{"id":"fac4940b-cc9a-41ac-9af1-7af3ba5f74d8","arxiv_id":"2505.13917","paper_version":1,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"The paper claims that Fisher information in Gaussian neural population codes produces a hyperbolic visual space, and fits that model to distance-judgment and horopter data.","lead":"This preprint proposes that the brain encodes object locations with Gaussian tuning curves, and that the Fisher information of those curves defines a hyperbolic visual space with the Fisher-Rao metric as its psychometric distance. A generalist might read it because it is an attempt to connect neural population codes to the geometry of visual space, a long-standing question in perception.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The central mechanism is an algebraic identity: in the 2D visual space, Eq. (23) reduces to 0 = 0 because T is defined via ρ = R + g, so the claimed role of Fisher information as a curvature-generating stress-energy tensor is not supported.","rationale":"The reader's REJECT is appropriate. The reader's weakest_assumption was the unsupported Gaussian tuning-curve assumption, while the rationale also flagged the field equation as an algebraic identity. I regard the field-equation vacuity as the more load-bearing concern because it undermines the claimed neural-origin mechanism even if the Gaussian assumption were granted. The paper correctly computes the Fisher–Rao metric for Gaussian distributions and correctly notes that this metric has constant negative curvature, but the central causal story—'Fisher information as stress-energy tensor creating curvature'—is not supported by Eqs. (20)–(23). An analytical re-derivation showing that Eq. (23) is an identity would settle this point without relying on any empirical data. The later empirical sections use iterative fitting of σe to the same Oleksiak data and borrow the horopter construction from an earlier preprint, so they cannot rescue the theoretical claim. No change to the reader's REJECT verdict is needed.","tokens_in":17138,"tokens_out":6225,"duration_ms":61228,"concrete_test":"Independently compute both sides of Eq. (23) for the metric in Eq. (12) using only the definitions in Eqs. (18)–(22). If the result is the identity 0 = 0 (equivalently, Tμν = 0 in 2D), the field equation is vacuous. A complementary check: repeat the derivation for a flat metric ds² = dμ² + dσ²; Eq. (23) must also hold identically, confirming that no metric is preferred by the equation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim—that Fisher information acts as a stress-energy tensor and thereby creates the curved visual space—rests on Eqs. (20)–(23). But Eq. (20) defines ρμν = Rμν + gμν, and Tμν is then defined as ρμν − (1/2)gμνρ. Substituting the definition gives Tμν = Gμν − (n−2)/2 gμν, which is precisely the left-hand side of Eq. (23). For the paper's n = 2 visual space the Einstein tensor vanishes identically, so Eq. (23) reduces to 0 = 0. The 'field equation' therefore imposes no constraint on the metric and cannot select hyperbolic geometry; the constant curvature R = −1 is simply a property of the Gaussian Fisher–Rao metric chosen in Eq. (12). Even before asking whether real tuning curves are Gaussian, the advertised neural-origin mechanism is vacuous. The empirical fit and horopter simulation in Sections 3.1 and 3.2 may be consistent with hyperbolic geometry, but they do not test the claimed causal role of Fisher information.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes that the non-Euclidean, hyperbolic structure of visual space arises from the Fisher information contained in Gaussian neural population codes. The authors model the population response to a point stimulus as a Gaussian tuning curve, compute the Fisher-Rao metric on the half-plane of parameters (mean, standard deviation), obtain constant scalar curvature R = -1, and introduce an information-geometric field equation (Eq. 23) in which Fisher information plays the role of a stress-energy tensor. They then fit the model to Oleksiak et al.'s distance-estimation data by iteratively adjusting the tuning-curve width, and they simulate Helmholtz horopters as geodesics of the hyperbolic visual space using Luneburg's transformation. The paper concludes that Fisher information creates the curved visual space and that the Fisher-Rao distance is the correct psychometric distance.","tokens_in":17275,"tokens_out":4741,"duration_ms":45556,"significance":"If the central claim were correct, the paper would provide a neural-origin mechanism for the long-studied hyperbolic geometry of visual space, connecting population coding, information geometry, and Luneburg's classical model. The mathematical computation of the Gaussian Fisher metric and its constant negative curvature is correct, and the formal connection between the Fisher-Rao upper half-plane and the Poincaré disk is a useful pedagogical observation. The paper also