pith. sign in

arxiv: 2606.29655 · v1 · pith:NXL3ERFDnew · submitted 2026-06-28 · 🧬 q-bio.NC · cs.NE· q-bio.QM

Geometric Stability of Neural Population Codes: Regional Variation, Behavioral Relevance, and Circuit Dependence

Prashant C. Raju This is my paper

Pith reviewed 2026-06-30 07:16 UTC · model grok-4.3

classification 🧬 q-bio.NC cs.NEq-bio.QM
keywords geometric stabilityrepresentational dissimilarity matricesneural population codesattractor networksbehavioral couplingregional hierarchyrecurrent connectivity
0
0 comments X

The pith

Geometric stability of neural population codes, measured as split-half RDM correlation, independently predicts trial-by-trial behavioral coupling and emerges from recurrent excitatory connections.

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

The paper claims that the consistency of pairwise distances among stimuli in a neural population across independent halves of the same recording session forms a distinct axis of representational reliability. This property, called geometric stability, is shown to be separate from both how much the average response to each stimulus drifts over time and how accurately stimuli can be decoded from the population. In data from dozens of brain regions, geometric stability tracks how closely neural activity on each trial aligns with the animal's choice, while drift measures show no such link. The regions with highest geometric stability follow a hierarchy opposite to the one seen for temporal stability, and a recurrent attractor network reproduces the effect when excitatory connections are strong enough to complete partial input patterns.

Core claim

Geometric stability, formalized as the Spearman rank correlation between split-half representational dissimilarity matrices (Shesha), is empirically dissociable from temporal stability and decoding accuracy. Across 229 area-session observations it predicts trial-by-trial neural-behavioral coupling (ρ = 0.18, p = 0.005) while centroid drift does not (ρ = 0.002, p = 0.976); the regional hierarchy runs opposite to temporal stability; and an attractor network model with recurrent excitatory coupling produces the observed split-half RDM consistency (ρ = +0.64, p = 0.010).

What carries the argument

Shesha, the Spearman rank correlation between split-half representational dissimilarity matrices, which quantifies the reproducibility of the pairwise distance structure among stimuli within a single session.

If this is right

  • Geometric stability predicts neural-behavioral coupling while centroid drift shows no relation.
  • The ordering of brain regions by geometric stability is roughly the reverse of their ordering by temporal stability.
  • Recurrent excitatory coupling in an attractor network is sufficient to generate the observed level of split-half RDM consistency.

Where Pith is reading between the lines

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

  • Manipulations that weaken recurrent excitation should reduce geometric stability without necessarily increasing centroid drift.
  • The same split-half RDM measure could be applied to test whether geometric stability changes during learning or across different task demands.
  • If the mechanism is general, similar regional patterns should appear in datasets from other sensory modalities.

Load-bearing premise

That the split-half RDM Spearman correlation provides a valid and independent measure of geometric stability that is not reducible to or confounded by other properties of the neural data, task design, or analysis choices in the Steinmetz et al. 2019 dataset.

What would settle it

If, after controlling for decoding accuracy and other population statistics in the same recordings, the correlation between geometric stability and trial-by-trial behavioral coupling falls to zero or reverses sign.

Figures

Figures reproduced from arXiv: 2606.29655 by Prashant C. Raju.

Figure 1
Figure 1. Figure 1: Geometric stability predicts neural-behavioral coupling; centroid drift does not. 7 [PITH_FULL_IMAGE:figures/full_fig_p007_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Recurrent circuitry predicts geometric stability in the olfactory hierar￾chy. Geometric stability (Shesha) in three groups from the Bolding and Franks (2018) PCX-1 dataset. OB: olfactory bulb recordings (n = 11); TeLC PCx: piriform cortex with recurrent connections silenced by tetanus toxin light chain (n = 7); Control PCx: contralateral intact piriform cortex (n = 5). Bars show mean with 95% bootstrap con… view at source ↗
Figure 3
Figure 3. Figure 3: Recurrent coupling increases geometric stability via pattern completion in a rate network model. 12 [PITH_FULL_IMAGE:figures/full_fig_p012_3.png] view at source ↗
read the original abstract

