{"id":"ca80ef8b-a06a-4151-8496-a21d040dabbe","arxiv_id":"2411.14635","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"A nonparametric relative entropy summary of intermittent time series, combined with BIC lag selection and CUSUM change-point detection, outperforms approximate entropy in locating complexity changes.","lead":"This paper develops a new statistic called relative entropy that summarizes each short time series segment into one number, then uses change-point detection on those numbers to find when complexity changes. In simulations and muscle fatigue data, it finds complexity change-points more reliably than approximate entropy, a common alternative.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Case 2 ApEn comparison is not head-to-head: ApEn is evaluated only on the 58.11% stationary subset while RlEn uses all 150 repetitions, and ApEn's r is unreported; the 'outperforms ApEn' claim is confounded.","rationale":"The abstract's central claim is that RlEn 'outperformed the ApEn' in change-point detection. That is a comparative empirical claim, so the comparison itself is the load-bearing element. In §4.2 the paper reports that only 58.11% of the logistic-transformed Case 2 series are stationary and then states that ApEn results are based entirely on the stationary series, while Table 3's RlEn results use all repetitions. This creates a matched-pairs violation: different sample sizes, different change-point locations, and no control for ApEn's r parameter. The enormous ApEn failure counts at several m values are exactly what one would expect if the CUSUM were run on a much shorter sequence whose change-point has moved. This is not an attack on the method's idea; it is a request to make the benchmark fair. The reader's CONDITIONAL verdict is reasonable, but the decisive condition should be the matched comparison, not only the stationarity assumption. A rerun on identical data with a grid of r values, or a statement that the repository already contains such a matched comparison, would resolve the concern. If the matched comparison reproduces the advantage, the paper's central claim stands; if not, the current Tables 3 and 6 overstate the result.","tokens_in":38295,"tokens_out":10044,"duration_ms":105605,"concrete_test":"Use the provided repository (github.com/Jieli12/RlEn) to rerun Case 2 with two matched protocols: (i) both RlEn and ApEn on the full 100 series per repetition; (ii) both methods on the same stationary-only subsequence, recomputing the true change-point in that subsequence. For ApEn, vary r over {0.05, 0.10, 0.15, 0.20, 0.25} × SD(segment) and also use the exact r from the original code if present; report MAD, failure count, and accuracy for each. If ApEn achieves failure rate below ~20% or accuracy comparable to RlEn under either matched protocol, the claimed outperformance is an artifact of unequal data and/or untuned r. If ApEn still fails at all reasonable r under both protocols, the concern is resolved.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In §4.2 (Case 2) the authors state that after logistic transformation only 58.11% of the time series are stationary and that 'ApEn results are totally based on the stationary time series,' while RlEn is applied to all 150 repetitions. This is not a matched comparison: ApEn's CUSUM input is roughly half the length of RlEn's, and the nominal change-point at 61 in the original ordering does not necessarily correspond to position 61 in the stationary subsequence. The large ApEn failure counts in Table 3 (e.g., 110/150 at m=2) could therefore be an artifact of reduced sample size and a shifted change-point rather than inferior complexity measurement. In addition, the ApEn tolerance parameter r is never reported; Pincus's 'suggestions' are not a unique setting, and ApEn is known to be highly sensitive to r. The same omission affects the real-data analysis in §5. Because the abstract's headline claim is an empirical outperformance claim, the current results do not establish it. This is a concrete, fixable flaw, not a theoretical quibble.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a nonparametric relative-entropy statistic (RlEn) as a scalar summary of each segment in a collection of intermittent time series, then applies a CUSUM-type change-point detector to the sequence of RlEn values. The lag order m is selected by BIC from a nonlinear autoregressive fit, and the entropy is estimated with jackknife kernel density estimators after a logistic transformation to [0,1]. The authors derive a Gaussian limit for the relative-entropy estimator under a fixed bandwidth and prove consistency of the BIC lag selection. They compare RlEn with approximate entropy (ApEn) in three simulation cases and in a real dataset of fatiguing muscle contractions, claiming that RlEn detects complexity change-points more accurately than ApEn.","tokens_in":38563,"tokens_out":5632,"duration_ms":58354,"significance":"If the comparison with ApEn were clean, the paper would fill a practical gap: ApEn is sensitive to two free parameters and is not transformation invariant, while RlEn has a nonparametric justification and comes with convergence-rate guarantees. The paper also ships reproducible code on GitHub, and the BIC and