{"id":"ab74de41-9ba8-4e52-9cd5-c942fd730d03","arxiv_id":"2508.01620","paper_version":3,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"IMU applies a classifier-level influence approximation to reweight unlearning updates, matching uniform gradient ascent forgetting depth while improving average model utility by 30% on vision and language tasks.","lead":"IMU introduces a retain-data-free machine unlearning method that uses influence functions to dynamically allocate forgetting strength to individual samples in the forget set. If effective, it could ease privacy compliance by letting models forget targeted data without storing retain sets or sacrificing much performance.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Classifier-level influence approximation may mis-rank forget-sample priorities, weakening the reweighting benefit over uniform GA.","rationale":"The reader's weakest assumption directly identifies the approximation's fidelity as the load-bearing point. The proposed correlation check is a minimal, falsifiable test that would confirm or refute whether the priority signal is reliable enough to support the headline empirical claim. No stronger internal inconsistency was visible from the abstract and stated method.","tokens_in":1704,"tokens_out":309,"duration_ms":24802,"concrete_test":"On a 2-layer MLP or small ResNet trained on a 5k-sample subset, compute exact influence values via full Hessian-vector products for 200 forget points; compare Spearman rank correlation and top-20 overlap with the paper's classifier-level scores. If correlation < 0.65 or top-20 overlap < 40%, rerun the IMU vs. uniform GA comparison on the same subset and check whether the utility gap shrinks below 10%.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that the classifier-level approximation (avoiding full Hessian) produces influence scores whose ordering closely matches true parameter influence on the forget objective. If the approximation only captures output-layer sensitivity while ignoring feature-extractor contributions, reweighting will over- or under-penalize samples, so the reported 30% utility gain at identical unlearning depth could be an artifact of the particular datasets or random seeds rather than a general property of the method.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The manuscript introduces IMU, an influence-guided machine unlearning method that operates using only the forget set. It departs from uniform gradient ascent by employing influence functions as a priority signal to dynamically reweight unlearning updates, with a classifier-level approximation introduced to avoid full-model Hessian inversion. The central claim is that this yields competitive unlearning performance while improving model utility by an average of 30% over standard uniform GA across vision and language tasks, thereby mitigating the utility-forgetting trade-off.","tokens_in":1816,"tokens_out":532,"duration_ms":32890,"significance":"If the empirical claims and approximation hold under rigorous controls, the work would be significant for practical retain-set-free unlearning in privacy-constrained settings. It builds on established influence-function literature with an efficiency-focused approximation and provides empirical validation on standard tasks, offering a concrete mechanism to allocate unlearning strength non-uniformly rather than treating all forget samples equally.","major_comments":[{"comment":"Abstract and §4 (experimental results): The claim that IMU 'maintains identical unlearning depth while enhancing model utility by an average of 30%' is load-bearing for the central contribution, yet the abstract and results summary provide no details on exact baselines, error bars, number of random seeds, statistical significance tests, or precise definitions of 'unlearning depth' and 'utility' metrics; without these, it is impossible to determine whether the reported gain is robust or an artifact of particular datasets.","section":"Abstract and §4"},{"comment":"§3.2 (classifier-level influence approximation): The method relies on this approximation to produce influence scores whose ordering matches true parameter influence on the forget objective, enabling the reweighting benefit over uniform GA. However, the manuscript does not provide approximation-error bounds, a direct comparison of classifier-level scores versus full-Hessian influence on the same forget samples, or an ablation showing that mis-ranking would not erode the 30% utility gain; this is load-bearing because the skeptic concern (output-layer sensitivity ignoring feature-extractor contributions) directly threatens the priority-signal validity.","section":"§3.2"}],"minor_comments":[{"comment":"Notation for the influence approximation could be clarified with an explicit statement of which layers are treated as the 'classifier' versus the feature extractor.","section":"§3"}],"recommendation":"major_revision","confidential_remarks":"The citation pattern appears light on recent retain-free MU baselines that also avoid retain data; a more thorough comparison section would strengthen the positioning."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive and detailed feedback. We address each major comment below, clarifying our empirical reporting and the theoretical basis for the approximation while committing to targeted revisions for improved rigor.","responses":[{"response":"We agree that greater specificity is needed to substantiate the central claim. In the revised manuscript we will expand both the abstract and §4 to list the exact baselines (uniform gradient ascent plus the retain-set-free methods cited in the related work), report all metrics as mean ± standard deviation over five independent random seeds, include paired t-test p-values for the utility gains, and define the metrics explicitly: unlearning depth is the accuracy drop on the forget set relative to the original model, while utility is the average accuracy on the retain and test sets. The 30% figure is the mean relative utility improvement aggregated across the vision and language benchmarks.","revision_made":"yes","referee_comment":"[Abstract and §4] Abstract and §4 (experimental results): The claim that IMU 'maintains identical unlearning depth while enhancing model utility by an average of 30%' is load-bearing for the central contribution, yet the abstract and results summary provide no details on exact baselines, error bars, number of random seeds, statistical significance tests, or precise definitions of 'unlearning depth' and 'utility' metrics; without these, it is impossible to determine whether the reported gain is robust or an artifact of particular datasets."