{"id":"99555dcc-1e19-4570-bb5f-131063627000","arxiv_id":"2405.05097","paper_version":8,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"Introduces joint distribution neurons based on Hierarchical Correlation Reconstruction for multidirectional propagation of values and densities, extending KAN with probabilistic modeling.","lead":"The paper proposes joint distribution neurons that model local probability densities using Hierarchical Correlation Reconstruction to enable bidirectional information flow and probabilistic outputs in neural networks. This could help artificial systems better match biological neurons in handling uncertainty, risk, and local learning.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"The joint density model ρ(x)=∑ a_j f_j(x) lacks specified basis functions or algorithms ensuring tractable conditioning (substitution + normalization) and local training for d>3.","rationale":"The identified concern is exactly the reader's weakest assumption about practical training and implementation cost; the absence of any concrete basis, algorithm, or complexity analysis in the proposal makes that assumption the single load-bearing point.","tokens_in":1837,"tokens_out":314,"duration_ms":15686,"concrete_test":"Choose any concrete basis {f_j} and finite B for d=4; implement the substitution-plus-normalization procedure to compute a 2-variable conditional; time the operation on 100 random queries and compare accuracy against a standard GMM with the same number of parameters.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that this linear basis expansion delivers bidirectional propagation via variable substitution for conditionals, moment-based distribution propagation, and local training (direct, tensor decomposition, or information bottleneck) without prohibitive cost. Conditioning on a joint density requires not only substitution but also normalization via marginalization integrals over the remaining variables; for generic f_j and multi-index set B in dimension d this is intractable unless the basis has special product/separable structure that is never stated. No explicit choice of f_j, training procedure, non-negativity/integral-to-1 enforcement, or complexity bound is supplied, so the three biological properties remain formal possibilities rather than demonstrated capabilities of the given functional form.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript proposes joint distribution neurons as an extension of Kolmogorov-Arnold Networks, in which each neuron models a local joint distribution via the linear expansion ρ(x)=∑_{j∈B} a_j f_j(x) for x∈[0,1]^d. The central claim is that this form supplies three missing biological properties: (1) bidirectional propagation obtained by variable substitution to produce conditionals, (2) propagation of full distributions represented as moment vectors (mean, variance, …), and (3) local training routes including direct fitting, tensor decomposition, and the information bottleneck. The same construction is suggested as a drop-in replacement for softmax layers in transformers.","tokens_in":1987,"tokens_out":639,"duration_ms":17114,"significance":"If the functional form could be equipped with concrete, tractable basis functions and training procedures that realize the three listed properties at scale, the work would supply a principled probabilistic primitive that unifies interpretation, uncertainty propagation, and locality of learning—potentially improving robustness and sample efficiency over standard MLPs or KANs. The absence of any such concrete realization, however, leaves the significance prospective rather than demonstrated.","major_comments":[{"comment":"Abstract: the claim that substitution of variables directly yields conditional distributions omits the marginalization integrals required for normalization; without an explicit product or separable structure on the unspecified f_j, these integrals are intractable for d>3 and therefore load-bearing for the bidirectional-propagation claim.","section":"Abstract"},{"comment":"Abstract: no choice of basis functions f_j, multi-index set B, non-negativity constraint, or normalization procedure for the coefficients a_j is supplied, rendering the three biological properties formal possibilities rather than demonstrated capabilities of the given expansion.","section":"Abstract"},{"comment":"Abstract: the information-bottleneck training route is asserted to be “local and promising,” yet no algorithm, objective, or complexity bound is derived that would show how the bottleneck can be optimized using only the linear coefficients a_j and the (unspecified) f_j.","section":"Abstract"},{"comment":"Abstract: the manuscript contains neither derivations, pseudocode, complexity analysis, nor any empirical result that would substantiate that the proposed neuron can be trained or evaluated at practical cost while preserving the claimed moment-propagation and conditioning properties.","section":"Abstract"}],"minor_comments":[{"comment":"Abstract: grammatical phrasing “a million of biological neurons” and “the proposed can predict” should be corrected.","section":"Abstract"},{"comment":"Abstract: citation markers (e.g., ~cite{RL}, ~cite{axon}) appear without an accompanying reference list or context.","section":"Abstract"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a short conceptual sketch rather than a completed technical paper; its fit for a full-length journal may be marginal unless the authors supply the missing algorithmic and empirical substance."