entropy components have independent foundations in Vieu (1995) and Hong and White (2005), so the method is not circularly constructed. However, the headline empirical claim is currently not established because the ApEn comparison in the key simulation is confounded and the ApEn tuning parameters are unreported; the theoretical results also do not cover the non-stationary operating conditions in which the method is applied.","major_comments":[{"comment":"The comparison in Case 2 is not matched: after logistic transformation only 58.11% of the series are stationary, and the paper states that ApEn results are based entirely on the stationary subset while RlEn is applied to all 150 repetitions. Because the change-point at 61 refers to the original ordering, it does not necessarily correspond to position 61 in the stationary subsequence, and the ApEn input vector is roughly half as long. The large ApEn failure counts (e.g., 110/150 for m=2) are therefore confounded with sample-size reduction and index misalignment. This undermines the abstract's headline claim that RlEn outperforms ApEn.","section":"§4.2, Table 3"},{"comment":"The ApEn tolerance parameter r is never reported: the description \"follow the suggestions in Pincus (1991)\" in §5 is not a unique specification, and r is known to strongly affect ApEn values. Without the r value, none of the ApEn comparisons in Tables 2, 3, 5, and 6 is reproducible or auditable. The authors should report r for every ApEn computation and ideally provide a sensitivity analysis over r.","section":"§4.1, §4.2, §5, and Tables 2, 3, 5, 6"},{"comment":"There is a gap between theory and implementation for the bandwidth: Theorem 3.1 gives a Gaussian limit for \\hat E(m,h) under rate conditions on a fixed h, but in §2 and Algorithm 2 h is chosen data-dependently by maximizing (or, as written in Algorithm 2, minimizing) \\hat E(\\hat m,h|X_j). No proof is given that the CLT survives data-driven h, and no sensitivity analysis over h is reported. Additionally, Algorithm 2's \"arg min_h\" contradicts the text's \"arg max_h\".","section":"§2 and Theorem 3.1"},{"comment":"The paper's invariance and noise-free selling points are broader than the results: Proposition B.1 establishes background-noise-freeness only for stationary Gaussian AR(p) processes, while the method is applied to general nonlinear and explicitly non-stationary series (41.89% of Case 2 series, and the real muscle data). Theorem 3.1 requires strict stationarity and a marginal density bounded away from zero on [0,1]; no result is provided for the non-stationary operating conditions, so the sentence \"our method is robust even for non-stationary time series as well\" (§4.2) is not supported by the theory.","section":"Appendix B and §3, in relation to §4.2 and §5"}],"minor_comments":[{"comment":"The objective should be arg max_h, not arg min_h, to match the text immediately before it.","section":"Algorithm 2, line 4"},{"comment":"The condition written \"nhm+13 → 0\" is ambiguous; it should be typeset as n h^{m+13} or the intended power should be stated explicitly.","section":"Theorem 3.1 and Theorems D.2–D.3"},{"comment":"The detected fraction for the mean method is 1/150, which is 0.67%, not 0.007% as printed.","section":"Table 2, row 'Mean'"},{"comment":"The sentence \"Subject 2's is just 21.84%\" appears to be a typo; Table 4 gives Subject 10 the 21.84% value.","section":"§5, paragraph after Table 5"},{"comment":"The leave-one-out estimator is denoted \\hat g^{-i} in the algorithm but is defined as \\hat F^{-i} in equation (4); the notation should be unified.","section":"Algorithm 1, line 10"}],"recommendation":"major_revision","confidential_remarks":"The reproducible code and the independent theoretical foundations of the BIC and entropy components are strengths. The main obstacle is the ApEn comparison: it is not head-to-head in Case 2 and the ApEn tuning parameters are not reported, so the central empirical claim cannot be accepted as stated. If the authors redo the comparison on identical series, report r for all ApEn runs, and either prove the bandwidth-selection result or add a sensitivity analysis, the paper would be suitable for publication. I do not see a load-bearing error that forces rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nYou should know two things about this paper. First, the core method is a genuine advance: the authors extend Hong and White's nonparametric relative entropy from pairwise variables to m-consecutive variables, derive a Gaussian limiting distribution, add a BIC-based lag selection with consistency, and wrap it in a CUSUM change-point procedure for intermittent time series. That is real work, and it is supported by a serious proof sketch and a tidy simulation study. Code is on GitHub, which helps. Second, the headline claim that RlEn outperforms ApEn is not actually supported by the evidence as presented, and the flaw is in the comparison design, not the theory.\n\nWhat the paper does well: the m-dimensional extension is not trivial—the proof has to handle overlapping blocks in the U-statistics, and they do that carefully. The BIC lag selection is standard but sensible, and the real-data application to muscle fatigue is a plausible use case. The transformation-invariance and noise-free properties are nicely motivated, and the AR(p) derivation in Appendix B is a useful concrete check.