{"id":"95c5bb71-ac5b-4ead-92ab-8a61dda2ca81","arxiv_id":"2508.18338","paper_version":2,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":5.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"A linear-time method (HCR) estimates mutual information and tests independence using sums of squared mixed moments, claimed to match or beat the slower HSIC.","lead":"This paper presents a faster way to measure how strongly two datasets depend on each other, replacing a slow matrix-based method with a linear-time calculation based on statistical moments. If the claim holds, it makes independence testing practical on very large datasets.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Finite mixed-moment truncation cannot guarantee MI approximation or HSIC-level sensitivity; unverified from abstract alone.","rationale":"The reader's verdict was UNVERDICTED based on abstract-only evidence, with the weakest assumption being that a finite set of low-order mixed moments sufficiently characterizes the joint distribution. I agree with that identification. My stress-test sharpens the concern: the approximation of MI as a sum of squares of a finite moment set is not merely unproved; it is mathematically unsound as a universal approximation, because mutual information can be large even when all low-order mixed moments vanish. This means the central claim about 'even higher sensitivity to dependencies' is at risk for certain dependency structures. However, since the full text is unavailable, I cannot determine whether the authors restrict the claim to a particular class of dependencies, include a correction term, or provide empirical evidence covering such cases. Therefore the appropriate verdict remains UNVERDICTED rather than ACCEPT or REJECT. The concrete test proposed would help settle whether the sensitivity claim holds in a basic adversarial case.","tokens_in":681,"tokens_out":6612,"duration_ms":79822,"concrete_test":"Generate n=10^5 samples from X~N(0,1), Y=X^2−1 (centered so E[XY]=0), a strongly dependent pair with zero correlation. Run HCR with the moment set implied by the abstract (correlation and homoscedasticity terms) and compute its test statistic and permutation p-value; compare with HSIC and a plug-in MI estimate under the same n. If HCR's p-value is not significant while HSIC's is, the sensitivity claim fails. For a stronger version, construct a discrete distribution that has zero moments up to the exact HCR truncation order but nonzero MI, and check whether HCR's sum-of-squares approximation differs from the true MI.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract's central claim is that MI can be approximated as the sum of squares of a finite set of mixed moments (starting with correlation and homoscedasticity), and that HCR detects dependencies with sensitivity at least matching HSIC at O(n) cost per feature. The load-bearing assumption is that this truncated moment set captures the dependence structure relevant to MI. This is not guaranteed and is generally false: MI is a functional of the entire joint distribution, and there exist dependent distributions with vanishing low-order mixed moments—or even identical finite moment sequences up to any fixed order—while having large MI. Thus the HCR statistic can be near zero for strongly dependent data, directly contradicting the 'even higher sensitivity' claim. The abstract provides no truncation error bound, no convergence theorem, and no explicit class of alternatives for which the approximation is valid. Also, O(n) is per feature; testing pairwise dependencies costs O(d^2) features, so the overall 'linear cost' statement does not straightforwardly extend to high-dimensional settings. Without the full text, none of these points can be checked. The correct status is unverified.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes HCR (Hierarchical Correlation Reconstruction) as a linear-cost alternative to HSIC for dependence testing and mutual information estimation. The abstract claims that HCR computes a single dependence feature—a mixed moment such as correlation or homoscedasticity—in O(n) time, approximates mutual information as a sum of squares of such features, and achieves sensitivity at least matching HSIC in tests, while also providing a joint distribution model at a chosen significance level. The abstract also acknowledges that the number of features scales as O(d^2) for pairwise dependencies and O(d^3) for triplewise dependencies. No derivations, proofs, or experimental details are included in the available abstract-only text.","tokens_in":989,"tokens_out":2866,"duration_ms":35634,"significance":"If the claims are correct, the paper would offer a practically important speedup: replacing HSIC's O(n^2.37) matrix-based cost with O(n) per feature could enable dependence testing on very large samples. The moment-sum approximation is conceptually simple, and the abstract is honest about the O(d^2) scaling for pairwise feature screening. However, the central assertion that a finite set of low-order mixed moments accurately approximates mutual information is not demonstrated in the abstract. Since the available text contains no derivation, no convergence result, and no experimental evidence, the significance is conditional on a full manuscript that substantiates these claims.","major_comments":[{"comment":"The load-bearing claim is that mutual information can be approximated as 'just sum of squares of such nontrivial mixed moments.' This is not derived, and finite sets of low-order mixed moments do not in general determine mutual information. For example, distributions can share the same low-order moments up to any fixed order while having arbitrarily large mutual information, or can have vanishing low-order moments while still being strongly dependent. Without a convergence theorem, a truncation error