{"id":"bea3de35-aa24-409c-a896-05a3b413ed7f","arxiv_id":"1909.05148","paper_version":1,"verdict":"REJECT","confidence":"HIGH","novelty_score":2.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"A survey of machine learning for optical communication that classifies many references by algorithm type, but contains factual inaccuracies and an unsupported first-time claim.","lead":"This paper surveys machine learning applications in optical communication, grouping reviewed works by supervised, unsupervised, and reinforcement learning algorithms. The survey's claim to be the first ML-view review is contradicted by prior surveys cited within the paper itself.","discovery_kind":"review","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The 'first ML-view survey' claim is contradicted by the paper's own reference [128], a prior ML-techniques tutorial, so the central novelty claim fails.","rationale":"The reader's verdict is REJECT with high confidence, and my stress-test agrees. The central claim is empirical novelty: no prior ML-for-OC survey took an ML viewpoint. The load-bearing assumption is that [128] and similar tutorials did not. That assumption is false on the face of the paper's own reference list: [128] is a 2016 JLT paper titled 'Machine learning techniques in optical communication,' and the paper cites it in §III.B. The paper never distinguishes its viewpoint from [128]'s and omits [128] from the Related Works list, where the sentence 'none of them had ML view' appears. This is not a matter of taste or classification; it is a falsifiable bibliographic claim that fails. The secondary claim of covering 'much more investigations' is similarly unsupported by any disclosed search method or comparison. Since the introduction and abstract use these claims to justify the survey, the central contribution is not reliable. I would not change the reader's verdict.","tokens_in":28151,"tokens_out":4771,"duration_ms":47384,"concrete_test":"Retrieve the full text of [128] (Zibar, Piels, Jones, Schäffer, JLT 2016) and [156] (Khan, Fan, Lu, Lau, JLT 2019) and tabulate their section structures. If either organizes the surveyed OC applications by ML algorithm class (e.g., supervised methods, unsupervised methods, reinforcement learning, or named algorithms such as SVM, ANN, k-means, PCA, Q-learning), then the abstract's 'for the first time... ML viewpoint' assertion is false. Additionally, count the unique OC application papers discussed in [128] and in this survey; if [128] or [156] covers a comparable or larger set, the 'much more investigations / more generality' claim is also unsupported. This is a bibliographic check and requires no new experiments.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract and Section I.A state that 'for the first time' ML-for-OC is reviewed from an ML viewpoint, because previous tutorials took an OC/network view and 'neglected ML view.' This claim requires that no prior survey organized the literature primarily by ML algorithm families. The paper's own bibliography contradicts that requirement: reference [128] is Zibar et al., 'Machine learning techniques in optical communication,' J. Lightwave Technol. 34(6), 2016, which is a prior journal tutorial on ML techniques in OC. The paper even uses [128] as a technical citation in §III.B, so it cannot be dismissed as unrelated. Reference [156] (Khan et al., 'An Optical Communication's Perspective on Machine Learning and Its Applications,' JLT 2019) is a second possible ML-oriented prior review, and [153] already surveys ML techniques in optical networks. No passage explains how an 'ML viewpoint' differs from [128] or why [128] does not count; the Related Works list [152]–[156] omits [128] entirely. Because the claimed firstness is the stated reason the paper should exist, this contradiction undermines the central contribution. The companion claim that this survey covers 'much more investigations' is also unquantified, with no disclosed search or selection protocol, so the 'more generality' half of the claim has no evidentiary basis.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript claims to be the first survey of machine learning for optical communication organized from a machine-learning viewpoint. It groups a large set of cited works into supervised learning (SVM, ANN, DNN, CNN, kNN, random forest, regression), unsupervised learning (clustering, expectation maximization, PCA), and reinforcement learning (Q-learning, deep reinforcement learning). For each algorithm it gives a brief definition and then summarizes applications in fiber optical communication, optical networks, and wireless optical communication. The paper asserts that previous tutorials adopted only an optical-communication/network view and that this survey is both novel in viewpoint and broader in coverage.","tokens_in":28413,"tokens_out":5368,"duration_ms":50025,"significance":"A careful ML-centric tutorial of machine learning for optical communication would be useful, because the field is growing quickly and many practitioners come from an optical-communication background. The paper's organization by algorithm family is sensible, and the reference list is broad and includes much recent 2018–2019 work. Those are real strengths of the organizing scheme. However, the paper's stated reason for existence is the claim of firstness, and that claim is contradicted by material in the paper's own bibliography. The tutorial content also contains mathematically malformed equations and several inaccurate or misleading statements about ML methods. Since the value of a survey rests on the reliability and completeness of its narrative, the paper in its current form cannot serve as a dependable reference.","major_comments":[{"comment":"The paper's central contribution claim—'for the first time, this paper reviews ML for OC literature from ML viewpoint'—is contradicted by its own reference list. Reference [128] (Zibar et al., 'Machine learning techniques in optical communication,' J. Lightwave Technol. 