{"id":"5f41a26c-3cf0-4579-ba38-aa72284d2810","arxiv_id":"2504.18444","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A bucket-and-boost estimator using least squares and a geometric median recovers the first T Markov parameters of an LTI system under heavy-tailed noise with O(sqrt(p log(1/delta)/N)) error and O((mT)^2 kappa log(1/delta)) samples.","lead":"This paper designs a system identification method for partially observed linear systems under heavy-tailed noise, where the noise only needs a finite second moment. It shows the estimator achieves error bounds with logarithmic dependence on the failure probability, nearly matching sub-Gaussian benchmarks.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the proof of Theorem 1 is coherent and the stated assumptions support the logarithmic-in-delta guarantee.","rationale":"The reader identified the finite fourth moment on the input as the weakest assumption. I agree it is the most fragile part of the proof, but it is not a load-bearing objection to correctness: the theorem explicitly assumes it, and the paper transparently notes the necessity question remains open. All other steps of the argument, including the block-wise Markov bounds for heavy-tailed noise, the invertibility of the regressor Gram matrix, and the geometric-median boosting argument, are technically sound. I checked the derivation of Lemmas 1–4 and found no internal inconsistency or missing condition. The central claim, that heavy-tailed noise with only finite second moment does not destroy the logarithmic dependence on the failure probability, is well supported. Therefore, the reader's ACCEPT verdict should remain unchanged.","tokens_in":13612,"tokens_out":29555,"duration_ms":272289,"concrete_test":"Re-derive Lemma 1 for an input distribution with finite second moment but infinite fourth moment (e.g., a Student-t distribution with 4 degrees of freedom), and check whether the sample-covariance lower bound lambda_min >= M sigma_u^2/2 can hold with probability 1-q/3 when M grows only polynomially in 1/q. If it fails, the fourth-moment assumption is necessary for the stated logarithmic dependence, confirming the theorem's scope.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I could not identify a gap that would invalidate the central claim. The proof structure is sound: Lemmas 1–3 provide per-bucket OLS error bounds via Markov's inequality, and Lemma 4 correctly shows that the geometric median reduces the failure probability from a constant to exp(-K/32) using Hoeffding. The only assumption that stands out as potentially fragile is the finite fourth moment on the input (Eq. 4), used in Lemma 1 to control the sample covariance of the regressors. The paper explicitly flags this and leaves the necessity open (Remark 2), so it is a scope limitation rather than an error. The noise processes genuinely require only finite second moments.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies non-asymptotic system identification of partially observed linear time-invariant systems from multiple independent rollouts when the process and measurement noises are heavy-tailed and have only finite second moments. The authors propose a bucketed ordinary-least-squares estimator whose outputs are combined via the geometric median with respect to the Frobenius norm, and prove Theorem 1, which states that the Markov parameter estimator satisfies a bound of the form ||Ghat-G|| <= ((sigma_v C1 + sigma_w C2)/sigma_u) sqrt(p log(1/delta)/N) with probability at least 1-delta, provided the number of buckets K is ceil(32 log(1/delta)), each bucket has M = O((mT)^2 sigma_tilde_u^4/sigma_u^4) rollouts, and the input is zero-mean i.i.d. with finite fourth moment. The proof is built on Lemmas 1-3, which use Markov's inequality to control the sample covariance, the process-noise term, and the measurement-noise term per bucket, and Lemma 4, which uses a geometric-median boosting argument and Hoeffding's inequality to convert constant per-bucket failure probabilities into a logarithmic dependence on delta.","tokens_in":13709,"tokens_out":13784,"duration_ms":129271,"significance":"If the main result holds, it is a meaningful advance: it shows that logarithmic dependence on the confidence parameter is achievable for partially observed LTI system identification under noise distributions that are only assumed to have finite second moments, nearly matching the rates previously available under Gaussian or sub-Gaussian assumptions. The algorithm is simple, the proof strategy is largely self-contained, and the paper is unusually transparent about the cost of removing sub-Gaussian assumptions, including the extra dimension factors and the role of the fourth-moment assumption on the input. The explicit discussion of the input kurtosis and the open question about whether the fourth moment is necessary is a strength. However, as detailed below, the proof of the boosting lemma contains a threshold error that is load-bearing for