bound, or an explicit class of distributions for which the approximation holds, the central claim is unsupported.","section":"Abstract, second sentence"},{"comment":"This sentence reports empirical superiority over HSIC, but the abstract gives no experimental protocol: no sample sizes, no baselines beyond HSIC, no definition of 'sensitivity,' and no quantitative comparison. As written, this is an unverifiable assertion. The full text must provide reproducible experiments and a precise sensitivity metric to support the claim of 'similar performance as HSIC' (or better).","section":"Abstract, phrase 'in tests of even higher sensitivity to dependencies'"},{"comment":"The abstract states that a single dependence feature is calculated in O(n) linear time, but also notes that the number of features is O(d^2) for pairwise dependencies. Thus the overall cost of a pairwise independence screen is O(n d^2), not O(n). If the paper's 'linear cost' claim refers to per-feature cost in n, this should be stated explicitly to avoid misleading readers about high-dimensional scaling.","section":"Abstract, O(n) complexity statement"},{"comment":"This claim is vague. It is unclear how a finite set of mixed moments determines a full joint distribution, or how a significance level is selected. The abstract does not explain whether HCR constructs a maximum-entropy distribution, a parametric model, or a nonparametric fit. This is another substantial claim that requires derivation and validation in the full text.","section":"Abstract, 'providing actual joint distribution model for chosen significance level'"}],"minor_comments":[{"comment":"The term 'mixed moments' is used without definition; in particular, 'homoscedasticity' is not a moment but a property of equal variances. Please clarify the moment-based definitions and the role of each feature.","section":"Abstract, terminology"},{"comment":"The word 'nontrivial' is ambiguous—does it mean nonzero, non-Gaussian, or independent of lower-order moments? A precise definition would help.","section":"Abstract, 'nontrivial'"},{"comment":"The title says 'similar performance as HSIC' while the abstract claims 'even higher sensitivity to dependencies.' These are different claims; they should be aligned or the discrepancy explained.","section":"Title and abstract"},{"comment":"Calling HSIC 'state-of-art' is subjective and may need a broader citation context. Also, the complexity O(n^2.37) is asymptotic; practical implementations often use approximations, so a brief qualification would be useful.","section":"Abstract, 'state-of-art method'"}],"recommendation":"uncertain","confidential_remarks":"This review is based on the abstract only, as no full text was available. The central claims are not verifiable from the abstract, and the moment-truncation approximation raises a fundamental correctness risk that cannot be resolved without the complete derivation. I recommend requesting the full manuscript before making a final decision. If the full paper does not include a convergence analysis or explicit distributional assumptions, the abstract's overclaim should be revised. I also note that 'HCR' may refer to prior work, so novelty disclosure should be checked."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"This is an abstract-only submission, so everything below is provisional. The core idea — approximate mutual information as a sum of squares of mixed moments and use it for independence testing at O(n) per feature — is concrete and, if it works, a genuinely useful complement to HSIC. The authors are also honest about the cost structure: O(d^2) for pairwise dependencies, O(d^3) for triplewise, which is more nuanced than the headline 'linear cost.' That transparency earns some credit.\n\nThe soft spots are mostly of the 'not shown' kind, since there is no math or data. The biggest is the moment truncation issue. A finite set of low-order mixed moments (starting with correlation and homoscedasticity) does not characterize dependence in general. There are distributions with zero correlation and identical low-order moments that have arbitrarily large mutual information, so the claimed 'even higher sensitivity' cannot hold universally. The abstract gives no truncation error bound, no convergence theorem, and no explicit class of distributions where the approximation is valid. It might work well in practice, but that is an empirical claim not supported by the abstract. Also, the comparison to HSIC is stated, not shown — no experiments, no synthetic data, no runtime measurements.\n\nThe stress-test note suggests the whole method may fail because MI depends on the full joint distribution. That is correct as a general statement, but it is not automatically fatal. Many successful independence tests use finite-dimensional summaries (e.g., distance correlation uses only pairwise distances). The paper needs to establish for which alternatives HCR has power and how the approximation behaves as the moment order grows. Without that, the paper is a research proposal, not a finished result.\n\nWho is this for? People working on large-scale independence testing and anyone using pairwise feature selection in high dimensions. It would make a decent workshop abstract. From the abstract alone, I would not cite it in my own work yet, but I would read the full paper if it appeared.