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive critique. The manuscript is a concise conceptual proposal introducing the joint-distribution neuron form and arguing that it formally enables three biological properties. We respond point-by-point below, acknowledging where the current text is limited to the general expansion and where concrete realizations remain future work.","responses":[{"response":"We agree that obtaining a properly normalized conditional from the joint expansion generally requires marginalization integrals. The manuscript states that substitution yields conditionals, but does not claim this is automatic for arbitrary bases; the intent is that, once a concrete basis admitting closed-form or efficient marginals is chosen, the same linear coefficients allow both forward and backward propagation. The current text leaves the required structure on f_j implicit, which is a limitation of the presentation.","revision_made":"no","referee_comment":"[Abstract] Abstract: the claim that substitution of variables directly yields conditional distributions omits the marginalization integrals required for normalization; without an explicit product or separable structure on the unspecified f_j, these integrals are intractable for d>3 and therefore load-bearing for the bidirectional-propagation claim."},{"response":"The manuscript deliberately presents the most general linear expansion that still permits the three listed operations (variable substitution, moment-vector propagation, and local coefficient updates). Specific bases (e.g., multivariate polynomials or wavelets on [0,1]^d), non-negativity constraints, and normalization schemes are indeed omitted because the paper’s scope is to establish the functional form and its qualitative advantages over standard KAN neurons. Concrete instantiations are required for implementation and are noted as future work.","revision_made":"no","referee_comment":"[Abstract] Abstract: no choice of basis functions f_j, multi-index set B, non-negativity constraint, or normalization procedure for the coefficients a_j is supplied, rendering the three biological properties formal possibilities rather than demonstrated capabilities of the given expansion."},{"response":"The claim is that the information-bottleneck objective can be expressed directly in terms of the coefficients a_j once the basis is fixed, because the modeled density is linear in those coefficients; this would in principle allow a local update without back-propagation through the rest of the network. No explicit algorithm or complexity analysis is supplied, as the manuscript only identifies the route as conceptually local. Deriving a practical optimizer is left for subsequent development.","revision_made":"no","referee_comment":"[Abstract] Abstract: the information-bottleneck training route is asserted to be “local and promising,” yet no algorithm, objective, or complexity bound is derived that would show how the bottleneck can be optimized using only the linear coefficients a_j and the (unspecified) f_j."},{"response":"The manuscript is a short conceptual note whose contribution is the identification of the linear joint-density expansion and the three formal properties it enables. It therefore contains no empirical results, pseudocode, or complexity bounds. We accept that demonstrating practical cost and preservation of the properties requires concrete bases, training procedures, and experiments, none of which are present.","revision_made":"no","referee_comment":"[Abstract] Abstract: the manuscript contains neither derivations, pseudocode, complexity analysis, nor any empirical result that would substantiate that the proposed neuron can be trained or evaluated at practical cost while preserving the claimed moment-propagation and conditioning properties."}],"tokens_in":1553,"tokens_out":722,"duration_ms":15766,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that the paper outlines an idea for neurons that store a local joint density as a linear combination of basis functions, then use substitution to get conditionals and moments to carry distributions forward. This is positioned as a way to add three bio-inspired traits to current networks: bidirectional propagation, variance handling, and local training options like information bottleneck. The framing is general enough that it could also apply to embeddings or other layers. That part is clear and connects to existing work on KAN and probabilistic models without obvious circularity. What the paper does reasonably is name concrete gaps in modern architectures relative to biology and suggest how a density model might close them. The stress-test concern lands: the given form ρ(x) = sum a_j f_j(x) does not come with any choice of f_j, any product structure that would make marginalization cheap, or any procedure for normalization and training. Conditioning therefore stays formal rather than practical, especially past low dimensions, and nothing in the text shows the three properties actually appear at reasonable cost. There are also no derivations, toy calculations, or experiments to test feasibility. The paper is therefore an idea rather than a method. Readers already working on bio-plausible or probabilistic extensions of KAN might find it useful as a starting point for discussion. It shows honest engagement with the cited biology references and does not overclaim results. I would bring it to a reading group to talk through whether any basis choice could make the operations tractable. I would not cite it in its current form. It deserves peer review because the direction is worth exploring and feedback could push the authors toward concrete realizations or bounds.","headline":"This is a high-level sketch for joint-distribution neurons in a KAN extension that could support bidirectional flow and moment propagation, but the functional form is left too open to deliver on the claims without major additional work.","tokens_in":2464,"tokens_out":415,"would_cite":false,"duration_ms":19892,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[],"headline":"HCR