the stated logarithmic-in-delta rate; the central claim appears defensible after a localized correction, but the manuscript in its current form does not establish Theorem 1 as written.","major_comments":[{"comment":"The proof of Lemma 4 defines epsilon = (sqrt(8)/M)(1/sigma_u)(sigma_v C1 + sigma_w C2), but this is not compatible with the per-bucket guarantee (13). Setting q = 1/8 in (13) gives ||Ghat_j - G|| <= 8(sigma_v C1 + sigma_w C2)/(sigma_u sqrt(M)) with probability at least 7/8, so the per-bucket failure probability can only be bounded by 1/8 for thresholds of order 1/sqrt(M). The chosen epsilon is of order 1/M, so the claim E[Z1] <= 1/8 used in the Hoeffding step is not established. Consequently, the final logarithmic-in-delta bound in (8) and (14) does not follow from the written proof, and the final substitution M = N/K cannot produce the stated sqrt(p log(1/delta)/N) rate from an epsilon of order 1/M. This is a load-bearing issue, although it appears repairable by choosing epsilon proportional to (sigma_v C1 + sigma_w C2)/(sigma_u sqrt(M)) and tracking the resulting constants.","section":"Section 4, Lemma 4"},{"comment":"In the proof of Lemma 1, the displayed equality E[sum_{i != l} (uhat(i))^2 (uhat(l))^2] = ((mT)^2 - mT) sigma_u^4 assumes independence of the coordinates of each input vector, whereas Assumption (4) only specifies the covariance and the marginal fourth moments. The same final bound can be obtained by applying Cauchy-Schwarz to each cross term, so the lemma remains valid after this correction, but the proof as written relies on a stronger assumption than stated.","section":"Section 4, Lemma 1 proof"}],"minor_comments":[{"comment":"There are several typographical errors, for example 'we device a novel algorithm' in Section 1 should be 'we devise a novel algorithm'; these should be corrected in revision.","section":"Throughout"},{"comment":"The proof of Lemma 4 mixes strict and non-strict inequalities when defining the events S_j and Z_j and when applying Lemma 6; the argument can be made fully rigorous by introducing a small slack in the thresholds and then taking the slack to zero.","section":"Section 4, Lemma 4"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this paper is a good theory contribution. It answers the open question from the Tsiamis et al. survey about whether logarithmic dependence on δ can survive when noise has only a finite second moment, and it does so for partially observed systems with inputs and measurement noise. The proof is a median-of-means argument on per-bucket OLS estimators; the three Markov-type lemmas and the boosting lemma check out. I read the appendix and found no circularity and no fitted quantities.\n\nWhat is new: earlier heavy-tailed results either need all moments (sub-Weibull) or a fully observed setting. This paper transfers geometric median boosting from robust mean estimation to system identification, and the finite-fourth-moment input assumption is a reasonable price for the generality. The paper is also honest about its own limitations: it flags the extra T·max(n,m,p) factors relative to the sub-Gaussian case and leaves open whether the fourth moment on the input is necessary.\n\nSoft spots, in proportion. The noise and input are assumed i.i.d. across time, which is a real restriction compared with the single-trajectory or correlated-noise literature. The theorem is stated as an error bound ~ sqrt(p log(1/δ)/N), but because N=MK and K~log(1/δ), the log cancels and the bound is essentially sqrt(p/M); that is standard median-of-means behavior (block size controls accuracy, number of blocks controls confidence), but the presentation could be clearer. No simulations; fine for a result of this type. Constants are not optimized and the T-dependencies are ugly, but they are explicit.\n\nI agree with the reader's verdict: the argument is coherent, the related-work discussion is accurate, and the stress-test note found no gap. This deserves a serious referee. I would expect acceptance after minor revisions, mainly to clarify the N=MK substitution and the role of M.","headline":"Solid, honest theory paper: first finite-sample guarantee for partially observed LTI identification under second-moment noise, with a clean median-of-means proof and no load-bearing flaw.","tokens_in":14206,"tokens_out":12949,"would_cite":true,"duration_ms":125596,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["93B30","62F35","93E12"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that a partially observed linear system's input-response matrices can be estimated with high confidence under noise that has only a finite variance, without any Gaussian tail assumption.","keywords":["system identification","Markov parameters","heavy-tailed noise","geometric median","boosting","finite-sample guarantees","partially observed LTI systems","robust estimation"],"falsifier":"Run the algorithm with a zero-mean input that has finite covariance but infinite fourth moment, such