

Figure 1
Figure 1. Figure 1: Observed and predicted latent centroid trajectories between M1 and M2 for the core Level [PITH_FULL_IMAGE:figures/full_fig_p018_1.png] view at source ↗
read the original abstract

Adaptive biomechanical systems may show similar observable gait performance while differing in latent organization and longitudinal behavior. This study examines whether an observed longitudinal transformation of gait organization can be approximated within a predictive latent-space framework, without claiming clinical prediction or causal occlusal effects. Using an exploratory single-subject design in a Parkinsonian participant, gait was recorded with instrumented insoles during two sessions separated by eleven weeks. Six occlusal observational probes were tested: natural occlusion, open-mouth disengagement, strong clenching, two vertical-dimension increases in centric relation, and one vertical-dimension increase with mandibular protrusion. Principal Component Analysis was used to construct a PC1--PC2 latent representation. A simplified supervised machine-learning model, implemented as a feed-forward neural network, was trained to approximate the observed M1--M2 transformation. The primary analysis focused on the three centric-relation conditions and tested whether the displacement hierarchy could be reproduced. The model preserved the ordering OC3 < ONL < OC2.5. The extended six-probe analysis also preserved the global structure of the exploratory displacement pattern, with OC3 and OC3P closely grouped and the highest displacements associated with OC2.5 and open-mouth disengagement. Held-out M2 and leave-condition-out analyses showed condition-dependent approximation variability. These findings do not establish generalizable prediction, therapeutic superiority, causal occlusal effects, or clinical viability forecasting. They support only the restricted conclusion that observed longitudinal latent transformations can be internally approximated within this single-subject dataset, providing a methodological bridge toward future multi-subject predictive viability models.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript reports an exploratory single-subject analysis of gait dynamics in a Parkinsonian participant across two sessions (M1 and M2, separated by eleven weeks) under six occlusal probes. PCA constructs a PC1-PC2 latent space from instrumented insole data; a feed-forward neural network is trained to approximate the observed M1-M2 transformation. The primary result is that the model reproduces the displacement ordering OC3 < ONL < OC2.5 for centric-relation conditions and preserves global structure across all probes in an extended analysis, while held-out and leave-condition-out checks show condition-dependent variability. The paper restricts its conclusion to internal approximation within this dataset, explicitly disclaiming generalizable prediction, causal effects, or clinical utility.

Significance. If the reported ordering and structure preservation reflect a genuine internal approximation rather than dataset-specific fitting, the work supplies a methodological template for modeling longitudinal latent changes in biomechanical gait organization via supervised ML on small corpora. This could serve as a bridge to multi-subject predictive viability frameworks, particularly given the explicit scoping that avoids overclaiming. The single-subject design and internal validation, however, constrain broader significance to proof-of-concept status.

major comments (2)
  1. [Abstract and primary analysis section] Abstract and primary analysis section: The feed-forward NN is trained to approximate the M1-M2 transformation using the identical single-subject PCA-derived points (two sessions, six probes) on which the observed ordering OC3 < ONL < OC2.5 was measured. This reduces the reproduction of the hierarchy to a direct fit of the measured change, undermining the claim of an 'internal predictive approximation' independent of the training data itself.
  2. [Results (held-out M2 and leave-condition-out analyses)] Results (held-out M2 and leave-condition-out analyses): These checks remain entirely within the same subject and two-session corpus, so they cannot distinguish a meaningful approximation of the longitudinal transformation from reproduction of session-specific artifacts or noise in the PCA axes. The absence of error bars, confidence intervals, or statistical tests on the preserved ordering leaves the strength of the hierarchy evidence unclear.
minor comments (2)
  1. [Methods] Clarify the precise definition and computation of 'displacement' (e.g., vector difference or Euclidean norm in PC1-PC2) at first use in the methods to avoid ambiguity in interpreting the ordering and global structure results.
  2. [Abstract] The abstract would benefit from stating the number of gait trials or steps per condition to better contextualize the extremely limited data regime for readers.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the detailed review and insightful comments on our exploratory single-subject analysis. We address the major comments below, providing clarifications on the scope of our internal approximation approach and indicating where revisions will be made to enhance transparency.

read point-by-point responses

  1. Referee: Abstract and primary analysis section: The feed-forward NN is trained to approximate the M1-M2 transformation using the identical single-subject PCA-derived points (two sessions, six probes) on which the observed ordering OC3 < ONL < OC2.5 was measured. This reduces the reproduction of the hierarchy to a direct fit of the measured change, undermining the claim of an 'internal predictive approximation' independent of the training data itself.