34(6), 2016) is a journal tutorial organized around machine-learning techniques applied to optical communication, and it is even cited in §III.B for technical content. The Related Works discussion in §I.A compares only [152]–[156] and does not explain why [128] lacks the 'ML view' that this paper claims to be new. The conclusion in §V repeats that all previous works had an 'Optical Communication view,' which is not supportable in the presence of [128], [153], and [156]. Because firstness is the stated motivation for the survey, this contradiction is load-bearing and undermines the central claim as written.","section":"Abstract; §I.A; §V"},{"comment":"The SVM kernel equations are mathematically malformed. Equation (1) is written as K(x,xi)=a0+sum_i ai<x,xi>_i with no exponent, Eq. (2) is written as a0+sum_i ai(<x,xi>)^d_i, which is not the standard polynomial kernel, and Eq. (3) is written as exp(-sum_i ai(x—xi)_i), which is not a Gaussian/RBF kernel because it lacks a squared norm and an appropriate scale parameter. The text also states that 'polynomial and exponential kernels calculate separating hyperplane' without completing the sentence. For a tutorial whose advertised value is teaching ML algorithms to OC researchers, these errors in the central mathematical definitions are substantive and need correction.","section":"§II.A, Eqs. (1)–(3)"},{"comment":"The companion claim that 'compared with other works, this survey reviews much more investigations; therefore, it has more generality' is unquantified and unverifiable. The paper gives no search strategy, no inclusion or exclusion criteria, no total number of papers screened, and no bibliometric comparison with [152]–[156]. Without such a protocol, 'much more' is an assertion rather than a documented property of the survey, and it cannot support the paper's stronger-generality claim.","section":"Abstract; §V"}],"minor_comments":[{"comment":"Section III contains a duplicated subsection heading: 'A. Peak search, c-means, hieratical clustering' is followed later by another 'A. k-means clustering,' and the subsection numbering then continues with B and C. The organization should be repaired and renumbered.","section":"§III"},{"comment":"There are numerous language errors that impede readability, including 'quiet necessary,' 'cavities of this field,' 'shielded on Fiber OC,' 'a bbeter choice,' 'instigation of k-means,' and 'hieratical clustering.' A careful proofreading pass is needed.","section":"Throughout"},{"comment":"References [32] and [34] are identical entries, and references [108] and [143] are also identical. These duplicates should be removed and the citations renumbered.","section":"References"},{"comment":"Equation (8) writes the kNN distance as D(R,Li)=sqrt((R-Li)^2), which is dimensionally inconsistent for vector inputs; it should be a vector norm such as ||R-Li||_2 summed over the feature dimensions.","section":"§II.E, Eq. (8)"},{"comment":"Figure numbering and in-text references are not fully aligned, and several subsections in Section III are labeled inconsistently; for example, the second 'A' heading in Section III should be a different letter and the text around Figs. 14 and 17 could be more explicitly tied to the discussion.","section":"Figures and headings"},{"comment":"Reference [74] is a self-cited arXiv preprint used in §II.C in support of a claim about the fiber Kerr effect. Because it is not peer-reviewed and is cited without independent verification, its status should be clearly flagged or the claim should be supported by a published source.","section":"§II.C, Ref. [74]"}],"recommendation":"reject","confidential_remarks":"The central novelty claim is contradicted by the paper's own reference [128], and the tutorial content has enough technical inaccuracies that I cannot see a path to acceptance without a fundamental reframing and a substantive technical revision. The self-citation of [74] and the omission of [128] from the Related Works section are additional editorial concerns. If the authors were to resubmit a thoroughly revised version that abandons the 'first' claim, provides a documented search methodology, corrects the kernel equations, and cleans up the organization and language, the result could be a useful survey, but that would be a different paper."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"I read the Amirabadi survey with the \"ML view\" label. Bottom line: the reference list is genuinely broad and the algorithm-by-algorithm organization could be handy for someone entering the field, but the paper cannot stand as a survey because its main novelty claim is contradicted by material it itself cites, and the tutorial content has multiple errors.\n\nWhat actually works: the paper groups the literature by ML algorithm family (SVM, ANN, DNN, CNN, kNN, random forest, regression, clustering, Q-learning, DRL) and then splits each into Fiber OC, Network, WOC. That makes it easy to see where a particular algorithm has been applied, and the coverage is indeed wide. If you want a quick way to find which papers used k-means in VLC, this list helps.