\n\nThe soft spots, in proportion: the Case 2 comparison is the biggest one. After logistic transformation only 58.11% of the series are stationary, and ApEn is evaluated only on that subset while RlEn runs on all 150 repetitions. That is not head-to-head. The stationary subset is about half the length, and the nominal change-point at 61 in the original ordering need not align with position 61 in the subsequence. The massive ApEn failure counts in Table 3 could simply be an artifact of reduced sample size and a shifted target. The ApEn tolerance r is never reported anywhere, which makes the comparison non-reproducible and raises the same concern for the real-data section. This is fixable—run both methods on the same stationary subset, report r, or match the inputs—but as it stands the empirical claim is confounded.\n\nTwo smaller issues: the asymptotic theory fixes the bandwidth while the implementation maximizes over it, which is a gap, though a common one. And the noise-free property is proven only for Gaussian AR(p), not claimed generally, yet the wording sometimes suggests more. The stationarity assumption is also violated in their own simulations, but they admit it and the robust behavior is at least plausible. I would not call these fatal; they are addressable in revision.\n\nWho should read this: anyone working on complexity measures for physiological time series or change-point detection for intermittent data. It deserves a serious referee, but the empirical section needs a redo before it can be trusted. My recommendation: engage with it, invite revision, and insist on a clean head-to-head comparison.","headline":"A real methodological extension of relative entropy to m-consecutive variables with a Gaussian limit, but the empirical claim that it beats ApEn is not yet established because the main comparison is confounded.","tokens_in":39042,"tokens_out":1766,"would_cite":true,"duration_ms":21186,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["62M10","62G07","62G20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that a nonparametric relative-entropy summary, with BIC-selected lag and CUSUM detection, locates complexity change-points in intermittent time series more accurately than approximate entropy.","keywords":["relative entropy","intermittent time series","change-point detection","approximate entropy","nonlinear autoregression","BIC lag selection","CUSUM","transformation invariant"],"falsifier":"Take two stationary AR(3) processes with a known change-point, generate intermittent series of length 500, then multiply the raw series by constants 0.1, 1, and 10 before the logistic transform and RlEn computation; if the detected change-point or the RlEn values shift by more than simulation error, the transformation-invariance claim fails.","tokens_in":38107,"feed_emoji":"📉","tokens_out":6020,"duration_ms":51516,"temperature":0.7,"pith_summary":"The paper develops RlEn, a nonparametric relative-entropy summary for each piece of an intermittent time series, and uses it to locate change-points in complexity across segments. It argues that RlEn has two properties the usual summaries lack: it is invariant to transformations of the signal and free of the background noise level. The lag order is selected by BIC, then a CUSUM statistic scans the sequence of RlEn values for breakpoints. Simulations and a fatigue study of muscle contractions indicate that RlEn locates complexity change-points more accurately than approximate entropy, especially when the lag order is misspecified.","feed_headline":"Relative entropy beats ApEn at spotting complexity shifts","feed_subtitle":"A BIC-chosen lag and a CUSUM scan convert each intermittent signal into one complexity number, with better accuracy than ApEn.","key_machinery":"The central object is the nonparametric relative entropy $\\hat E(m,h|X)$ for $m$ consecutive variables, a Kullback-Leibler divergence between the joint density of $m+1$ consecutive observations and the product of the $m$-history density and the marginal density. The paper estimates all three densities with leave-one-out Jackknife kernel estimators, selects the lag $m$ by BIC on a Nadaraya-Watson nonlinear autoregression, and then feeds the resulting scalar sequence into a CUSUM change-point detector. This scalar map is what turns each intermittent time series into one comparable number, so that breakpoints among segments become breakpoints in the mean of the RlEn sequence.","core_discovery":"The central claim is that mapping each intermittent time series to its nonparametric relative entropy $E = E_f[\\log(f(X_{i;m+1})/(g(X_{i;m})g_1(X_{i+m})))]$, with the lag $m$ chosen by a BIC criterion, produces a scalar sequence whose CUSUM change-points coincide with the times at which the underlying process loses complexity, and do so more accurately than approximate entropy. The estimator uses leave-one-out Jackknife kernel density estimates; under strict stationarity and a marginal density bounded away from zero on $[0,1]$, the paper proves a normal limiting distribution at rate $\\sqrt{n h^{(m+1)/2}}$ and consistency of the BIC lag selector. The method