as a multivariate $t$ distribution with four degrees of freedom, choosing $K=\\lceil32\\log(1/\\delta)\\rceil$ and any finite bucket size $M$. The theorem's proof specifically requires $M\\gtrsim(mT)^2(\\tilde\\sigma_u^4/\\sigma_u^4)$, which diverges when $\\tilde\\sigma_u^4$ is infinite; if the empirical failure frequency still stays below $\\delta$ with an error matching the bound, the fourth-moment assumption is unnecessary, while an ill-conditioned $U_jU_j^\\top$ or a degraded rate would confirm it is load-bearing.","tokens_in":13419,"feed_emoji":"⚙️","tokens_out":15563,"duration_ms":139717,"temperature":0.7,"pith_summary":"This paper asks whether the parameters of a partially observed linear time-invariant system can still be identified with high confidence when both the process and measurement noise have only a finite second moment—no Gaussian or sub-Gaussian tail assumption. It answers yes, using an estimator that splits $N$ independent input-output rollouts into $K$ buckets, fits a least-squares model to each bucket, and combines the $K$ weak fits with a geometric median taken in the Frobenius norm. The main theorem gives, with probability at least $1-\\delta$, an error bound of order $\\sqrt{p\\log(1/\\delta)/N}$ for the first $T$ Markov parameters, recovering the logarithmic dependence on the failure probability that previous light-tailed analyses achieved. The price is a mild increase in the required number of rollouts and an extra dimensional factor, with the finite fourth moment required only on the input process, not on the noise.","feed_headline":"System ID keeps logarithmic confidence under two-moment noise","feed_subtitle":"A bucket-and-geometric-median estimator nearly matches sub-Gaussian guarantees without light tails.","key_machinery":"The load-bearing mechanism is boosting by geometric median in the matrix space $\\mathbb{R}^{p\\times mT}$ equipped with the Frobenius norm (the entrywise Euclidean norm): $\\hat G=\\arg\\min_{\\theta}\\sum_{j=1}^K\\|\\theta-\\hat G_j\\|_F$. Its stability lemma (Lemma 6 in the paper) states that if the median is far from the truth, then a positive fraction of the bucket estimators must themselves be far from the truth; with only $1/8$ of buckets failing, $K=\\lceil32\\log(1/\\delta)\\rceil$ buckets push the overall failure probability below $\\delta$. Because the noise is only heavy-tailed, the per-bucket analysis replaces Chernoff-type bounds with Markov's inequality on squared Frobenius norms, using the variance additivity identity for independent matrix summands. The finite fourth moment of the input enters precisely where $\\mathbb{E}[\\|\\hat u\\|^4]\\le(mT)^2\\tilde\\sigma_u^4$ is used to prove that the per-bucket Gram matrix $U_jU_j^\\top$ is invertible with the stated sample size.","core_discovery":"The paper's central claim is that, under zero-mean i.i.d. noise with covariance matrices $\\sigma_w^2 I_n$ and $\\sigma_v^2 I_p$, and a zero-mean i.i.d. input with covariance $\\sigma_u^2 I_m$ and finite fourth moment $\\tilde\\sigma_u^4$, the estimator described above achieves, with probability at least $1-\\delta$, $$\\|\\hat G-G\\|\\le \\frac{\\sigma_v C_1+\\sigma_w C_2}{\\sigma_u}\\sqrt{\\frac{p\\log(1/\\delta)}{N}},$$ where $C_1=c_1T^{1.5}\\sqrt{pm}$, $C_2=c_2\\|F\\|T^{2.5}\\sqrt{nm}$, $K=\\lceil32\\log(1/\\delta)\\rceil$, $M\\ge c_4(mT)^2(\\tilde\\sigma_u^4/\\sigma_u^4)$, and $N=MK$. Here $G=[D\\;CB\\;CAB\\;\\cdots\\;CA^{T-2}B]$ is the block of the first $T$ Markov parameters, and $\\hat G$ is the geometric median of the per-bucket least-squares estimates. The bound matches the multi-trajectory sub-Gaussian rate up to a factor of order $T\\max(n,m,p)$, and the Ho-Kalman realization step then transfers the logarithmic confidence to estimates of the original system matrices $(A,B,C,D)$ up to similarity transformation.","pith_inferences":["A direct stress test suggested by the proof is to replace the input distribution by one with finite variance and infinite fourth moment, such as a multivariate $t$ distribution with four degrees of freedom; the theorem's mechanism predicts that the invertibility lemma breaks, so either the error bound degrades or the geometric median still rescues it, and either outcome would settle the paper's op","The same bucket-and-median architecture could be adapted to the single-trajectory setting by chopping one long rollout into weakly dependent segments, but the temporal correlations would need an explicit decorrelation argument that the paper does not provide.","The kurtosis ratio $\\tilde\\sigma_u^4/\\sigma_u^4$ entering $M$ is an estimable quantity, so a practitioner could estimate it from input data before collecting outputs and choose the number of rollouts accordingly, making the sample-size prescription data-driven.","Because the geometric-median lemma tolerates a minority of bad buckets, the estimator is plausibly robust not only to heavy tails but also to a fraction of entirely corrupted rollouts, although the paper