{"id":"ba1cc23c-e685-403d-9134-b1d8835e71da","arxiv_id":"2508.02414","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":4.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"The abstract proposes ASMR, a hyperparameter-free angular-distance filter for malfunctioning clients in federated learning, but the submitted full text is an unrelated knowledge-graph paper.","lead":"A federated-learning defense called ASMR is proposed to filter out bad client updates by their angular distance from the others. The abstract claims no hyperparameters and no knowledge of how many clients are malfunctioning, but the submitted full text is a different paper.","discovery_kind":"unclear","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Supplied full text does not describe ASMR; the central claims are absent, so no technical evaluation of the method is possible.","rationale":"The reader correctly found that the full text does not contain the ASMR method and assigned UNVERDICTED. My stress-test pass confirms this mechanically: the supplied body is a different paper by different authors on a different topic, so the abstract's claims have no supporting method, definitions, or experiments in the artifact. The weakest-assumption analysis in the reader's report focuses on the substantive risk that angular distance may not separate all malfunction types; that is a reasonable scientific concern, but it is downstream of a more basic issue: no algorithm is present to analyze. I therefore partially agree: the reader identified the same underlying problem in the rationale, but the formal 'weakest_assumption' field targets a methodological premise rather than the missing-artifact problem. No further technical objection can be raised without fabricating an analysis of an absent algorithm. The verdict should remain UNVERDICTED, and the concrete next step is to retrieve the correct arXiv source and verify whether the ASMR description and histopathology experiments actually exist. If the correct full text is subsequently provided, the angular-separation assumption and the hyperparameter-free dynamic-boundary claim should be stress-tested on that text.","tokens_in":9524,"tokens_out":2036,"duration_ms":25711,"concrete_test":"Obtain the actual arXiv:2508.02414 PDF or HTML from arXiv's own source files and search for the strings 'ASMR', 'angular distance', and 'histopathological'. If none of these appears, or if the text does not define an angular-distance exclusion rule with a dynamic threshold, then the abstract's central claims have no corresponding method or experiments in the artifact, and the work remains unverdictable.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The abstract claims ASMR dynamically excludes malfunctioning clients by angular distance, requires no hyperparameters, and needs no knowledge of the number of malfunctioning clients, with experiments on a histopathological image-classification task. For these claims to be assessable, the full text must define the update space, the angular-distance criterion, the exclusion rule, and the dynamic decision boundary, and must report the corresponding experiments. The provided full text is arXiv:2508.02413v2, 'Improving Knowledge Graph Understanding with Contextual Views' by Christou and Shimizu, a human-computer-interaction study of a knowledge-graph browser. It contains no mention of ASMR, no angular-distance-based exclusion rule, no dynamic decision boundary, and no histopathological federated-learning experiment. The central claim therefore has no supporting artifact in the text under review. This is not a defect in ASMR's reasoning as such; it is an unreviewability condition: there is no method description or experiment to stress-test. The reader's weakest assumption about angular separation reliability is plausible, but it cannot even be examined because the algorithm itself is missing. The appropriate disposition is UNVERDICTED, not ACCEPT or REJECT, since no substantive argument is available for evaluation.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The abstract proposes ASMR, a federated-learning defense that dynamically excludes malfunctioning clients by angular distance, claims no hyperparameters or prior knowledge of the number of malfunctioning clients, and reports experiments on histopathological image classification. The submitted full text is a different paper, \"Improving Knowledge Graph Understanding with Contextual Views\" by Christou and Shimizu, which presents the InK Browser for knowledge-graph exploration and a user study with statistical tests. The full text contains no mention of ASMR, no angular-distance criterion, no dynamic exclusion rule, and no histopathology or federated-learning experiments. Consequently, the manuscript does not supply the content needed to evaluate its stated central claim.","tokens_in":9724,"tokens_out":4740,"duration_ms":53325,"significance":"If the claimed result held, ASMR would be a practically valuable contribution to federated-learning robustness, since current defenses often require knowing the number of malicious clients or tuning hyperparameters. However, the manuscript as submitted provides no algorithm, derivation, or experimental evidence for