demonstrates the possibility of reproducing psychophysical distance-error data and horopter shapes within a hyperbolic-geometry framework. However, the advertised causal mechanism—that Fisher information acts as a stress-energy tensor and thereby generates curvature—is an algebraic identity rather than a physical or neural mechanism, and the empirical agreement is obtained by fitting a free parameter. These issues undermine the paper's central claim as stated.","major_comments":[{"comment":"The central mechanism is an algebraic identity. Defining ρμν = Rμν + gμν (Eq. 20) and Tμν = ρμν − (1/2)gμνρ (Eq. 22), one obtains by substitution Tμν = Gμν − (n−2)/2 gμν, where Gμν is the Einstein tensor. Hence Eq. (23) is identically satisfied for every Riemannian metric and imposes no constraint on the geometry. In the paper's n = 2 visual space the Einstein tensor vanishes identically, so Eq. (23) reduces to 0 = 0. Consequently the claim that Fisher information 'creates' the curved visual space is not supported: the constant negative curvature R = −1 follows directly from the Gaussian ansatz in Eq. (12), not from any dynamical equation involving information. The statement in the text that 'the curvature of the space changes when information appears in it' is also unsupported because no information-dependent variation of the metric is ever computed.","section":"Section 3, Eqs. (20)–(23)"},{"comment":"The claimed empirical verification is a curve fit, not a prediction. The authors first compute distance errors using σe from Eq. (24), find disagreement beyond 4°, and then 'keep on modifying the value of σe iteratively till the difference between the simulated and experimental values of the error reaches a minimum.' The reported correlation 0.9224 between simulated and experimental curves is therefore obtained by tuning a free parameter per data set. The subsequent attentional-gain analysis (Eq. 25) does not provide an independent test; it solves for the attentional width σA using the already-fitted σR. This procedure cannot distinguish the Fisher-Rao metric from any other distance function with a suitably adjusted width parameter.","section":"Section 3.1, Steps 4–5 and Figure 6"},{"comment":"The load-bearing assumption that neural tuning curves are exactly Gaussian, with the same width parameter retained from retinal ganglion cells through dLGN to LIP, is asserted rather than derived. The DOG-based response equations (3)–(5) contain factors of (1 − exp(...)) and incomplete gamma functions; the step to a normalized Gaussian population code is an additional idealization. Because the Fisher metric (Eq. 12) and the scalar curvature R = −1 depend entirely on this Gaussian form, the paper's central geometric results rest on an unvalidated phenomenological assumption. No neural-recording evidence or formal derivation is supplied to justify the claim that the Gaussian shape is preserved unchanged along the dorsal pathway.","section":"Section 2 and Step 4, Section 3.1"}],"minor_comments":[{"comment":"There is a typographical error: 'Enviol et al.' should be 'Einevoll et al.' (reference [24]).","section":"Section 2, references"},{"comment":"The sentence 'Further to note that the interaction of attentional influence and the neural response has two contrastive natures' is duplicated verbatim in consecutive paragraphs; one occurrence should be deleted.","section":"Section 3.1"},{"comment":"The caption refers to 'σe as given by Equation (20)', but the relevant formula for σe is Equation (24); the cross-reference should be corrected.","section":"Figure 6 caption"},{"comment":"The expansion of the Kullback-Leibler divergence contains unclear notation ('1/P2(x; θ)') and the signs in the intermediate expression are not clearly justified; a cleaner derivation with explicit Taylor expansion would improve readability.","section":"Eq. (9)"},{"comment":"Reference [57] appears to have the same title as reference [56] (Zajaczkowska), which is likely a citation error; the actual title of [57] should be verified.","section":"Reference [57]"}],"recommendation":"reject","confidential_remarks":"The paper contains a correct computation of the Gaussian Fisher-Rao metric and its constant negative curvature, and the connection to Luneburg's horopter model is interesting. However, the central claimed mechanism—Fisher information as a stress-energy tensor generating curvature—is an algebraic identity that imposes no constraint, and the empirical support relies on iterative parameter fitting. These are load-bearing issues that cannot be repaired by local revisions. A revised paper that reframes the result as a phenomenological model of Gaussian population codes implying hyperbolic visual space could be reconsidered."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper is a competent restatement of a standard result in information geometry, wrapped in a neural-origin story that does not survive contact with its own equations.