{"id":"cbec486a-5992-448a-bdcd-5b96e0aaacd5","arxiv_id":"2607.12417","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":2,"one_line_summary":"PolarBM and LogPolarBM are complex-valued Boltzmann machines in polar and log-polar coordinates that couple phase to amplitude and report better audio spectral modeling than conventional models including DNNs.","lead":"The paper introduces PolarBM and LogPolarBM, Boltzmann machines that model complex data in polar (amplitude–phase) form so phase can depend on amplitude. If the claimed gains hold, they offer a more natural probabilistic model for spectra, radar, and other complex-valued signals than real-valued or independent-phase baselines.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.5","headline":"Abstract-only review leaves the central empirical claim (superior accuracy of PolarRBM/LogPolarRBM over DNNs via amplitude-phase coupling) uninspectable; no equations, experiments, or data are available to verify it.","rationale":"The Reader already correctly identified that the abstract asserts an empirical superiority driven by amplitude-phase coupling without supplying any inspectable evidence. Because the full text is unavailable, no deeper technical flaw (e.g., an invalid normalization of the complex density or a non-integrable energy) can be confirmed or refuted. The appropriate stress-test outcome is therefore to leave the verdict UNVERDICTED and to treat the missing experimental and mathematical detail as the single load-bearing concern. Agreement with the Reader is complete; no adjustment of the verdict is warranted until the full paper can be examined.","tokens_in":2197,"tokens_out":471,"duration_ms":4171,"concrete_test":"Obtain the full PDF (or arXiv source) and re-evaluate the experimental section: extract the precise energy functions of PolarRBM/LogPolarRBM, the PW-NCCG derivation, the audio datasets, model sizes, and the quantitative metrics versus the DNN baselines. If the superiority disappears under matched capacity/training or if the PW-NCCG special-case reductions do not hold algebraically, the central claim weakens; otherwise the UNVERDICTED status can be revised.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The strongest claim is that PolarRBM and LogPolarRBM achieve superior modeling accuracy on audio by explicitly coupling phase to amplitude (via PolarBM/LogPolarBM densities and the PW-NCCG conditional that recovers Rice/Nakagami/noncentral-chi margins). That claim is load-bearing for the paper's significance, yet the available text is only the abstract: no energy functions, no sampling/inference procedure, no dataset description, no architecture sizes or training protocols, no quantitative tables, and no comparison methodology against the DNN baselines. Consequently it is impossible to check whether the reported superiority is produced by the claimed inductive bias rather than by unmatched capacity, optimization differences, or evaluation choices. The reader's weakest_assumption correctly flags this gap; with only the abstract there is no further internal inconsistency to probe.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The manuscript proposes PolarBM, a Boltzmann machine for complex-valued variables in polar (amplitude–phase) coordinates, so that phase is explicitly conditioned on amplitude, together with LogPolarBM, which places amplitude on a logarithmic scale. LogPolarBM is claimed to induce a power-weighted noncentral complex Gaussian (PW-NCCG) conditional whose amplitude margins recover the Rice, Nakagami, and noncentral-chi distributions as special cases. Restricted variants PolarRBM and LogPolarRBM are introduced for practical use. The abstract asserts that, by modeling amplitude–phase dependence, these RBMs attain superior modeling accuracy on audio signals relative to conventional models, including deep neural networks, and suggests broader applicability to other complex-valued domains.","tokens_in":2365,"tokens_out":823,"duration_ms":16358,"significance":"If the distributional constructions and the empirical superiority claims hold under full scrutiny, the work would supply a useful inductive bias for energy-based modeling of complex signals (audio, wireless, quantum) and a flexible conditional family (PW-NCCG) that unifies several classical amplitude laws. Explicit amplitude–phase coupling and the restricted practical variants are potentially valuable contributions to complex-valued machine learning. Those strengths cannot yet be credited as established, because the material under review is only the abstract: no energy functions, sampling/inference procedures, datasets, metrics, architecture sizes, or quantitative comparisons are available to inspect.","major_comments":[{"comment":"The load-bearing empirical claim—that PolarRBM and LogPolarRBM achieve superior modeling accuracy on audio versus conventional models including DNNs by virtue of explicit amplitude–phase coupling—is asserted without any quantitative results, baseline definitions, metrics, error bars, datasets, capacity controls, or training/evaluation protocols in the material provided. Without those elements the central claim cannot be assessed, and it is impossible to separate the proposed inductive bias from unmatched capacity or optimization