    Authors: We agree that the neural network is trained and evaluated on the observed points from the same dataset, which aligns with our explicit framing as an internal approximation rather than an independent or out-of-sample prediction. The goal was to determine whether a feed-forward model could reproduce the specific displacement ordering observed in the centric-relation conditions. This serves as a proof-of-concept for approximating longitudinal latent transformations within a small, controlled corpus. We will revise the abstract and primary analysis section to more clearly state that the approximation is descriptive of the training data's structure and does not claim independence from it. revision: partial

  2. Referee: Results (held-out M2 and leave-condition-out analyses): These checks remain entirely within the same subject and two-session corpus, so they cannot distinguish a meaningful approximation of the longitudinal transformation from reproduction of session-specific artifacts or noise in the PCA axes. The absence of error bars, confidence intervals, or statistical tests on the preserved ordering leaves the strength of the hierarchy evidence unclear.

    Authors: We acknowledge that all analyses are confined to the single-subject, two-session dataset, which inherently limits the ability to separate genuine longitudinal patterns from potential session-specific effects or PCA artifacts. The held-out and leave-condition-out checks were performed to explore variability across conditions, as described. Given the small sample size, we did not include statistical tests or error bars on the ordering, as these would require assumptions not supported by the design. We will add explicit language in the results and discussion sections to highlight this limitation and reinforce the exploratory, non-generalizable nature of the findings. revision: yes

Circularity Check

1 steps flagged

NN trained to approximate observed M1-M2 shift reproduces hierarchy by construction

specific steps
  1. fitted input called prediction [Abstract]
    "A simplified supervised machine-learning model, implemented as a feed-forward neural network, was trained to approximate the observed M1--M2 transformation. The primary analysis focused on the three centric-relation conditions and tested whether the displacement hierarchy could be reproduced. The model preserved the ordering OC3 < ONL < OC2.5."

    The neural network is explicitly supervised-trained on the observed M1-M2 transformation extracted from the same PCA-derived latent space of the two-session single-subject recordings. Any reproduction of the displacement ordering or global structure is therefore a direct consequence of minimizing error against the training target rather than constituting an independent prediction or first-principles result.

full rationale

The paper derives its central result by first applying PCA to the single-subject gait data to obtain a PC1-PC2 latent space, then training a feed-forward neural network in a supervised manner to approximate the observed M1-to-M2 transformation within that same space. The reported success in preserving the OC3 < ONL < OC2.5 ordering and global displacement structure is a direct outcome of this supervised fitting process on the identical limited dataset (two sessions, six probes). Although the authors qualify the conclusion as restricted to internal approximation and note condition-dependent variability in held-out checks, the framing of the fit as a 'predictive' step and the test of whether the hierarchy 'could be reproduced' reduces the claimed approximation to a tautological reproduction of the training target. No independent external benchmark or parameter-free derivation breaks the loop.

Axiom & Free-Parameter Ledger

1 free parameters · 1 axioms · 0 invented entities

Abstract provides insufficient detail to enumerate all free parameters or axioms; the central claim rests on the assumption that the chosen latent space and model architecture meaningfully represent gait organization changes.

free parameters (1)
  • Neural network parameters
    Weights and biases fitted to approximate the observed M1-M2 transformation on the single-subject data.
axioms (1)
  • domain assumption PCA on gait features yields a latent space that captures relevant organization changes under occlusal constraint
    Invoked when constructing the PC1-PC2 representation used for the approximation task.

pith-pipeline@v0.9.0 · 5839 in / 1326 out tokens · 44839 ms · 2026-05-20T20:39:49.557668+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

32 extracted references · 32 canonical work pages · 2 internal anchors

  1. [1]

    Observable Performance Does Not Fully Reflect System Organization: A Multi-Level Analysis of Gait Dynamics Under Occlusal Constraint

    Raynal J, Slangen P, Margerit J. Observable Performance Does Not Fully Reflect System Or- ganization: A Multi-Level Analysis of Gait Dynamics Under Occlusal Constraint. arXiv. 2026. doi:10.48550/arXiv.2605.00778

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.00778 2026

  2. [2]

    From Organization to Viability: A Multi-Level Analysis of Gait Dynamics Under Occlusal Constraint

    Raynal J, Slangen P, Raynal E, Margerit J. From Organization to Viability: A Multilevel Anal- ysis of Gait Dynamics Under Occlusal Constraint. arXiv. 2026. arXiv:2605.13893 [q-bio.OT]. doi:10.48550/arXiv.2605.13893

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2605.13893 2026

  3. [3]

    Human balance and posture control during standing and walking

    Winter DA. Human balance and posture control during standing and walking. Gait Posture. 1995;3(4):193–214. doi:10.1016/0966-6362(96)82849-9

    work page doi:10.1016/0966-6362(96)82849-9 1995

  4. [4]

