{"id":"08ff2b22-5cb8-400a-9a95-97dcdd1e8528","arxiv_id":"2606.04651","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"Unsupervised Hebbian networks show a sharp spectral transition in the interaction matrix that marks the onset of generalization when dataset quality allows separation of informative and noisy eigenvalue bulks.","lead":"The paper derives the asymptotic spectrum of the interaction matrix in an unsupervised Hebbian network from noisy realizations of hidden vectors using random matrix theory and replica methods, identifying a sharp spectral transition that controls generalization. A smart generalist might read it to see how spectral properties can predict when such networks recover ground-truth structure from corrupted data.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Asymptotic RMT/replica spectrum may not capture finite-N effects on generalization","rationale":"The reader's weakest assumption is precisely the load-bearing step: thermodynamic-limit spectrum → finite-size generalization. No other internal inconsistency is visible from the abstract and stated claim; the concern is therefore the same one.","tokens_in":1619,"tokens_out":290,"duration_ms":28746,"concrete_test":"Numerically diagonalize the Hebbian matrix for N=256,512,1024 at fixed noise variance σ and sample count M/N=α; locate the parameter value where the spectral gap opens; compare to the analytic transition line; then run regularized dynamics and measure ground-truth overlap—if the finite-N transition point deviates by >10% or the overlap jump disappears, the prediction fails.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The central claim requires that the sharp spectral transition (two bulks vs merged) derived in the N→∞ limit via RMT and replicas directly predicts reconstruction performance from corrupted samples once regularization is added. This holds only if the infinite-volume spectrum accurately governs the finite-N, finite-M regime for the specific model of noisy ground-truth vectors. Finite-size fluctuations, 1/N corrections to the bulk edges, or sample-to-sample variability in the interaction matrix could shift or smear the transition, breaking the predictive link even when the asymptotic formulas are exact.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript considers unsupervised Hebbian networks with pairwise interactions derived from noisy realizations of hidden ground-truth vectors. Combining random matrix theory and replica methods, the authors derive the asymptotic spectrum of the interaction matrix and identify a sharp spectral transition between a two-bulk phase (informative and noisy directions separated) and a single-bulk phase. They claim that, with regularization, the emergence of this spectral split predicts the network's ability to reconstruct ground-truth vectors from corrupted samples.","tokens_in":1745,"tokens_out":303,"duration_ms":34685,"significance":"If the central results hold, the paper provides a spectral criterion for generalization in this class of networks, extending beyond classical Hopfield models to unsupervised structure extraction. The application of standard RMT and replica techniques to derive the spectrum is a methodological strength, offering potential insights into the role of eigenvalue distributions in learning performance.","major_comments":[{"comment":"The central claim that the spectral transition (two bulks vs. merged) predicts generalization performance with regularization assumes the N\to∞ RMT/replica spectrum governs finite-N, finite-M reconstruction from corrupted samples. No finite-size analysis, 1/N corrections, or numerical checks validating that the transition remains predictive (rather than smeared by fluctuations) are provided for the specific model of noisy ground-truth vectors.","section":"Asymptotic spectrum derivation and generalization claim"}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for highlighting the distinction between the asymptotic analysis and finite-size behavior. We respond point by point below.","responses":[{"response":"We agree the spectrum derivation and transition are obtained exactly in the thermodynamic limit via RMT and replicas. The manuscript contains no finite-N corrections, 1/N analysis, or numerical checks for the noisy ground-truth model. The central claim is that the sharp asymptotic transition supplies the spectral criterion for generalization (with regularization) in the large-system regime where the theoretical tools apply. We maintain that this limit result is the appropriate and rigorous statement of the criterion; finite-size smearing is outside the scope of the asymptotic analysis presented.","revision_made":"no","referee_comment":"[Asymptotic spectrum derivation and generalization claim] The central claim that the spectral transition (two bulks vs. merged) predicts generalization performance with regularization assumes the N→∞ RMT/replica spectrum governs finite-N, finite-M reconstruction from corrupted samples. No finite-size analysis, 1/N corrections, or numerical checks validating that the transition remains predictive (rather than smeared by fluctuations) are provided for the specific model of noisy ground-truth vectors."}],"tokens_in":13539,"tokens_out":260,"duration_ms":168557,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main result is a derivation of the asymptotic spectrum of the interaction matrix built from noisy ground-truth vectors, showing a sharp transition between two bulks and a merged bulk that, with regularization, tracks whether the network reconstructs the hidden vectors.\n\nThis is new for this class of unsupervised Hebbian nets. Earlier work on Hopfield-like models focused on memory capacity, not on extracting latent structure from corrupted samples, so the spectral criterion tied to generalization is a distinct step.\n\nThe approach is standard but executed cleanly: RMT plus replicas give the bulk edges as functions of dataset size and noise level, which is useful for organizing when the informative directions separate from noise.