differences.","section":"Abstract (experimental claims)"},{"comment":"The claim that LogPolarBM yields a PW-NCCG conditional whose amplitude margins include Rice, Nakagami, and noncentral chi as special cases is distributionally load-bearing for the paper’s theoretical contribution, yet no energy function, conditional derivation, or parameter-reduction argument appears in the available text. The special-case recoveries therefore remain unverified.","section":"Abstract (PW-NCCG / distributional claim)"},{"comment":"The abstract motivates superiority by the amplitude–phase dependence built into PolarBM/LogPolarBM, but supplies no independent justification or ablation that this structure—not architecture size, training protocol, or evaluation setup—is what drives any reported gains. That premise is the weakest load-bearing assumption of the work and cannot be checked from the abstract alone.","section":"Abstract (inductive-bias motivation)"}],"minor_comments":[{"comment":"The abstract does not name the audio corpora, evaluation metrics (e.g., log-likelihood, reconstruction or spectral measures), or the specific DNN/baseline architectures, which would help readers gauge the intended empirical scope.","section":"Abstract"},{"comment":"Notation for the free parameters of PolarBM/LogPolarBM and of the PW-NCCG (weights, biases, power/scale) is not introduced; a brief parameter list would clarify what is learned versus fixed by construction.","section":"Abstract"}],"recommendation":"uncertain","confidential_remarks":"Only the abstract was supplied for review; the full manuscript (equations, algorithms, experiments) was not available. A definitive soundness or novelty judgment relative to prior complex-valued RBMs and energy-based models is therefore impossible. I recommend obtaining the complete paper before any accept/reject decision; with the present material the only responsible recommendation is uncertain."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"Punchline: this is a methods paper that puts Boltzmann machines in polar and log-polar coordinates so phase is explicitly conditioned on amplitude, plus a restricted form and a PW-NCCG conditional whose amplitude margins include Rice, Nakagami, and noncentral chi. That distributional unification is the cleanest part of the abstract. The load-bearing empirical claim—that PolarRBM/LogPolarRBM beat conventional models including DNNs on audio by virtue of that coupling—is asserted, not shown here.\n\nWhat is actually new, on the text we have: formulating the energy so the complex density lives in polar (and log-polar) coordinates with amplitude→phase dependence, naming the restricted variants for practice, and deriving a flexible conditional that recovers standard amplitude laws as special cases. That is a coherent inductive bias for spectra and other complex signals where amplitude and phase are not independent. Credit where due: the abstract is clear about the physical motivation and does not hide that the models are energy-based with free weights/biases. Self-contained special-case recovery of classical distributions is useful if the derivations hold in the full paper.\n\nSoft spots, in proportion: we only have the abstract. No energy functions, sampling/inference details, datasets, architecture sizes, training protocols, metrics, tables, or ablations. So we cannot tell whether the reported accuracy edge comes from the amplitude–phase coupling or from unmatched capacity, optimization, or evaluation choices. The weakest assumption the reader flagged is real: the abstract motivates the bias but does not independently justify that it, rather than setup, drives the gains. That is a gap in the available text, not proof the experiments are wrong. Circularity burden looks low; free BM parameters are expected.\n\nWho this is for: people working on complex-valued probabilistic models, audio spectral modeling, and maybe wireless or other polar-signal domains who care about energy-based models with interpretable conditionals. A serious referee should see the full math and experiments. I would not cite from the abstract alone, and I would not put it in next week’s reading group without the paper body, but it is not desk-reject material if the derivations and comparisons are actually there. Send it to peer review; the method idea is sharp enough to deserve a careful look, with the empirical superiority treated as a claim to verify rather than a settled fact.","headline":"Polar/log-polar BMs with amplitude-conditioned phase and a PW-NCCG that recovers classical amplitude laws look like a real methods contribution; the DNN-beating accuracy claim is uncheckable from the abstract alone.","tokens_in":3065,"tokens_out":589,"would_cite":false,"duration_ms":9540,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A Boltzmann machine in polar coordinates models amplitude and phase together so that phase depends on amplitude, improving how complex audio spectra are learned.","keywords":["complex-valued Boltzmann machine","polar coordinates","amplitude-phase modeling","LogPolarBM","restricted Boltzmann machine","audio spectral modeling","PW-NCCG","energy-based models"],"falsifier":"Train PolarRBM or LogPolarRBM and matched-capacity real-valued or complex baselines on the same