    Postural orientation and equilibrium: what do we need to know about neural control of balance to prevent falls? Age Ageing

    Horak FB. Postural orientation and equilibrium: what do we need to know about neural control of balance to prevent falls? Age Ageing. 2006;35(Suppl 2):ii7–ii11. doi:10.1093/ageing/afl077

    work page doi:10.1093/ageing/afl077 2006

  5. [5]

    Gait dynamics in Parkinson’s disease: common and distinct behavior among stride length, gait variability, and fractal-like scaling

    Hausdorff JM. Gait dynamics in Parkinson’s disease: common and distinct behavior among stride length, gait variability, and fractal-like scaling. Chaos. 2009;19(2):026113. doi:10.1063/1.3147408

    work page doi:10.1063/1.3147408 2009

  6. [6]

    Human movement variability, nonlinear dynamics, and pathology: is there a connection? Hum Mov Sci

    Stergiou N, Decker LM. Human movement variability, nonlinear dynamics, and pathology: is there a connection? Hum Mov Sci. 2011;30(5):869–888. doi:10.1016/j.humov.2011.06.002

    work page doi:10.1016/j.humov.2011.06.002 2011

  7. [7]

    Movement disorders in people with Parkinson disease: a model for physical therapy

    Morris ME. Movement disorders in people with Parkinson disease: a model for physical therapy. Phys Ther. 2000;80(6):578–597. doi:10.1093/ptj/80.6.578

    work page doi:10.1093/ptj/80.6.578 2000

  8. [8]

    Gait impairments in Parkinson’s disease

    Mirelman A, Bonato P, Camicioli R, Ellis TD, Giladi N, Hamilton JL, et al. Gait impairments in Parkinson’s disease. Lancet Neurol. 2019;18(7):697–708. doi:10.1016/S1474-4422(19)30044-4

    work page doi:10.1016/s1474-4422(19)30044-4 2019

  9. [9]

    Free-living gait characteristics in ageing and Parkinson’s disease: impact of environment and ambulatory bout length

    Del Din S, Godfrey A, Galna B, Lord S, Rochester L. Free-living gait characteristics in ageing and Parkinson’s disease: impact of environment and ambulatory bout length. J Neuroeng Rehabil. 2016;13:46. doi:10.1186/s12984-016-0154-5

    work page doi:10.1186/s12984-016-0154-5 2016

  10. [10]

    The bliss of motor abundance

    Latash ML. The bliss of motor abundance. Exp Brain Res. 2012;217(1):1–5. doi:10.1007/s00221- 012-3000-4

    work page doi:10.1007/s00221- 2012

  11. [11]

    Constraints on the development of coordination

    Newell KM. Constraints on the development of coordination. In: Wade MG, Whiting HTA, editors. Motor Development in Children: Aspects of Coordination and Control. Dordrecht: Martinus Nijhoff

    work page

  12. [12]

    Dynamic Patterns: The Self-Organization of Brain and Behavior

    Kelso JAS. Dynamic Patterns: The Self-Organization of Brain and Behavior. Cambridge (MA): MIT Press; 1995. doi:10.7551/mitpress/2411.001.0001

    work page doi:10.7551/mitpress/2411.001.0001 1995

  13. [13]

    Coordination

    Turvey MT. Coordination. Am Psychol. 1990;45(8):938–953. doi:10.1037/0003-066X.45.8.938

    work page doi:10.1037/0003-066x.45.8.938 1990

  14. [14]

    Principal component analysis: a review and recent developments

    Jolliffe IT, Cadima J. Principal component analysis: a review and recent developments. Philos Trans A Math Phys Eng Sci. 2016;374(2065):20150202. doi:10.1098/rsta.2015.0202

    work page doi:10.1098/rsta.2015.0202 2016

  15. [15]

    Representation learning: a review and new perspectives

    Bengio Y , Courville A, Vincent P. Representation learning: a review and new perspectives. IEEE Trans Pattern Anal Mach Intell. 2013;35(8):1798–1828. doi:10.1109/TPAMI.2013.50

    work page doi:10.1109/tpami.2013.50 2013

  16. [16]

    A path towards autonomous machine intelligence

    LeCun Y . A path towards autonomous machine intelligence. OpenReview. 2022. Available from: https://openreview.net/forum?id=BZ5a1r-kVsf

    work page 2022

  17. [17]

    Deep Learning

    Goodfellow I, Bengio Y , Courville A. Deep Learning. Cambridge (MA): MIT Press; 2016. 30

    work page 2016

  18. [18]