\n\nThe soft spot is exactly the one in the stress-test note. Everything is derived in the N to infinity limit, and the claim that the split predicts reconstruction performance assumes the asymptotic spectrum governs finite-N, finite-M behavior. Finite-size fluctuations or sample variability in the matrix could shift or blur the transition, and the abstract gives no indication of numerical checks or 1/N corrections to test that. That gap makes the predictive part tentative rather than settled.\n\nThe math itself looks solid for the thermodynamic case, with no obvious circularity.\n\nThis is for theorists working on statistical mechanics of unsupervised learning and Hebbian models. Someone already using replica methods on neural nets would get a concrete spectral handle to compare against.\n\nIt deserves peer review because the claim is specific, the methods match the problem, and the finite-size issue is addressable in revision rather than fatal.","headline":"The paper derives a spectral transition via RMT and replicas that controls generalization in this unsupervised Hebbian model, but the link from asymptotic spectrum to finite-sample performance is the main open question.","tokens_in":2216,"tokens_out":398,"would_cite":false,"duration_ms":33591,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A sharp spectral transition in the interaction matrix controls when unsupervised Hebbian networks generalize from noisy data.","keywords":["Hebbian networks","spectral transition","generalization","random matrix theory","replica methods","unsupervised learning","noisy data","interaction matrix"],"falsifier":"For fixed dataset size and noise level, compute the eigenvalues of a finite interaction matrix and check whether the observed reconstruction error drops precisely at the noise threshold where the predicted spectral split appears.","tokens_in":2533,"feed_emoji":"","tokens_out":614,"duration_ms":37113,"temperature":0.7,"pith_summary":"This paper examines unsupervised Hebbian networks built from noisy realizations of hidden ground-truth vectors. It derives the asymptotic spectrum of the interaction matrix via random matrix theory and replica methods. Depending on data quality and size, the spectrum shows either two separated bulks or a single merged bulk. The transition between these regimes, when paired with regularization, determines the network's ability to reconstruct the ground-truth vectors from corrupted samples.","feed_headline":"Spectral split predicts generalization in Hebbian nets","feed_subtitle":"A sharp transition between split and merged eigenvalue bulks marks when networks reconstruct hidden vectors from corrupted samples.","key_machinery":"The asymptotic eigenvalue spectrum of the interaction matrix, obtained through random matrix theory and replica methods in the thermodynamic limit, which splits into separate bulks or merges into one to signal generalization onset.","core_discovery":"In an unsupervised Hebbian network whose pairwise interactions come from noisy realizations of hidden ground-truth vectors, the interaction matrix develops an asymptotic spectrum that either splits into two distinct bulks (informative and noisy) or remains a single merged bulk. The onset of the split marks the transition to a regime where the network can extract latent structure and generalize. When regularization is applied, networks in the split phase reconstruct the underlying vectors from corrupted samples, while the merged phase loses this capability.","pith_inferences":["The eigenvalue spectrum could serve as a diagnostic tool to monitor generalization during training without requiring ground-truth labels.","The same spectral criterion might apply to other unsupervised models whose weights arise from noisy high-dimensional data.","Finite-size scaling of the transition point could be measured numerically to guide practical network design."],"forward_implications":["The network extracts latent structure and generalizes beyond the training set precisely when the spectrum splits.","Regularization enables reconstruction of ground-truth vectors only in the presence of the spectral split.","The split or merged phase is determined by the quality and size of the accessible dataset.","In the merged phase the distinction between informative and noisy directions is lost and generalization fails."],"fun_headline_variants":["Spectral transition controls Hebbian generalization","Split bulks mark generalization in Hebbian nets","Merged bulk phase loses Hebbian generalization","Spectral split with regularization predicts reconstruction"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The spectrum derived in the thermodynamic limit via random matrix theory and replica methods accurately describes finite-size network behavior and generalization performance.","fun_headline_variants_meta":{"raw":{"variants":["Spectral transition controls Hebbian generalization","Split bulks mark generalization in Hebbian nets","Merged bulk phase loses Hebbian generalization","Spectral split with regularization predicts reconstruction"]},"model":"grok-4.3","cost_usd":0.008665,"raw_usage":{"total_tokens":3873,"prompt_tokens":599,"num_sources_used":0,"completion_tokens":54,"cost_in_usd_ticks":86649500,"prompt_tokens_details":{"text_tokens":599,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":3220,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":599,"tokens_out":54,"duration_ms":44345,"temperature":1.0,"reasoning_tokens":3220,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T03:32:31.712920+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"For fixed dataset size and noise level, compute the eigenvalues of a finite interaction matrix and check whether the observed reconstruction error drops precisely at the noise threshold where the predicted spectral split appears.","supporting_citations":[],"review_version":1}