moment-expansion neurons for bidirectional density propagation share no machinery with RS J-cost or distinction-forcing chain","alignment":"orthogonal","rationale":"The paper's central object is the linear expansion rho(x) = sum a_j f_j(x) on orthonormal polynomial product bases, with a_j interpreted as mixed moments, enabling substitution-based conditioning and moment-vector propagation. This is a standard functional approximation technique in ML, unrelated to the RS recognition cost J(x) = 1/2(x + 1/x) - 1, its functional-equation uniqueness (Cost/FunctionalEquation.washburn_uniqueness_aczel), the phi-ladder constants, or the single-distinction forcing of spacetime (Foundation/RealityFromDistinction, AbsoluteFloorClosure). No RS theorem is invoked or paralleled; the domain (neural density models) lies outside RS scope.","tokens_in":53022,"confidence":"high","tokens_out":189,"duration_ms":5453,"cache_read_input_tokens":38528,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Joint distribution neurons model local densities to enable bidirectional propagation, moment-based uncertainty handling, and local training alternatives.","keywords":["joint distribution neurons","hierarchical correlation reconstruction","bidirectional propagation","Kolmogorov-Arnold networks","moment propagation","local training","information bottleneck","neural network architecture"],"falsifier":"A controlled benchmark in which networks built from these neurons require substantially more parameters or training time than MLPs or KANs to reach the same test accuracy on a standard classification or regression task.","tokens_in":2725,"feed_emoji":"🧠","tokens_out":626,"duration_ms":25242,"temperature":0.7,"pith_summary":"The paper proposes extending KAN-style neurons with an explicit model of the joint distribution over their inputs, written as a linear combination of basis functions over the unit hypercube. This model supports substituting observed values to recover conditional expectations or distributions for the remaining variables, propagating vectors of moments such as mean and variance, and training through direct fitting, tensor methods, or an information-bottleneck objective. A sympathetic reader would care because the construction directly targets three gaps between current artificial networks and biological ones: unidirectional flow, deterministic activation, and global back-propagation.","feed_headline":"Joint-density neurons enable bidirectional and probabilistic flow","feed_subtitle":"Modeling local input correlations lets networks compute conditionals, carry variance, and train without global back-propagation.","key_machinery":"The joint distribution representation ρ(x) = sum_{j in B} a_j f_j(x) that encodes correlations among inputs and supplies conditional values or moments on demand.","core_discovery":"Neurons containing the joint-density model ρ(x) = sum a_j f_j(x) for x in [0,1]^d allow repair of missing inputs by conditional evaluation, propagate distributions via moment vectors, and admit local training procedures including direct optimization and information-bottleneck updates, while remaining compatible with existing architectures such as transformers.","pith_inferences":["Such neurons could support decentralized or continual learning scenarios where only local statistics are updated.","Interpreting transformer features as moments suggests a route to uncertainty-aware attention mechanisms.","The approach opens a concrete path for testing whether explicit joint-density modeling improves robustness on tasks that reward risk sensitivity."],"forward_implications":["Inputs can be repaired on the fly by solving for the conditional distribution given the observed coordinates.","Uncertainty can be propagated forward by carrying vectors of moments rather than single point estimates.","Training rules other than back-propagation become available, including direct fitting of the coefficients a_j and local information-bottleneck objectives.","The same representation can replace softmax layers in embedding models by treating learned features as mixed moments of an underlying joint density."],"fun_headline_variants":["Joint density neurons enable bidirectional value propagation","Neurons with local joint densities compute conditionals and variances","Joint density models allow multidirectional propagation of moments","Joint distribution neurons allow local training without backpropagation"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The joint distribution model can be trained and evaluated at practical cost while preserving accuracy comparable to standard layers.","fun_headline_variants_meta":{"raw":{"variants":["Joint density neurons enable bidirectional value propagation","Neurons with local joint densities compute conditionals and variances","Joint density models allow multidirectional propagation of moments","Joint distribution neurons allow local training without backpropagation"]},"model":"grok-4.3","cost_usd":0.0098,"raw_usage":{"total_tokens":4391,"prompt_tokens":727,"num_sources_used":0,"completion_tokens":57,"cost_in_usd_ticks":97999500,"prompt_tokens_details":{"text_tokens":727,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3607,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":727,"tokens_out":57,"duration_ms":21343,"temperature":1.0,"reasoning_tokens":3607,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-24T00:54:51.927105+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A controlled benchmark in which networks built from these neurons require substantially more parameters or training time than MLPs or KANs to reach the same test accuracy on a standard classification or regression task.","supporting_citations":[],"review_version":1}