Current models of representational reliability in neural populations focus on temporal stability: whether population centroids are preserved across sessions and days. This framing leaves a fundamental question unanswered: how reliably does the pairwise distance structure among stimuli reproduce across independent observations within a session? We argue that this property, geometric stability, constitutes an independent axis of representational analysis that existing frameworks do not capture. We formalize geometric stability as the Spearman rank correlation between split-half representational dissimilarity matrices (Shesha) and show that it is empirically dissociable from both temporal stability and decoding accuracy. Across 229 area-session observations spanning 68 brain regions in a visual discrimination task (Steinmetz et al. 2019), geometric stability predicts trial-by-trial neural-behavioral coupling ($\rho = 0.18$, $p = 0.005$) while centroid drift does not ($\rho = 0.002$, $p = 0.976$). The regional hierarchy, with striatum most stable ($\bar{S} = 0.44$) and hippocampus least ($\bar{S} = 0.19$), runs roughly opposite to the temporal stability hierarchy. Directionally consistent olfactory data (Bolding \& Franks 2018) motivate an attractor network model in which recurrent excitatory coupling amplifies split-half RDM consistency by completing stimulus patterns from sparse feedforward input ($\rho = +0.64$, $p = 0.010$), providing a circuit-level account of how geometric stability emerges. These results establish geometric stability as a functionally relevant, circuit-dependent property of neural population codes, orthogonal to temporal drift measures and complementary to recent accounts of how recurrent connectivity balances representational stability with sequential dynamics in hippocampal circuits.

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 proposes geometric stability, measured as the Spearman rank correlation between split-half representational dissimilarity matrices (Shesha), as an independent property of neural population codes. Using data from 229 area-session observations in a visual discrimination task (Steinmetz et al. 2019), it demonstrates that Shesha is dissociable from temporal stability and decoding accuracy, predicts trial-by-trial neural-behavioral coupling (ρ = 0.18, p = 0.005) unlike centroid drift, shows regional variation opposite to temporal stability, and is accounted for by an attractor network model with recurrent excitatory coupling that reproduces the observed consistency (ρ = +0.64, p = 0.010).

Significance. If the geometric stability measure proves independent of confounds, the work would establish a new, functionally relevant axis of representational analysis orthogonal to temporal drift, with direct links to behavior and a plausible circuit mechanism. This could complement existing frameworks focused on stability over time and provide insights into how recurrent connectivity shapes population codes.

major comments (2)
  1. [Empirical analyses of Steinmetz et al. 2019 dataset (results describing the 229 observations and behavioral correlations] The claim that geometric stability is empirically dissociable from temporal stability/decoding accuracy and predicts behavior (ρ = 0.18) across the 229 area-session observations rests on Shesha being unconfounded by area-specific differences in trial count, neuron number, or stimulus sampling. No controls, partial correlations, or regressions for these factors are described, which is load-bearing because split-half RDM correlations are known to be sensitive to recording quality metrics that vary across regions (e.g., striatum vs. hippocampus) in the Steinmetz et al. 2019 dataset.
  2. [Attractor network model section] The attractor network model is motivated by the observed empirical patterns and reports a correlation (ρ = +0.64) generated from the model itself; this is load-bearing for the circuit-dependence claim because it may reflect parameter choices rather than an independent prediction, and no out-of-sample test or parameter-free derivation is provided.
minor comments (2)
  1. The abstract states the regional means (striatum ar{S} = 0.44, hippocampus ar{S} = 0.19) but does not specify the exact statistical test or correction used for the regional hierarchy comparison.
  2. Clarify whether the olfactory data (Bolding & Franks 2018) enters the analysis only directionally or is used for any quantitative comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major point below and indicate where revisions will be made.

read point-by-point responses

  1. Referee: [Empirical analyses of Steinmetz et al. 2019 dataset (results describing the 229 observations and behavioral correlations] The claim that geometric stability is empirically dissociable from temporal stability/decoding accuracy and predicts behavior (ρ = 0.18) across the 229 area-session observations rests on Shesha being unconfounded by area-specific differences in trial count, neuron number, or stimulus sampling. No controls, partial correlations, or regressions for these factors are described, which is load-bearing because split-half RDM correlations are known to be sensitive to recording quality metrics that vary across regions (e.g., striatum vs. hippocampus) in the Steinmetz et al. 2019 dataset.

    Authors: We agree that additional controls are needed to fully establish independence from recording-quality confounds. The current manuscript shows dissociation from temporal stability and decoding accuracy but does not report partial correlations or regressions controlling for trial count, neuron number, or stimulus sampling. In the revised manuscript we will add these analyses and report whether the behavioral correlation and regional hierarchy remain significant after controlling for the listed factors. revision: yes

  2. Referee: [Attractor network model section] The attractor network model is motivated by the observed empirical patterns and reports a correlation (ρ = +0.64) generated from the model itself; this is load-bearing for the circuit-dependence claim because it may reflect parameter choices rather than an independent prediction, and no out-of-sample test or parameter-free derivation is provided.

