{"id":"cb05839e-b70b-4d16-87f9-da69928fd030","arxiv_id":"2605.29826","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"LDKE framework localizes fact-specific layers and disentangles inputs to improve generalization and locality in multimodal knowledge editing for MLLMs.","lead":"The paper introduces LDKE, a framework that localizes edits to specific layers in multimodal models and disentangles relevant inputs to prevent unwanted changes. This targets reliability issues when updating knowledge in AI systems that handle both text and images.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"Reader's weakest_assumption directly matches the unverified implementability of the two modules. Since full text is unavailable for deeper inspection, the UNVERDICTED/low-confidence stance requires no adjustment.","tokens_in":1711,"tokens_out":231,"duration_ms":17860,"concrete_test":"Re-run the main results table (whatever reports generalization and locality metrics) after ablating the Fast Localization module entirely; if performance drops to baseline levels on related-context queries while locality remains high, the localization step is load-bearing as claimed.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's assessment already flags low confidence due to abstract-only review. With the full manuscript referenced but not supplied here, no concrete technical flaw in the central claim (localization plus disentanglement yielding precise generalization without locality loss) can be isolated. The two failure modes are presented as diagnosed causes, and the modules are described as direct remedies; absent the actual implementation details, equations, or experimental controls, the argument cannot be shown internally inconsistent or unsupported on its own terms.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper identifies two failure modes in existing Multimodal Knowledge Editing (MKE) methods for MLLMs—Causal Misalignment (edits confined to specific samples) and Feature Entanglement (unintended changes to coupled irrelevant information)—and proposes Localized and Disentangled Knowledge Editing (LDKE). LDKE uses a Fast Localization module to identify and update critical layers and a Disentanglement Classifier to route target-relevant inputs, claiming this yields precise edits that generalize to related contexts while preserving high locality, as shown in experiments across benchmarks and MLLMs.","tokens_in":1784,"tokens_out":498,"duration_ms":31136,"significance":"If the experimental claims hold with proper controls and baselines, LDKE could meaningfully advance knowledge editing for MLLMs by providing a more targeted mechanism that reduces side effects and improves generalization, which is valuable for applications requiring reliable factual updates in multimodal systems.","major_comments":[{"comment":"Abstract: The central claim of 'superior performance' and 'extensive experiments' demonstrating better edit propagation and locality is load-bearing, yet the text provides no quantitative results, baselines, error bars, or implementation details, preventing verification of whether the modules actually resolve the diagnosed failure modes without new trade-offs.","section":"Abstract"},{"comment":"The assumption that Causal Misalignment and Feature Entanglement are the dominant causes (and that the proposed modules address them without side effects) is not isolated empirically; without ablation studies or controls showing these are primary over other factors, the motivation for the specific Fast Localization and Disentanglement Classifier design remains under-supported.","section":"Introduction / Method"}],"minor_comments":[{"comment":"Clarify the exact routing mechanism of the Disentanglement Classifier with pseudocode or an equation, as the high-level description leaves implementation ambiguous.","section":"Method"},{"comment":"Ensure all benchmarks and MLLMs used are explicitly listed with citation, and add a limitations section discussing potential computational overhead of the localization step.","section":"Experiments"}],"recommendation":"major_revision","confidential_remarks":"The manuscript appears to be at a stage where core experimental evidence is missing from the provided text, raising questions about readiness for this venue; the citation pattern and novelty relative to prior MKE work should be checked against recent arXiv submissions in the area."},"author_rebuttal":{"model":"grok-4.3","summary":"We appreciate the referee's constructive feedback on our manuscript. We address each major comment below and outline the planned revisions.","responses":[{"response":"We agree that the abstract would benefit from including key quantitative results to support the claims. In the revised manuscript, we will update the abstract to highlight specific metrics from our experiments, such as improvements in edit generalization and locality compared to baselines. The full results with error bars, baselines, and implementation details are presented in the Experiments section.","revision_made":"yes","referee_comment":"[Abstract] Abstract: The central claim of 'superior performance' and 'extensive experiments' demonstrating better edit propagation and locality is load-bearing, yet the text provides no quantitative results, baselines, error bars, or implementation details, preventing verification of whether the modules actually resolve the diagnosed failure modes without new trade-offs."},{"response":"The failure modes are identified through analysis of existing methods' behaviors on multimodal data, as detailed in the Introduction. To strengthen the empirical isolation of these factors, we will incorporate additional ablation studies in the revision that separately control for localization and disentanglement effects, showing their specific role in the observed issues and the design's effectiveness.","revision_made":"yes","referee_comment":"[Introduction / Method] The assumption that Causal Misalignment and Feature Entanglement are the dominant causes (and that the proposed modules address them without side effects) is not isolated empirically; without ablation studies or controls showing these are primary over other factors, the motivation for the specific Fast Localization and Disentanglement Classifier design remains under-supported."}],"tokens_in":1310,"tokens_out":351,"duration_ms":22781,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that this paper names two concrete problems in multimodal knowledge editing—edits that fail to spread to related queries and edits that leak into unrelated but visually similar content—and offers a framework to target them directly.