\n\nWhat is genuinely good: the Fisher-Rao metric for Gaussian distributions is computed correctly, the scalar curvature R = -1 is correct, and the paper makes a clean connection between this geometry and Luneburg's classical hyperbolic model of visual space. If you need a short derivation of why the Gaussian Fisher-Rao metric is hyperbolic, this is a readable source.\n\nWhere it falls apart is the advertised mechanism. Equation (23) is presented as a field equation showing that Fisher information acts as a stress-energy tensor and creates curvature. But T is defined via rho = R + g, so the equation is an algebraic identity. In two dimensions the Einstein tensor vanishes identically, so Eq. (23) becomes 0 = 0. The paper does not derive the hyperbolic geometry from Fisher information; it simply observes that the Gaussian Fisher-Rao metric is hyperbolic. This is a load-bearing flaw, not a cosmetic one. The stress-test note is right.\n\nThe empirical validation is also weak. The distance-error correlation of 0.9224 comes from iteratively adjusting sigma_e until the simulated curve matches Oleksiak's data, so it is a calibration statistic, not a prediction. The horopter simulation imports Luneburg's transformation and the authors' own prior preprint, so it does not independently confirm the new mechanism. No code, data, or error bars are provided, and the post hoc attention mechanism for separations beyond 4 degrees is another free parameter. The paper also assumes, without neural evidence or a formal derivation, that Gaussian tuning curves survive unchanged from retina through dLGN to LIP; if that assumption fails, the whole structure changes.\n\nWho is this paper for? Possibly someone looking for a neat exposition of Fisher-Rao geometry and its formal resemblance to Luneburg's model. As a research contribution claiming a neural origin of non-Euclidean visual space, it does not hold up.\n\nRecommendation: I would not send this to peer review. The central claim is an identity, and the empirical support is a fit. A referee would spend time uncovering exactly what the authors could have checked themselves. If the authors want to make a real contribution, they need a nontrivial derivation from neural population dynamics and a parameter-free prediction against new data.","headline":"The paper dresses a standard Fisher-Rao geometry result in a neural-origin story, but the central field equation is an algebraic identity and the empirical fit is calibration; the thesis does not survive close reading.","tokens_in":17934,"tokens_out":2755,"would_cite":false,"duration_ms":27254,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper argues that visual space is intrinsically hyperbolic because Fisher information in Gaussian neural population codes acts like a stress-energy tensor, forcing constant negative curvature and fixing the Fisher-Rao distance as the…","keywords":["visual space","information geometry","Fisher information","hyperbolic geometry","neural population code","psychometric distance","Helmholtz horopter","Muller-Lyer illusion"],"falsifier":"Record or simulate population responses to point stimuli at several eccentricities and estimate the local Fisher information from the empirical tuning curves; if the metric does not scale as $1/\\sigma^2$ or the scalar curvature is not $-1$, the central claim fails. More directly, if one can find a condition in which tuning curves are clearly non-Gaussian but perceived visual space remains hyperbolic, or vice versa, the proposed neural origin of the curvature is refuted.","tokens_in":16797,"feed_emoji":"🧠","tokens_out":6008,"duration_ms":56947,"temperature":0.7,"pith_summary":"This paper argues that the non-Euclidean character of visual space is not an assumption but a consequence of how neurons encode location. Treating each point stimulus as a Gaussian tuning curve, with mean position and width as the encoding parameters, the population code defines a statistical manifold. The Fisher-Rao metric of that manifold is $ds_f^2 = (d\\mu^2 + 2d\\sigma^2)/\\sigma^2$, whose scalar curvature is $R = -1$: visual space is hyperbolic. The paper uses this metric as the psychometric distance and reproduces published data on the error in estimating the distance between two dots, and derives Helmholtz horopters as geodesics of the hyperbolic space. If the account is right, the geometry of visual space is fixed by the Gaussian form of neural tuning curves, giving a neural origin for a classical psychophysical conjecture.","feed_headline":"Visual space is hyperbolic, and bell-shaped neural codes explain why","feed_subtitle":"A Gaussian population code fixes the Fisher-Rao metric, gives curvature -1, and matches psychophysical distance data.","key_machinery":"The engine is the identification of a point stimulus with a Gaussian probability density $p(x;\\mu,\\sigma)$ and the visual space with the upper half-plane $H_f = \\{(\\mu,\\sigma) \\in \\mathbb{R}^2 : \\sigma > 0\\}$. The Fisher-Rao metric computed from this density is $ds_f^2 = (d\\mu^2 + 2d\\sigma^2)/\\sigma^2$, which is the Poincar\\'e upper half-plane metric after rescaling $\\omega = \\mu + i\\sqrt{2}\\sigma$; the Poincar\\'e disk model is obtained by a M\\'obius transformation. This metric carries the argument because its constant negative curvature $R=-1$ is the claimed neural origin of hyperbolic visual space, and its geodesics are the claimed loci of the Helmholtz horopter. The metric also supplies the psychometric distance, since for fixed width it reduces to $|\\mu_1-\\mu_2|/\\sigma$, which the paper uses to simulate distance-estimation biases.","core_discovery":"The central claim is that Fisher information contained in the neural population code serves as an energy-momentum tensor, generating the space-dependent metric tensor and therefore a curved visual space. With Gaussian tuning curves the Fisher-Rao metric becomes $ds_f^2 = (d\\mu^2 + 2d\\sigma^2)/\\sigma^2$, and the scalar curvature computed from the Christoffel symbols and Ricci tensor is constant, $R = -1$, so the visual space is a hyperbolic manifold. The paper recasts the geometric field equation as $R_{\\mu\\nu} - \\frac{1}{2}g_{\\mu\\nu}R - \\frac{1}{2}(n-2)g_{\\mu\\nu} = T_{\\mu\\nu}$, where $T_{\\mu\\nu}$ is a stress-energy tensor built from Fisher information, so the presence of information changes curvature. In this space, the Fisher-Rao distance between two Gaussian codes with the same width is $|\\mu_1-\\mu_2|/\\sigma$, and the paper adopts this as the psychometric distance. Simulated distance-estimation errors match the experimental curve up to roughly $4^\\circ$ of separation, and beyond that an attention-induced narrowing of the tuning width is added; the resulting correlation with the psychophysical data is $0.9224$. The Helmholtz horopter is derived as a geodesic of this hyperbolic space, matching data for three observers.","pith_inferences":["A testable consequence of the paper's logic is that manipulating tuning width without changing stimulus position, for example by adaptation or pharmacological intervention, should measurably distort perceived distance in the direction predicted by the $1/\\sigma$ Fisher-Rao distance.","If the argument is read as a general mechanism rather than a vision-specific one, the same Fisher-Rao construction would predict non-Euclidean perceptual geometry in any modality whose population codes are approximately Gaussian, such as auditory or haptic space.","The derivation depends on the Gaussian form of tuning curves, so a natural extension is to compute the Fisher-Rao metric and curvature for empirically measured non-Gaussian population codes; the sign and magnitude of the curvature would then be a physiological observable rather than an assumption.","The field-equation analogy suggests that the perceived curvature of space could be read as a proxy for the local information content of the neural code, connecting information-theoretic measures of coding efficiency to psychophysical geometry."],"forward_implications":["Visual space in frameless or dark-room conditions is hyperbolic with constant curvature $-1$, so Euclidean geometry is not the right model for perceived spatial relations in such conditions.","The Fisher-Rao distance is the appropriate psychometric distance, meaning that perceived separation between two points should scale inversely with the width of the underlying neural tuning curves.","The Helmholtz horopter is a geodesic of this hyperbolic space, and the same construction can account for parallel and distance alleys and the M\\\"uller-Lyer illusion in one framework.","Any factor that changes the tuning width $\\sigma$, such as attention, eccentricity, or receptive-field size, changes the metric and hence the perceived distance, giving a concrete neural handle on geometric visual illusions.","If the tuning curves deviate from Gaussian under rich background cues, the visual space loses constant curvature, which the paper connects to measured curvature varying from elliptic in near space to hyperbolic in far space."],"supporting_citations":[{"why":"Supplies the classical hyperbolic constant-curvature model of binocular visual space and the coordinate transformations between physical and visual space that this paper grounds in neural population codes.","marker":"[3]"},{"why":"Provides the Difference-of-Gaussians model of retinal and dLGN spatial receptive fields whose half-wave-rectified response is taken to produce the Gaussian population code.","marker":"[24]"},{"why":"Establishes the population-code convention that normalized tuning curves are Gaussian, so that mean and standard deviation parametrize the code.","marker":"[27]"},{"why":"Provides the two-dot distance-estimation data in fixate condition and the eccentricity-dependent receptive-field scaling used in the simulation.","marker":"[31]"},{"why":"Justifies using the distance between