is designed to be transformation invariant and background noise-free, properties that mean, variance, entropy, and conditional entropy summaries do not have. In the muscle-fatigue application, the detected change-point separates contractions into two groups whose mean RlEn differs highly significantly.","pith_inferences":["If the invariance claim holds operationally, RlEn could serve as a unit-free feature for cross-subject or cross-device comparisons in clinical monitoring, where force, torque, or EEG scales differ across recordings.","The paper's own numbers indicate a potential gap: after the logistic transform only 58.11% of the AR(3) series are stationary, yet RlEn still performs well; a natural test is whether the asymptotic normality and BIC consistency survive under non-stationarity or whether those simulations are too mild to expose the gap.","One could extend the method to multivariate intermittent series by replacing the scalar RlEn with a vector of pairwise or block RlEn values, though the current CUSUM framework would not directly accept such a vector.","Because RlEn is computed per segment, the two-stage detector propagates any BIC lag-selection error into the entropy values; the paper's robustness results suggest the propagation is mild, but a formal sensitivity analysis is not provided."],"forward_implications":["Because RlEn is transformation invariant and background noise-free, change-point detections should not move when the measurement unit or noise variance of the intermittent signals changes.","With $m$ selected by BIC, RlEn keeps a small mean absolute distance from the true change-point even when $m$ is misspecified, whereas ApEn's accuracy collapses for larger $m$ in the AR(3) simulation.","The method detects a statistically reliable fatigue change-point in all 11 subjects of the muscle dataset, while ApEn fails to detect one for Subject 4 and gives an unreliable point for Subject 1.","The limiting Gaussian distribution of $\\hat E(m)$ justifies using the RlEn sequence as input to standard CUSUM-type change-point procedures."],"supporting_citations":[{"why":"Supplies the definition of relative entropy and the transformation-invariance property on which RlEn is built.","marker":"Kullback and Leibler (1951)"},{"why":"Defines approximate entropy, the baseline method against which RlEn is compared in simulations and the real data analysis.","marker":"Pincus (1991)"},{"why":"Provides the asymptotic distribution theory for nonparametric entropy measures that the paper extends from pairwise variables to m-consecutive variables.","marker":"Hong and White (2005)"},{"why":"Supplies the nonlinear autoregressive order-selection framework whose BIC consistency proof the paper adapts.","marker":"Vieu (1995)"},{"why":"Contributes the model-selection asymptotics used to prove consistency of the BIC lag selector.","marker":"Shao (1997)"},{"why":"Introduces the Jackknife kernel used to reduce boundary bias in the kernel density estimators for RlEn.","marker":"John (1984)"},{"why":"Provides the optimal CUSUM change-point detection method with linear computational cost used to find breakpoints in the RlEn sequence.","marker":"Killick et al. (2012)"}],"fun_headline_variants":["New relative entropy method outshines ApEn for complexity shifts","Complexity loss? This entropy-based CUSUM finds it better than ApEn","BIC lag + CUSUM on relative entropy beats ApEn in tests","Relative entropy pinpoints complexity loss in intermittent signals","Fatigue monitoring: relative entropy beats ApEn for change-points"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method's proof assumes each time series is strictly stationary and its distribution never touches zero on the unit interval, but the simulations and the real muscle data include non-stationary series, so the theoretical guarantee does not cover the conditions under which the method is actually used.","fun_headline_variants_meta":{"raw":{"variants":["New relative entropy method outshines ApEn for complexity shifts","Complexity loss? This entropy-based CUSUM finds it better than ApEn","BIC lag + CUSUM on relative entropy beats ApEn in tests","Relative entropy pinpoints complexity loss in intermittent signals","Fatigue monitoring: relative entropy beats ApEn for change-points"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000321,"raw_usage":{"total_tokens":1797,"prompt_tokens":926,"completion_tokens":871,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":542,"completion_tokens_details":{"reasoning_tokens":780}},"tokens_in":542,"tokens_out":871,"duration_ms":7615,"temperature":1.0,"reasoning_tokens":780,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T15:04:00.782128+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take two stationary AR(3) processes with a known change-point, generate intermittent series of length 500, then multiply the raw series by constants 0.1, 1, and 10 before the logistic transform and RlEn computation; if the detected change-point or the RlEn values shift by more than simulation error, the transformation-invariance claim fails.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Contributes the model-selection asymptotics used to prove consistency of the BIC lag selector."}],"review_version":1}