    Adam: a method for stochastic optimization

    Kingma DP, Ba J. Adam: a method for stochastic optimization. In: International Conference on Learning Representations; 2015

    work page 2015

  19. [19]

    Pattern Recognition and Machine Learning

    Bishop CM. Pattern Recognition and Machine Learning. New York: Springer; 2006

    work page 2006

  20. [20]

    Error correction, sensory prediction, and adaptation in motor control

    Shadmehr R, Smith MA, Krakauer JW. Error correction, sensory prediction, and adaptation in motor control. Annu Rev Neurosci. 2010;33:89–108. doi:10.1146/annurev-neuro-060909-153135

    work page doi:10.1146/annurev-neuro-060909-153135 2010

  21. [21]

    Body awareness: construct and self-report measures

    Mehling WE, Gopisetty V , Daubenmier J, Price CJ, Hecht FM, Stewart A. Body awareness: construct and self-report measures. PLoS One. 2009;4(5):e5614. doi:10.1371/journal.pone.0005614

    work page doi:10.1371/journal.pone.0005614 2009

  22. [22]

    Movement-based embodied contemplative practices: definitions and paradigms

    Schmalzl L, Crane-Godreau MA, Payne P. Movement-based embodied contemplative practices: definitions and paradigms. Front Hum Neurosci. 2014;8:205. doi:10.3389/fnhum.2014.00205

    work page doi:10.3389/fnhum.2014.00205 2014

  23. [23]

    Dental occlusion, body posture and temporomandibular disorders: where we are now and where we are heading for

    Manfredini D, Castroflorio T, Perinetti G, Guarda-Nardini L. Dental occlusion, body posture and temporomandibular disorders: where we are now and where we are heading for. J Oral Rehabil. 2012;39(6):463–471. doi:10.1111/j.1365-2842.2012.02291.x

    work page doi:10.1111/j.1365-2842.2012.02291.x 2012

  24. [24]

    Dental occlusion and body posture: no detectable correlation

    Perinetti G. Dental occlusion and body posture: no detectable correlation. Gait Posture. 2006;24(2):165–168. doi:10.1016/j.gaitpost.2005.08.004

    work page doi:10.1016/j.gaitpost.2005.08.004 2006

  25. [25]

    Occlusion and posture: myth or reality? J Oral Rehabil

    Michelotti A, Farella M. Occlusion and posture: myth or reality? J Oral Rehabil. 2010;37(5):317–

    work page 2010

  26. [26]

    doi:10.1111/j.1365-2842.2010.02072.x

    work page doi:10.1111/j.1365-2842.2010.02072.x 2010

  27. [27]

    Dental occlusion modifies gaze and posture stabilization in human subjects

    Gangloff P, Louis JP, Perrin PP. Dental occlusion modifies gaze and posture stabilization in human subjects. Neurosci Lett. 2000;293(3):203–206. doi:10.1016/S0304-3940(00)01516-5

    work page doi:10.1016/s0304-3940(00)01516-5 2000

  28. [28]

    Unilateral trigeminal anaesthesia modifies postural control in human subjects

    Gangloff P, Perrin PP. Unilateral trigeminal anaesthesia modifies postural control in human subjects. Neurosci Lett. 2002;330(2):179–182. doi:10.1016/S0304-3940(02)00779-6

    work page doi:10.1016/s0304-3940(02)00779-6 2002

  29. [29]

    Dental occlusion and postural control in adults

    Tardieu C, Dumitrescu M, Giraudeau A, Blanc JL, Cheynet F, Borel L. Dental occlusion and postural control in adults. Neurosci Lett. 2009;450(2):221–224. doi:10.1016/j.neulet.2008.12.005

    work page doi:10.1016/j.neulet.2008.12.005 2009

  30. [30]

    A short latency vestibulomasseteric reflex evoked by electrical vestibular stimulation in healthy humans

    Deriu F, Tolu E, Rothwell JC. A short latency vestibulomasseteric reflex evoked by electrical vestibular stimulation in healthy humans. J Physiol. 2003;553(Pt 1):267–279. doi:10.1113/jphysiol.2003.047274

    work page doi:10.1113/jphysiol.2003.047274 2003

  31. [31]

    Single-case experimental designs: a systematic review of published research and current standards

    Smith JD. Single-case experimental designs: a systematic review of published research and current standards. Psychol Methods. 2012;17(4):510–550. doi:10.1037/a0029312

    work page doi:10.1037/a0029312 2012

  32. [32]

    Single-Case Research Designs: Methods for Clinical and Applied Settings

    Kazdin AE. Single-Case Research Designs: Methods for Clinical and Applied Settings. 2nd ed. New York: Oxford University Press; 2011. 31

    work page 2011