\n\nThe soft spots are serious. The abstract and Section I claim \"for the first time\" this reviews ML for OC from an ML viewpoint because prior tutorials took an OC/network view. But reference [128] is Zibar et al., \"Machine learning techniques in optical communication\" (JLT 2016), a review organized around ML techniques. The paper itself cites [128] in Section III.B. So the firstness claim collapses, and the Related Works list doesn't mention [128] at all. The same applies to the claim that this survey covers \"much more investigations\": there is no search protocol, no inclusion criteria, and no quantitative comparison.\n\nThe tutorial portions also need work. Equations (1)-(3) for SVM kernels are malformed: (1) looks like a linear expansion, (2) is missing a power on the whole dot product, and (3) lacks a normalizing scale; the variable ranges are not given. There is a duplicated \"A.\" subsection heading in Section III (both \"Peak search, c-means, hierarchical clustering\" and \"k-means clustering\" are labeled A), and some statements are simply wrong (e.g., \"SVM is a linear classifier\" without qualification, and the definition of kernelized SVM is muddled). Self-citation [74] is used to support a claim about DNN and hyperparameter tuning, and while self-citation isn't a flaw per se, here it is an unpublished preprint that is not independently verified.\n\nProportionately: this is not a case of one weak section. The errors are in the pedagogy and the novelty claim is load-bearing. But the reference list is real and the taxonomy is usable.\n\nIf this comes across my desk as an editor, I would not desk reject it outright. I'd send it to a referee with instructions to focus on the novelty claim and the correctness of the tutorial equations, and with the expectation of major revision. The author has done the work of assembling a broad bibliography; the paper just needs to drop the false firstness and clean up the technical content. As it stands, I would not cite it.","headline":"Broad bibliography, but the 'first ML-view' claim is contradicted by the paper's own reference [128], and the tutorial equations need correction.","tokens_in":28824,"tokens_out":2983,"would_cite":false,"duration_ms":29292,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"This survey maps machine learning for optical communication by algorithm family, claiming to be the first ML-side review of the field.","keywords":["machine learning","optical communication","survey","supervised learning","unsupervised learning","reinforcement learning","equalization","optical performance monitoring"],"falsifier":"Read reference [128], a prior review of machine learning techniques in optical communication, and compare the number of unique investigations collected here with the counts in [156] and [152]; if an earlier survey already organises the field by algorithm family or covers more studies, the paper's 'first ML viewpoint' and 'more generality' claims fall.","tokens_in":27995,"feed_emoji":"📡","tokens_out":7205,"duration_ms":155924,"temperature":0.7,"pith_summary":"This paper tries to establish that the research field joining machine learning to optical communication is best understood from the algorithm's side: group the papers by the machine-learning method being used (SVM, neural networks, decision forests, clustering, principal component analysis, Q-learning, and so on), and then look at what optical-communication jobs that method has been asked to do. The survey assembles more than a hundred studies into this ML-first taxonomy, separates them into supervised, unsupervised, and reinforcement learning, and within each family reviews applications in fiber links, optical networks, and wireless optical channels. Its stated reason for caring is practical: most people working on ML for optical communication are optical engineers, not ML specialists, so an algorithm-organised map helps them choose a method. The paper also claims this ML viewpoint is taken for the first time here and that its coverage is broader than earlier surveys.","feed_headline":"An algorithm-first map of machine learning for optical links","feed_subtitle":"Studies grouped into supervised, unsupervised, and reinforcement learning, so optical engineers can pick a method directly.","key_machinery":"The organising device is a two-level taxonomy: first the learning paradigm (supervised, unsupervised, reinforcement), then the algorithm family (SVM, ANN, DNN, CNN, kNN, random forest, regression; k-means and other clustering, EM, PCA; Q-learning, deep RL), and within each family a subdivision by application domain (fiber links, optical networks, wireless optical communications). The taxonomy does the argument's work: it turns a scattered set of experiments into a reusable lookup table, and it makes absences visible, such as the lack of semi-supervised learning in the field.","core_discovery":"The paper's central claim is that the ML-for-optical-communication literature can be and should be organised by the machine-learning algorithm rather than by the optical system or network layer. After brief definitions of each algorithm, it walks through supervised methods (support vector machines, artificial neural networks, deep neural networks, convolutional neural networks, k-nearest neighbours, random forests, regression), unsupervised methods (clustering variants, expectation-maximisation, principal component analysis), and reinforcement learning (Q-learning and deep reinforcement learning), noting for each where it has been applied in fiber, network, and wireless-optical settings. Read in the author's intended way, the survey establishes a gap map: many standard ML tools—extreme learning machines, semi-supervised