audio spectral datasets with identical protocols; if the polar models do not show lower negative log-likelihood or better reconstruction metrics, the central superiority claim fails.","tokens_in":3006,"feed_emoji":"🔊","tokens_out":780,"duration_ms":7820,"temperature":0.7,"pith_summary":"Most machine learning treats complex data such as audio spectra by splitting them into independent real and imaginary parts, or into amplitude and phase that are learned separately. That discards the physical link between how loud a component is and how its phase behaves. This paper introduces PolarBM, an energy-based model that works directly in polar coordinates so the probability of a complex value makes phase depend on amplitude. For audio, LogPolarBM further places amplitude on a logarithmic scale, matching human hearing and producing a flexible conditional density called the power-weighted noncentral complex Gaussian (PW-NCCG). The restricted versions, PolarRBM and LogPolarRBM, are practical machines that can be trained on real data. The authors argue that this inductive bias alone yields higher modeling accuracy on audio signals than conventional restricted Boltzmann machines and even deep neural networks, and that the same construction applies wherever complex measurements appear.","feed_headline":"Polar Boltzmann machine ties phase to amplitude for better audio models","feed_subtitle":"Restricted polar and log-polar RBMs beat conventional and deep baselines by keeping the physical link.","key_machinery":"PolarBM and LogPolarBM: energy-based densities over complex variables in polar or log-polar form that make phase depend on amplitude; their restricted variants (PolarRBM, LogPolarRBM) and the PW-NCCG conditional whose amplitude margins recover Rice, Nakagami, and noncentral-chi laws as special cases.","core_discovery":"By defining a complex-valued Boltzmann machine whose energy is written in polar (or log-polar) coordinates, the model induces a joint density in which phase is conditioned on amplitude; the resulting restricted machines capture physically meaningful amplitude-phase coupling and outperform conventional real-valued and deep baselines on audio spectral modeling.","pith_inferences":["If the amplitude-conditioned phase prior is the real source of the gains, similar polar reparameterizations should improve other energy-based or generative models beyond Boltzmann machines.","Log-polar amplitude modeling may give a principled alternative to ad-hoc log-magnitude preprocessing in modern complex neural audio architectures.","Failure of the polar models on signals whose phase is known to be amplitude-independent would cleanly isolate when the inductive bias helps versus hurts."],"forward_implications":["Restricted PolarRBM and LogPolarRBM become practical density models for complex audio spectra that keep amplitude-phase coupling.","The PW-NCCG conditional supplies a single parametric family that specializes to Rice, Nakagami, and noncentral-chi amplitude laws used in acoustics and communications.","The same polar construction can be applied without change to other complex-valued domains such as wireless channels or quantum measurements.","Because phase is explicitly amplitude-dependent, downstream tasks that need consistent phase (source separation, synthesis) inherit that structure for free."],"fun_headline_variants":["PolarBM couples phase to amplitude for stronger audio spectral models","Complex Boltzmann machine models phase conditioned on amplitude","LogPolar RBM ties log-amplitude to phase and beats deep baselines","Restricted polar BMs capture physical amplitude-phase links in audio","Polar and log-polar RBMs outperform conventional models on spectra"],"cache_read_input_tokens":128,"weakest_assumption_plain":"The claim that the physically important structure of complex audio is captured mainly by conditioning phase on amplitude inside an energy-based polar model, and that this bias—not model size or training details—explains the reported gains over deep networks.","fun_headline_variants_meta":{"raw":{"variants":["PolarBM couples phase to amplitude for stronger audio spectral models","Complex Boltzmann machine models phase conditioned on amplitude","LogPolar RBM ties log-amplitude to phase and beats deep baselines","Restricted polar BMs capture physical amplitude-phase links in audio","Polar and log-polar RBMs outperform conventional models on spectra"]},"model":"grok-4.5","effort":"low","cost_usd":0.004188,"raw_usage":{"total_tokens":1278,"prompt_tokens":822,"num_sources_used":0,"completion_tokens":85,"cost_in_usd_ticks":41880000,"prompt_tokens_details":{"text_tokens":822,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":371,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":822,"tokens_out":85,"duration_ms":3611,"temperature":1.0,"reasoning_tokens":371,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-15T06:15:02.551445+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Train PolarRBM or LogPolarRBM and matched-capacity real-valued or complex baselines on the same audio spectral datasets with identical protocols; if the polar models do not show lower negative log-likelihood or better reconstruction metrics, the central superiority claim fails.","supporting_citations":[],"review_version":1}