this claim. There is no code, no formal proof, and no reproducible evaluation to credit. The significance of the paper cannot be assessed from the artifact under review.","major_comments":[{"comment":"The abstract's central claims—that ASMR dynamically excludes malfunctioning clients based on angular distance, requires no hyperparameters, and needs no knowledge of the number of malfunctioning clients—are entirely unsupported by the submitted text. Sections 1–8 describe an unrelated manuscript on the InK Browser for knowledge graphs, with no ASMR method, no update-space definition, no angular-distance metric, no exclusion rule, and no dynamic decision-boundary procedure. The artifact therefore does not contain the method or the evidence it promises.","section":"Abstract vs. full text"},{"comment":"The abstract claims that experiments showcase ASMR's detection capabilities on a histopathological image-classification task and that findings on dynamically adapting decision boundaries are presented. The Results section reports only a user study comparing task accuracy and completion time for knowledge-graph navigation; there are no federated-learning or histopathology results. No table or figure in the manuscript addresses ASMR, so the experimental claim cannot be checked.","section":"Section 6"}],"minor_comments":[{"comment":"Section 5.2.5 repeats Section 5.2.1 and Section 5.2.6 repeats Section 5.2.2 nearly verbatim; these duplicates should be removed in any revision.","section":"Section 5.2.5–5.2.6"},{"comment":"The phrase \"For ouf follow up responses\" contains a typo ('ouf' should be 'our').","section":"Section 5.2.1"}],"recommendation":"reject","confidential_remarks":"The mismatch between abstract and full text appears to be a file-submission error: the full text is a different manuscript entirely. If a correct ASMR manuscript exists, the editor may wish to contact the authors before final disposition. As submitted, however, there is no technical content to referee, and the appropriate disposition is to reject or require resubmission."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe paper you sent me isn't the paper you think it is. The abstract describes ASMR, a federated-learning defense that filters clients by angular distance and claims to need no hyperparameters and no knowledge of the number of bad clients. The full text is arXiv:2508.02413v2, \"Improving Knowledge Graph Understanding with Contextual Views,\" an unrelated HCI study of a knowledge-graph browser. The artifact under review contains zero content from the claimed work: no method, no experiments, no histopathology results, no related work, no equations. There is nothing to evaluate technically.\n\nWhat is genuinely new here, as far as I can tell from the abstract alone, is the specific claim of a hyperparameter-free dynamic decision boundary for angular-distance filtering. That is a plausible incremental contribution within a well-established family of robust-aggregation defenses. The abstract is clearly written and honest about its prerequisites — that alone is a small point in its favor. But none of this can be confirmed from the submitted text.\n\nThe soft spot is not a technical flaw in the method; it is an absent submission. The mismatch is objective and mechanical. The reader's concern about whether angular distance reliably separates all malfunction modes is reasonable, but it cannot even be examined because the algorithm itself is missing. A serious referee would have nothing to check: no derivations, no experimental details, no comparison baselines, no statistical analysis.\n\nMy take: this should be desk-rejected as submitted, not because the underlying idea is obviously wrong, but because the submission does not contain the paper it claims to be. The authors should resubmit with the correct full text. If the actual ASMR paper exists and matches the abstract, it may well deserve a proper review — I just cannot judge it from this envelope.\n\nFor you: skip this for the reading group and don't cite it. There's no content to engage with. If the correct version appears later, pass it along and I'll give it a real look.","headline":"The submitted file is the wrong paper — the ASMR abstract has no matching full text, so there is nothing to review.","tokens_in":10227,"tokens_out":1697,"would_cite":false,"duration_ms":20077,"reading_group":"no","serious_thinker":"unclear","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"A federated-learning defense that needs no attack count.","keywords":["federated learning","client malfunction","angular distance","robust aggregation","dynamic decision boundary","poisoning defense","histopathological image classification"],"falsifier":"Run ASMR on a benchmark where an attacker crafts a malfunctioning update that is nearly collinear with the healthy update directions; if the update is not excluded and global accuracy degrades significantly, the core angular-separation premise fails. A second check is to sweep the fraction of malfunctioning clients continuously and see whether the no-hyperparameter dynamic boundary still excludes the right set at every fraction.","tokens_in":9337,"feed_emoji":"🛡️","tokens_out":3307,"duration_ms":38239,"temperature":0.7,"pith_summary":"This