learning, most deep reinforcement learning—are barely used in optical communication, and many optical applications have not yet been touched by ML. The author further claims that earlier surveys looked at the field from the optical-communication side, and that this is the first review to put the ML taxonomy first while covering a larger set of investigations.","pith_inferences":["The 'first ML viewpoint' claim should be read as a claim about emphasis, not an absolute about the literature: at least one prior review cited in the paper, reference [128], already discusses machine learning techniques for optical communication, so the real novelty is the algorithm-first organisation and the breadth of coverage rather than the viewpoint itself.","A quantitative follow-up could turn the survey's gap map into a table: for each algorithm family, count papers, report testbeds and data sets, and state reported performance gains; that would let optical engineers compare methods directly.","The taxonomy predicts that the next high-yield directions are the pairs with the fewest entries, such as deep reinforcement learning for dynamic network control and kernel-based equalizers for wireless optical links, since the paper lists only one or two examples per pair.","Any successful semi-supervised optical-communication demonstration would directly fill a hole the survey identifies, because the paper reports no semi-supervised investigations in the field."],"forward_implications":["An optical engineer who knows the impairment but not the ML toolbox can look up algorithms by family and find reported applications for detection, equalization, performance monitoring, modulation-format identification, fault prediction, routing, and bandwidth allocation.","The algorithm-first map exposes underused methods—extreme learning machines, kernel methods, and deep reinforcement learning appear in only a handful of papers—so the survey functions as a research agenda.","Because the taxonomy separates fiber, network, and wireless-optical applications, future surveys can keep the same skeleton and insert new papers without reorganising the field.","The reported lack of semi-supervised learning points to a concrete opening: tasks where labels are expensive, such as live-network monitoring, could be tackled with semi-supervised methods."],"supporting_citations":[{"why":"Supplies the definitions of SVMs and margin/regularization concepts used throughout the supervised-learning sections.","marker":"[1]"},{"why":"Prior comprehensive survey of AI in optical networks that the paper positions as taking a network-layer view rather than an ML view.","marker":"[152]"},{"why":"Prior overview of machine learning in optical networks, another baseline that the paper says lacks the ML-first organisation.","marker":"[153]"},{"why":"Prior tutorial on ML for network automation that the paper contrasts with its own algorithm-first approach.","marker":"[154]"},{"why":"Prior review focused on ML for fiber-induced nonlinearity in CO-OFDM, used to show the existing physical-layer viewpoint.","marker":"[155]"},{"why":"Prior review of ML from the optical-communication perspective, cited as the main example of an OC-view survey.","marker":"[156]"}],"fun_headline_variants":["ML-first tour of optical comm's untapped algorithms","Optical engineers: see ML methods, not just systems","The first ML-centric map of optical communication","New survey flips the view: ML before optics"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The paper's claim of being the first ML-view survey assumes that none of the earlier surveys it cites, including reference [128], already organised the literature by machine-learning method; if any did, the 'first time' claim loses its footing.","fun_headline_variants_meta":{"raw":{"variants":["ML-first tour of optical comm's untapped algorithms","Optical engineers: see ML methods, not just systems","The first ML-centric map of optical communication","New survey flips the view: ML before optics"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000221,"raw_usage":{"total_tokens":1483,"prompt_tokens":1011,"completion_tokens":472,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":627,"completion_tokens_details":{"reasoning_tokens":410}},"tokens_in":627,"tokens_out":472,"duration_ms":434266,"temperature":1.0,"reasoning_tokens":410,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:58:19.695495+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Read reference [128], a prior review of machine learning techniques in optical communication, and compare the number of unique investigations collected here with the counts in [156] and [152]; if an earlier survey already organises the field by algorithm family or covers more studies, the paper's 'first ML viewpoint' and 'more generality' claims fall.","supporting_citations":[{"cited_title":"J., Merayo, N., Singh, S","cited_arxiv_id":null,"evidence_quote":"Prior comprehensive survey of AI in optical networks that the paper positions as taking a network-layer view rather than an ML view."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Prior overview of machine learning in optical networks, another baseline that the paper says lacks the ML-first organisation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Prior tutorial on ML for network automation that the paper contrasts with its own algorithm-first approach."},{"cited_title":"N., Fan, Q., Lu, C., & Lau, A","cited_arxiv_id":null,"evidence_quote":"Prior review of ML from the optical-communication perspective, cited as the main example of an OC-view survey."}],"review_version":1}