    Authors: The model is presented as a mechanistic demonstration that recurrent excitatory coupling can produce the observed geometric stability, reproducing both the empirical ρ = +0.64 match and the directional consistency from the olfactory dataset. We acknowledge that parameters were selected to align with the data and that no out-of-sample test is provided. In revision we will explicitly state the model's role as an existence proof, add a parameter-sensitivity analysis, and discuss the absence of out-of-sample validation as a limitation. revision: partial

Circularity Check

0 steps flagged

No significant circularity; results grounded in external data

full rationale

The derivation relies on empirical application of Shesha (split-half RDM Spearman) to the external Steinmetz et al. 2019 dataset across 229 observations, with reported dissociations (from temporal stability/decoding) and behavioral correlation (ρ=0.18) that are directly testable against that data. The attractor model is motivated by a separate external study (Bolding & Franks 2018) and produces a reported correlation (ρ=+0.64) from simulation; no equations or steps in the provided text reduce the central claims to self-definition, fitted inputs renamed as predictions, or self-citation chains. The analysis is self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Based on abstract only; the central claim rests on the domain assumption that representational dissimilarity matrices capture meaningful stimulus relationships and that split-half correlation isolates geometric structure. No explicit free parameters, axioms, or invented entities are detailed in the abstract.

pith-pipeline@v0.9.1-grok · 5843 in / 1218 out tokens · 54905 ms · 2026-06-30T07:16:04.846633+00:00 · methodology

discussion (0)

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

Reference graph

Works this paper leans on

30 extracted references · 6 canonical work pages · 3 internal anchors

  1. [1]

    Aitken, K., Garrett, M., Olsen, S., & Mihalas, S. (2022). The geometry of representa- tional drift in natural and artificial neural networks.PLOS Computational Biology, 18(11), e1010716

    2022

  2. [2]

    A., & Franks, K

    Bolding, K. A., & Franks, K. M. (2018). Recurrent cortical circuits implement concentration-invariant odor coding.Science,361(6407)

    2018

  3. [3]

    A., Nagappan, S., Han, B.-X., Wang, F., & Franks, K

    Bolding, K. A., Nagappan, S., Han, B.-X., Wang, F., & Franks, K. M. (2020). Recurrent 24 circuitry is required to stabilize piriform cortex odor representations across brain states.eLife,9

    2020

  4. [4]

    Deitch, D., Rubin, A., & Ziv, Y . (2021). Representational drift in the mouse visual cortex.Current Biology,31(19), 4327–4339.e6

    2021

  5. [5]

    Diedrichsen, J., & Kriegeskorte, N. (2017). Representational models: A common framework for understanding encoding, pattern-component, and representational- similarity analysis.PLOS Computational Biology,13(4), e1005508

    2017

  6. [6]

    H., Shahbazi, M., & Kriegeskorte, N

    Diedrichsen, J., Berlot, E., Mur, M., Sch ¨utt, H. H., Shahbazi, M., & Kriegeskorte, N. (2021). Comparing representational geometries using whitened unbiased-distance- matrix similarity.Neurons, Behavior , Data analysis, and Theory,5(3)

    2021

  7. [7]

    N., Pettit, N

    Driscoll, L. N., Pettit, N. L., Minderer, M., Chettih, S. N., & Harvey, C. D. (2017). Dy- namic Reorganization of Neuronal Activity Patterns in Parietal Cortex.Cell,170(5), 986–999.e16

    2017

  8. [8]

    N., Duncker, L., & Harvey, C

    Driscoll, L. N., Duncker, L., & Harvey, C. D. (2022). Representational drift: Emerging theories for continual learning and experimental future directions.Current Opinion in Neurobiology,76, 102609

    2022

  9. [9]

    A., Perich, M

    Gallego, J. A., Perich, M. G., Chowdhury, R. H., Solla, S. A., & Miller, L. E. (2020). Long-term stability of cortical population dynamics underlying consistent behavior. Nature Neuroscience,23(2), 260–270

    2020

  10. [10]

    Geva, N., Deitch, D., Rubin, A., & Ziv, Y . (2023). Time and experience differentially 25 affect distinct aspects of hippocampal representational drift.Neuron,111(15), 2357– 2366.e5

    2023

  11. [11]

    Haberly, L. B. (2001). Parallel-distributed Processing in Olfactory Cortex: New Insights from Morphological and Physiological Analysis of Neuronal Circuitry.Chemical Senses,26(5), 551–576

    2001

  12. [12]

    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

    1982

  13. [13]

    T., Mosser, C.-A., & Brandon, M

    Keinath, A. T., Mosser, C.-A., & Brandon, M. P. (2022). The representation of context in mouse hippocampus is preserved despite neural drift.Nature Communications, 13(1)