paper, in its abstract, introduces Angular Support for Malfunctioning Client Resilience (ASMR), a defense for federated learning that excludes malfunctioning clients by their angular distance in update space. The claimed advance is that ASMR needs no hyperparameters and no knowledge of how many clients are malfunctioning, which would remove two impractical prerequisites of current defenses. The abstract reports experiments on a histopathological image classification task showing detection capabilities and findings on dynamically adapting decision boundaries. The supplied full text is a different manuscript about a knowledge-graph browsing tool; the ASMR method, its experiments, and its references are not present in that text, so the pith here rests on the abstract alone.","feed_headline":"A federated-learning defense that needs no attack count","feed_subtitle":"ASMR excludes malfunctioning clients by update angle, no hyperparameters needed.","key_machinery":"The central object is the angular distance between client update vectors in model-parameter space. ASMR uses this distance as the signal that identifies malfunctioning clients, and the load-bearing mechanism is a decision boundary that adapts dynamically, which is what allows the method to claim it needs neither hyperparameters nor an a priori malfunctioning-client count. Only the abstract describes this machinery; the full text includes no derivation or algorithmic details.","core_discovery":"The paper claims that client updates in federated learning can be screened by their angular distance from other updates, and that a dynamically adapted decision boundary can separate healthy from malfunctioning clients without a preset threshold. ASMR is presented as the mechanism that performs this screening: clients whose updates lie at anomalous angular distance are excluded, and the boundary adapts as training proceeds. If true, this would make robust federated learning feasible in settings where the number of malfunctioning clients is unknown, covering technical faults, disadvantageous training data, and malicious attacks alike. The full text supplied with this paper does not contain that method.","pith_inferences":["A natural test beyond the reported setting is to vary the attack: label flipping, targeted model poisoning, and scaling attacks change the angular signature of the malfunctioning client, and the method's stated coverage of all three is only credible if it performs on each.","The 'no hyperparameters' claim is stronger than it may seem; any dynamic boundary rule encodes an implicit criterion, such as a gap or quantile in the angular-distance distribution, so the paper would need to show that the rule is not silently tuned to the dataset.","If the angular-separation assumption holds, the same geometric screening could be reused in other distributed-optimization settings beyond federated learning, such as Byzantine-resilient decentralized SGD."],"forward_implications":["Deployments can drop the common requirement of knowing the number of attackers before training starts.","The defense can adapt when the number of malfunctioning clients changes mid-training, since the boundary is dynamic.","A single detection rule would cover technical failures, poor local data, and malicious updates.","The method is tested on an image classification task with histopathological data."],"supporting_citations":[],"fun_headline_variants":["Exclude malfunctioning FL clients by angle","Federated learning defense with no attack count","Angle distance spots bad clients in federated learning","No hyperparameters needed for client screening","Dynamic angular exclusion for robust federated learning"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The method assumes that malfunctioning client updates point in a detectably different direction from healthy updates, and that a dynamic boundary with no hyperparameters can always locate that split.","fun_headline_variants_meta":{"raw":{"variants":["Exclude malfunctioning FL clients by angle","Federated learning defense with no attack count","Angle distance spots bad clients in federated learning","No hyperparameters needed for client screening","Dynamic angular exclusion for robust federated learning"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000372,"raw_usage":{"total_tokens":1920,"prompt_tokens":809,"completion_tokens":1111,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":425,"completion_tokens_details":{"reasoning_tokens":1044}},"tokens_in":425,"tokens_out":1111,"duration_ms":12591,"temperature":1.0,"reasoning_tokens":1044,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T04:56:55.501274+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run ASMR on a benchmark where an attacker crafts a malfunctioning update that is nearly collinear with the healthy update directions; if the update is not excluded and global accuracy degrades significantly, the core angular-separation premise fails. A second check is to sweep the fraction of malfunctioning clients continuously and see whether the no-hyperparameter dynamic boundary still excludes the right set at every fraction.","supporting_citations":[],"review_version":1}