does not analyze adversarial corruption."],"forward_implications":["Via the Ho-Kalman realization step, the $\\log(1/\\delta)$ dependence transfers from the Markov parameters to estimates of $(A,B,C,D)$ up to similarity, so downstream model-based control can inherit high-probability model-error bounds under heavy-tailed noise.","The per-bucket sample requirement $M\\gtrsim(mT)^2(\\tilde\\sigma_u^4/\\sigma_u^4)$ depends only on the input distribution, and replacing the heavy-tailed input by a Gaussian one removes the kurtosis factor and reduces $M$ to order $mT$.","The error bound has the same $\\sqrt{\\log(1/\\delta)/N}$ scaling as the sub-Gaussian multi-trajectory baseline, so the cost of dropping light-tailed assumptions is only polynomial in the system dimensions, not exponential in $\\log(1/\\delta)$.","Because each rollout starts from zero and rollouts are independent, the result holds without mixing or stability assumptions on $A$; the only system-dependent constant entering through the process-noise term is $\\|F\\|$."],"supporting_citations":[{"why":"Supplies the multi-trajectory least-squares formulation for Markov parameters and the sub-Gaussian benchmark whose logarithmic failure dependence this paper recovers under heavy tails.","marker":"[6]"},{"why":"Provides the geometric-median robustness lemma used as Lemma 6, the property that makes the boosting step convert weak per-bucket estimates into exponentially small failure probability.","marker":"[9]"},{"why":"Provides the Ho-Kalman robustness analysis that lets the paper pass from the Markov-parameter bound to estimates of the system matrices up to similarity transformation.","marker":"[8]"},{"why":"Earlier fully observed heavy-tailed identification that introduced the kurtosis-dependent sample complexity and the variance-additivity lemma reused in the per-bucket bounds.","marker":"[25]"},{"why":"Gaussian multi-trajectory sample-complexity analysis that defines the light-tailed baseline and the downstream model-based control motivation.","marker":"[13]"}],"fun_headline_variants":["Heavy-tailed noise? System ID still gets log confidence","Boosting yields near-optimal rates with only two moments","System ID under two-moment noise: log confidence kept","Robust ID matches sub-Gaussian bounds without light tails","Two-moment noise: sub-Gaussian rates via boosting"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the excitation input has a finite fourth moment; the proof's guarantee that the least-squares fit in each bucket is well posed uses an average of the fourth power of the input, and without that moment the stated sample-size condition has no force.","fun_headline_variants_meta":{"raw":{"variants":["Heavy-tailed noise? System ID still gets log confidence","Boosting yields near-optimal rates with only two moments","System ID under two-moment noise: log confidence kept","Robust ID matches sub-Gaussian bounds without light tails","Two-moment noise: sub-Gaussian rates via boosting"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000714,"raw_usage":{"total_tokens":3233,"prompt_tokens":993,"completion_tokens":2240,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":609,"completion_tokens_details":{"reasoning_tokens":2159}},"tokens_in":609,"tokens_out":2240,"duration_ms":17728,"temperature":1.0,"reasoning_tokens":2159,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-16T10:17:33.270850+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run the algorithm with a zero-mean input that has finite covariance but infinite fourth moment, such as a multivariate $t$ distribution with four degrees of freedom, choosing $K=\\lceil32\\log(1/\\delta)\\rceil$ and any finite bucket size $M$. The theorem's proof specifically requires $M\\gtrsim(mT)^2(\\tilde\\sigma_u^4/\\sigma_u^4)$, which diverges when $\\tilde\\sigma_u^4$ is infinite; if the empirical failure frequency still stays below $\\delta$ with an error matching the bound, the fourth-moment assumption is unnecessary, while an ill-conditioned $U_jU_j^\\top$ or a degraded rate would confirm it is load-bearing.","supporting_citations":[{"cited_title":"Geometric median and robust estimat ion in banach spaces","cited_arxiv_id":null,"evidence_quote":"Provides the geometric-median robustness lemma used as Lemma 6, the property that makes the boosting step convert weak per-bucket estimates into exponentially small failure probability."},{"cited_title":"Non-asymptotic identifica tion of lti systems from a single trajectory","cited_arxiv_id":null,"evidence_quote":"Provides the Ho-Kalman robustness analysis that lets the paper pass from the Markov-parameter bound to estimates of the system matrices up to similarity transformation."},{"cited_title":"On the sam- ple complexity of the linear quadratic regulator","cited_arxiv_id":null,"evidence_quote":"Gaussian multi-trajectory sample-complexity analysis that defines the light-tailed baseline and the downstream model-based control motivation."}],"review_version":1}