\n\nRecommendation: if the full text contains a derivation of the moment-sum approximation, a bound or a clearly stated regime of validity, and solid experiments against HSIC, it deserves a serious referee. If this abstract is all there is, it's a desk-reject. I'd ask for the full text before making the call.","headline":"Promissory abstract: the HCR-based MI estimator is plausible and worth a full-text look, but the central truncation claim is unproven.","tokens_in":1338,"tokens_out":2043,"would_cite":false,"duration_ms":29874,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"HCR estimates mutual information at linear cost and matches HSIC sensitivity.","keywords":["mutual information","independence test","HSIC","hierarchical correlation reconstruction","mixed moments","linear complexity"],"falsifier":"Take two joint distributions that are constructed to be indistinguishable up to all moments included in some fixed HCR depth, but have different true mutual information (e.g., a distribution where dependence appears only in third-order interactions). If HCR with that depth gives the same dependence score for both while HSIC or a direct MI estimate distinguishes them, the approximation is shown to miss higher-order structure.","tokens_in":668,"feed_emoji":"📊","tokens_out":3769,"duration_ms":40374,"temperature":0.7,"pith_summary":"This paper tries to establish that Hierarchical Correlation Reconstruction (HCR) can replace HSIC for dependence testing at a fraction of the cost. Instead of multiplying large kernel matrices, HCR builds a description of the joint distribution from mixed moments—starting with correlation and homoscedasticity—and approximates mutual information as the sum of squares of those non-trivial moments. Each such feature costs O(n) time, so the full test scales linearly with sample size, whereas HSIC is superlinear. If correct, this makes sensitivity-equal or better dependence tests practical for very large datasets, while also returning a model of the joint distribution at a chosen significance level.","feed_headline":"HCR tests independence in linear time, rivaling HSIC","feed_subtitle":"Sum-of-squares mixed moments approximate mutual information cheaply, enabling large-scale dependence tests.","key_machinery":"The central machinery is the HCR expansion: a hierarchy of mixed moments (first correlation, then variance/homoscedasticity, then higher-order products) whose squared values sum to an approximation of mutual information. Each mixed moment is a normalized centred moment between the two variables; its computation is a single O(n) pass over the data. The hierarchy is what makes the linear cost possible and what connects low-order descriptive statistics to an information-theoretic quantity.","core_discovery":"The paper's central claim is that mutual information between two samples can be approximated by a sum of squares of normalized mixed moments—features that jointly encode correlation, homoscedasticity, and higher-order interactions. HCR (Hierarchical Correlation Reconstruction) computes these features in O(n) time each and aggregates them into a dependence measure that in tests shows equal or higher sensitivity than HSIC. The same features provide an explicit generative model of the joint distribution at a chosen significance level, which HSIC does not give.","pith_inferences":["The same moment-based MI approximation could be used as a training objective or regularizer in representation learning, since it is differentiable and linear-cost.","A natural stress test is to compare HCR against HSIC on distributions engineered to differ only in high-order moments; if the approximation misses those, its sensitivity advantage would vanish.","The hierarchical construction hints at a whole family of estimators: higher-order moment features could be pruned adaptively to the observed data, yielding an automatic complexity score."],"forward_implications":["Dependence testing on datasets with millions of points becomes feasible on ordinary hardware, since each feature costs O(n) rather than O(n^2.37).","The sum-of-squares approximation gives a direct way to estimate mutual information without density estimation or binning.","The moment hierarchy lets users choose a trade-off: O(d^2) features for pairwise dependencies, O(d^3) if triplewise interactions matter, and so on.","HCR yields a joint distribution model at the chosen significance level, so the same computation that tests independence also describes the dependence structure."],"supporting_citations":[],"fun_headline_variants":["Linear-time HCR rivals HSIC for dependence tests","HCR: sum-of-squares moments approximate mutual info","Independence test in O(n) matches HSIC sensitivity","HCR: faster independence tests with joint distribution"],"cache_read_input_tokens":2688,"weakest_assumption_plain":"The approximation assumes that low-order mixed moments—up to whatever depth the hierarchy is stopped—capture enough of the joint distribution that the sum of their squares stands in for the true mutual information; if real dependencies live in the omitted high-order moments, the estimate is biased and the test may miss them.","fun_headline_variants_meta":{"raw":{"variants":["Linear-time HCR rivals HSIC for dependence tests","HCR: sum-of-squares moments approximate mutual info","Independence test in O(n) matches HSIC sensitivity","HCR: faster independence tests with joint distribution"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000811,"raw_usage":{"total_tokens":3366,"prompt_tokens":688,"completion_tokens":2678,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":432,"completion_tokens_details":{"reasoning_tokens":2614}},"tokens_in":432,"tokens_out":2678,"duration_ms":24339,"temperature":1.0,"reasoning_tokens":2614,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-05T16:36:53.885018+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take two joint distributions that are constructed to be indistinguishable up to all moments included in some fixed HCR depth, but have different true mutual information (e.g., a distribution where dependence appears only in third-order interactions). If HCR with that depth gives the same dependence score for both while HSIC or a direct MI estimate distinguishes them, the approximation is shown to miss higher-order structure.","supporting_citations":[],"review_version":1}