    2022

  14. [14]

    Kriegeskorte, N., Mur, M., & Bandettini, P. (2008). Representational similarity analy- sis – connecting the branches of systems neuroscience.Frontiers in Systems Neuro- science. doi:10.3389/neuro.06.004.2008

    work page doi:10.3389/neuro.06.004.2008 2008

  15. [15]

    Li, Q., Sorscher, B., & Sompolinsky, H. (2024). Representations and generalization in artificial and brain neural networks.Proceedings of the National Academy of Sciences,121(27)

    2024

  16. [16]

    L., McNaughton, B

    McClelland, J. L., McNaughton, B. L., & O’Reilly, R. C. (1995). Why there are com- plementary learning systems in the hippocampus and neocortex: Insights from the successes and failures of connectionist models of learning and memory.Psychologi- cal Review,102(3), 419–457. 26

    1995

  17. [17]

    B., Mu ˜noz, M

    Morales, G. B., Mu ˜noz, M. A., & Tu, Y . (2025). Representational drift and learning- induced stabilization in the piriform cortex.Proceedings of the National Academy of Sciences,122(29)

    2025

  18. [18]

    Natrajan, M., & Fitzgerald, J. E. (2025). Stability through plasticity: Finding robust memories through representational drift.Proceedings of the National Academy of Sciences,122(45)

    2025

  19. [19]

    Nili, H., Wingfield, C., Walther, A., Su, L., Marslen-Wilson, W., & Kriegeskorte, N. (2014). A Toolbox for Representational Similarity Analysis.PLoS Computational Biology,10(4), e1003553

    2014

  20. [20]

    Raju, P. C. (2026a). Geometric Stability: The Missing Axis of Representations.arXiv preprint arXiv:2601.09173

    work page internal anchor Pith review Pith/arXiv arXiv

  21. [21]

    Raju, P. C. (2026b). Shesha: Self-Consistency Metrics for Representational Stability. Zenodo. doi:10.5281/zenodo.18227453

    work page doi:10.5281/zenodo.18227453

  22. [22]

    Raju, P. C. (2026c). The Geometric Canary: Predicting Steerability and Detecting Drift via Representational Stability.arXiv preprint arXiv:2604.17698

    work page internal anchor Pith review Pith/arXiv arXiv

  23. [23]

    Raju, P. C. (2026d). Geometric Coherence of Single-Cell CRISPR Perturbations Reveals Regulatory Architecture and Predicts Cellular Stress.arXiv preprint arXiv:2604.16642

    work page internal anchor Pith review Pith/arXiv arXiv

  24. [24]

    E., O’Leary, T., & Harvey, C

    Rule, M. E., O’Leary, T., & Harvey, C. D. (2019). Causes and consequences of repre- sentational drift.Current Opinion in Neurobiology,58, 141–147. 27

    2019

  25. [25]

    E., Ohashi, S

    Schoonover, C. E., Ohashi, S. N., Axel, R., & Fink, A. J. P. (2021). Representational drift in primary olfactory cortex.Nature,594(7864), 541–546

    2021

  26. [26]

    A., Zatka-Haas, P., Carandini, M., & Harris, K

    Steinmetz, N. A., Zatka-Haas, P., Carandini, M., & Harris, K. D. (2019). Distributed coding of choice, action and engagement across the mouse brain.Nature,576(7786), 266–273

    2019

  27. [27]

    Wagner, M., Chen, Y ., Karuvally, A., Cameron, M., & Sejnowski, T. J. (2026). Bal- ancing stability and flow in hippocampal networks via inductive bias and learned symmetry breaking.bioRxiv. doi:10.64898/2026.02.06.704443

    work page doi:10.64898/2026.02.06.704443 2026

  28. [28]

    Walther, A., Nili, H., Ejaz, N., Alink, A., Kriegeskorte, N., & Diedrichsen, J. (2016). Reliability of dissimilarity measures for multi-voxel pattern analysis.NeuroImage, 137, 188–200

    2016

  29. [29]

    Weiss, O., & Coen-Cagli, R. (2025). Measuring Stimulus Information Transfer Be- tween Neural Populations Through the Communication Subspace.Neural Computa- tion,37(9), 1600–1647

    2025

  30. [30]

    D., Cocker, E

    Ziv, Y ., Burns, L. D., Cocker, E. D., Hamel, E. O., Ghosh, K. K., Kitch, L. J., Gamal, A. E., & Schnitzer, M. J. (2013). Long-term dynamics of CA1 hippocampal place codes.Nature Neuroscience,16(3), 264–266. 28

    2013