\n\nWhat is new is the explicit split into Causal Misalignment and Feature Entanglement, plus the two modules meant to fix them: a Fast Localization step that picks which layers to change and a Disentanglement Classifier that decides which inputs should trigger an edit. The claim is that this combination lets edits generalize while keeping high locality, which existing MKE methods miss.\n\nThe paper does a clean job of laying out why current approaches fall short on those two axes. That diagnosis is useful on its own for anyone already working in the area.\n\nThe obvious soft spot is the complete absence of results. The abstract asserts better performance across benchmarks and models but shows no tables, no baselines, no error bars, and no implementation specifics. Without those, the central claim cannot be checked, and we have no way to know whether the modules add cost or create fresh side effects.\n\nThis is for people already doing knowledge editing on vision-language models. A reader who needs practical fixes for MLLM maintenance might pick up the failure-mode framing even if the method itself needs more testing.\n\nIf the full paper contains proper experiments with controls and comparisons, it is worth sending to review. Right now the idea is coherent but the evidence is missing.","headline":"LDKE formalizes two failure modes in MKE and adds localization plus disentanglement modules, but the abstract supplies zero numbers or controls to check if they work.","tokens_in":2276,"tokens_out":373,"would_cite":false,"duration_ms":15508,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"A framework localizes fact-specific layers in multimodal models and disentangles relevant inputs to make knowledge edits generalize without unintended changes.","keywords":["multimodal knowledge editing","knowledge editing","multimodal large language models","model editing","disentanglement","localization","causal misalignment","feature entanglement"],"falsifier":"An experiment in which the Fast Localization module and Disentanglement Classifier produce no measurable gain in edit propagation or locality compared with baseline editing methods on the same benchmarks would falsify the central claim.","tokens_in":2611,"feed_emoji":"🤖","tokens_out":610,"duration_ms":17080,"temperature":0.7,"pith_summary":"The paper seeks to overcome two failure modes in multimodal knowledge editing where updates stay confined to single examples and mix up unrelated visual or semantic features. It formalizes Causal Misalignment and Feature Entanglement as the root problems, then introduces modules that locate the right layers for editing and route inputs to protect unrelated knowledge. If successful, edits would spread correctly to related queries while leaving other model behavior intact. This would matter for keeping large multimodal systems current without repeated full retraining or widespread side effects. Experiments on benchmarks claim better propagation of changes and higher locality than prior methods.","feed_headline":"Multimodal editing localizes layers and separates inputs for broader fact updates","feed_subtitle":"The method fixes generalization failures by targeting specific layers and routing inputs to avoid side effects on unrelated knowledge.","key_machinery":"Fast Localization module that identifies critical layers for efficient updates, paired with a Disentanglement Classifier that routes inputs to preserve unrelated knowledge.","core_discovery":"LDKE achieves precise and generalized editing by localizing fact-specific model layers and disentangling target-relevant inputs from irrelevant ones, with superior performance in propagating edits to related contexts while maintaining high locality.","pith_inferences":["The same localization-plus-disentanglement pattern might transfer to non-multimodal language models or vision-only systems facing similar editing issues.","If the modules prove stable, they could reduce reliance on expensive full-model retraining for keeping deployed multimodal systems up to date.","Combining the approach with parameter-efficient fine-tuning techniques could further lower the cost of repeated edits."],"forward_implications":["Edits propagate accurately to logically related queries while unrelated but visually or semantically linked information stays unchanged.","The method applies across multiple benchmarks and different multimodal large language models without loss of locality.","Updates become confined to fact-specific layers rather than affecting the entire model.","Input routing prevents feature entanglement that previously caused unintended alterations."],"fun_headline_variants":["LDKE localizes layers and disentangles inputs for MLLM edits","Localized layers disentangle inputs in multimodal knowledge editing","Disentangling inputs via layer localization in multimodal LLMs","LDKE achieves localized and disentangled edits in multimodal models","Layer localization separates inputs for generalized multimodal edits"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The two failure modes of Causal Misalignment and Feature Entanglement are the main reasons existing methods fail at generalization and locality, and the new modules can fix them without creating fresh problems or trade-offs.","fun_headline_variants_meta":{"raw":{"variants":["LDKE localizes layers and disentangles inputs for MLLM edits","Localized layers disentangle inputs in multimodal knowledge editing","Disentangling inputs via layer localization in multimodal LLMs","LDKE achieves localized and disentangled edits in multimodal models","Layer localization separates inputs for generalized multimodal edits"]},"model":"grok-4.3","cost_usd":0.004376,"raw_usage":{"total_tokens":2158,"prompt_tokens":600,"num_sources_used":0,"completion_tokens":69,"cost_in_usd_ticks":43762000,"prompt_tokens_details":{"text_tokens":600,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1489,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":600,"tokens_out":69,"duration_ms":10817,"temperature":1.0,"reasoning_tokens":1489,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-29T07:54:21.046500+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"An experiment in which the Fast Localization module and Disentanglement Classifier produce no measurable gain in edit propagation or locality compared with baseline editing methods on the same benchmarks would falsify the central claim.","supporting_citations":[],"review_version":1}