neural response distributions as the retinal or psychometric distance between stimuli.","marker":"[34]"},{"why":"Supplies the prior derivation of the geodesic equation in the upper half-plane and the M\\'obius transformation to the Poincar\\'e disk used for the horopter simulation.","marker":"[36]"},{"why":"Provides an experimental measurement reporting negative curvature for frameless visual space, supporting the hyperbolic signature the model predicts.","marker":"[37]"},{"why":"Supplies the experimental horopter data and personal constants for three observers used to validate the simulated horopter curves.","marker":"[56]"}],"fun_headline_variants":["Fisher information warps visual space into hyperbolic geometry","Neural codes fix visual space as hyperbolic with curvature -1","Bell-shaped neural codes make visual space hyperbolic, curvature -1","Fisher information acts as energy-momentum, curving visual space","Curved visual space emerges from Fisher information in neural codes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole construction rests on treating a point stimulus as encoded by a bell-shaped Gaussian tuning curve whose peak marks position and whose width marks uncertainty, and on assuming that this bell shape survives unchanged from the retina through the dLGN to the parietal cortex.","fun_headline_variants_meta":{"raw":{"variants":["Fisher information warps visual space into hyperbolic geometry","Neural codes fix visual space as hyperbolic with curvature -1","Bell-shaped neural codes make visual space hyperbolic, curvature -1","Fisher information acts as energy-momentum, curving visual space","Curved visual space emerges from Fisher information in neural codes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001182,"raw_usage":{"total_tokens":4938,"prompt_tokens":1055,"completion_tokens":3883,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":671,"completion_tokens_details":{"reasoning_tokens":3800}},"tokens_in":671,"tokens_out":3883,"duration_ms":26439,"temperature":1.0,"reasoning_tokens":3800,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:08:05.508034+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Record or simulate population responses to point stimuli at several eccentricities and estimate the local Fisher information from the empirical tuning curves; if the metric does not scale as $1/\\sigma^2$ or the scalar curvature is not $-1$, the central claim fails. More directly, if one can find a condition in which tuning curves are clearly non-Gaussian but perceived visual space remains hyperbolic, or vice versa, the proposed neural origin of the curvature is refuted.","supporting_citations":[{"cited_title":"Mathematical analysis of binocular vision","cited_arxiv_id":null,"evidence_quote":"Supplies the classical hyperbolic constant-curvature model of binocular visual space and the coordinate transformations between physical and visual space that this paper grounds in neural population codes."},{"cited_title":"Mathematical models for the spatial receptive-field organization of nonlagged x-cells in dorsal lateral geniculate nucleus of cat","cited_arxiv_id":null,"evidence_quote":"Provides the Difference-of-Gaussians model of retinal and dLGN spatial receptive fields whose half-wave-rectified response is taken to produce the Gaussian population code."},{"cited_title":"Information processing with population codes","cited_arxiv_id":null,"evidence_quote":"Establishes the population-code convention that normalized tuning curves are Gaussian, so that mean and standard deviation parametrize the code."},{"cited_title":"Distance estimation is influenced by encoding conditions","cited_arxiv_id":null,"evidence_quote":"Provides the two-dot distance-estimation data in fixate condition and the eccentricity-dependent receptive-field scaling used in the simulation."},{"cited_title":"Retinal metric: a stimulus distance measure derived from population neural responses","cited_arxiv_id":null,"evidence_quote":"Justifies using the distance between neural response distributions as the retinal or psychometric distance between stimuli."},{"cited_title":"Representation of 2D frame less visual space as a neural manifold and its information geometric interpretation","cited_arxiv_id":"2011.13585","evidence_quote":"Supplies the prior derivation of the geodesic equation in the upper half-plane and the M\\'obius transformation to the Poincar\\'e disk used for the horopter simulation."},{"cited_title":"Curvature of binocular visual space","cited_arxiv_id":null,"evidence_quote":"Provides an experimental measurement reporting negative curvature for frameless visual space, supporting the hyperbolic signature the model predicts."},{"cited_title":"Experimental test of luneburg’s theory","cited_arxiv_id":null,"evidence_quote":"Supplies the experimental horopter data and personal constants for three observers used to validate the simulated horopter curves."}],"review_version":1}