},{"response":"Section 3.2 supplies a theoretical argument that, for cross-entropy losses, the output-layer influence ordering is preserved under the approximation because the forgetting objective is dominated by the final linear layer. We acknowledge that empirical corroboration would further address the skeptic concern. In revision we will add (i) a direct comparison of classifier-level versus full-Hessian influence scores on a CIFAR-10 forget subset, reporting Spearman rank correlation, and (ii) an ablation contrasting influence-based reweighting against random reweighting to quantify impact on the utility gain. Tight general error bounds are difficult to derive for non-convex deep networks; we will instead expand the limitations discussion to state the assumptions under which the ordering is expected to hold.","revision_made":"partial","referee_comment":"[§3.2] §3.2 (classifier-level influence approximation): The method relies on this approximation to produce influence scores whose ordering matches true parameter influence on the forget objective, enabling the reweighting benefit over uniform GA. However, the manuscript does not provide approximation-error bounds, a direct comparison of classifier-level scores versus full-Hessian influence on the same forget samples, or an ablation showing that mis-ranking would not erode the 30% utility gain; this is load-bearing because the skeptic concern (output-layer sensitivity ignoring feature-extractor contributions) directly threatens the priority-signal validity."}],"tokens_in":1447,"tokens_out":605,"duration_ms":34097,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing here is a retain-free unlearning method that uses influence functions to reweight forget samples instead of applying the same update to all of them, plus a classifier-level approximation to skip full Hessian inversion. This is the core departure from uniform gradient ascent and from earlier geometric or auxiliary-statistic approaches in the retain-free space. The paper does a clean job laying out why heterogeneous influence matters and why the approximation keeps things tractable for larger models. If the reported 30% utility improvement at matched unlearning depth holds, it would be a practical step for settings where retain data cannot be kept. The experiments span vision and language tasks and position the method as competitive, which is a reasonable scope for this kind of work. The soft spots are in the evidence. The abstract gives the headline numbers but no visible baselines, error bars, statistical tests, or exact controls, so it is hard to judge how general the gain really is. The classifier-level approximation is load-bearing; if it only tracks output-layer sensitivity and misses feature-extractor contributions, the reweighting could over- or under-penalize samples and the advantage over uniform GA might shrink or vanish on other data. That concern from the stress test is worth checking in the full experiments. This paper is for researchers working on practical machine unlearning under privacy constraints. Readers who need retain-free options would find the formulation worth reading. It has a clear enough idea and motivation to deserve a serious referee, mainly to verify the approximation and the experimental claims. I would send it to peer review after the authors add more detail on the runs and the approximation validation.","headline":"IMU uses influence functions for dynamic reweighting in retain-free unlearning and claims a 30% utility gain over uniform GA, but the experimental backing is thin.","tokens_in":2310,"tokens_out":402,"would_cite":false,"duration_ms":40171,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":{"model":"grok-4.3","evidence":[{"relation":"unclear","rs_module":"IndisputableMonolith/Cost/FunctionalEquation.lean","rs_theorem":"washburn_uniqueness_aczel","paper_passage":"To circumvent the prohibitive cost of full-model Hessian inversion, we introduce a theoretically grounded classifier-level influence approximation... estimate the influence value at the classifier."},{"relation":"unclear","rs_module":"IndisputableMonolith/Foundation/ArithmeticFromLogic.lean","rs_theorem":"absolute_floor_iff_bare_distinguishability","paper_passage":"IMU automatically adjusts the unlearning strength for each forgetting data point proportionally to its influence score"}],"headline":"IMU influence reweighting operates in ML unlearning domain with no RS-shaped J-cost or ratio-symmetric structure","alignment":"orthogonal","rationale":"Paper's core is classifier-level influence approximation (Eq. 3-4) to reweight GA loss for heterogeneous forget samples, avoiding full Hessian. This is standard ML influence machinery (Koh & Liang) with practical approximations; no golden-ratio identities, cosh-cost J(x), 8-tick periodicity, or parameter-free forcing from a single distinction appear. RS theorems (e.g., washburn_uniqueness_aczel, reality_from_one_distinction) derive physics constants from J-cost; IMU makes no such claims or structural parallels.","tokens_in":60111,"confidence":"high","tokens_out":315,"duration_ms":13001,"cache_read_input_tokens":38528,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"IMU reweights unlearning updates using influence scores on forget samples alone to match uniform gradient ascent forgetting while raising average model utility by 30%.","keywords":["machine unlearning","influence functions","gradient ascent","forget set","model utility","privacy","reweighting","approximation"],"falsifier":"Replace the influence-derived weights with random weights of the same distribution and measure whether the utility gain over uniform gradient ascent disappears while forgetting metrics remain unchanged.","tokens_in":2617,"feed_emoji":"🧠","tokens_out":588,"duration_ms":48648,"temperature":0.7,"pith_summary":"Machine unlearning removes the effects of chosen training points from a model without full retraining. Existing retain-free techniques forget every sample with equal force, which often harms performance on data the model should keep. IMU instead computes an influence score for each forget sample and uses it to scale the strength of the unlearning step applied to that sample. A classifier-level approximation supplies the scores without inverting the full model Hessian. Experiments show the result is equally thorough forgetting but noticeably better retained accuracy on vision and language tasks.","feed_headline":"Influence scores raise unlearning utility 30% at same forgetting depth","feed_subtitle":"IMU weights each forget sample by its influence and applies stronger updates only where they matter most, using no retain data.","key_machinery":"Classifier-level influence approximation that ranks each forget sample by its contribution to the unlearning objective and dynamically scales the gradient update strength for that sample.","core_discovery":"By treating forget samples as heterogeneous rather than uniform, IMU allocates stronger gradient-ascent updates to those that most support the forgetting goal. The classifier-level influence approximation supplies the ranking signal at low cost. This produces unlearning that reaches the same depth as standard uniform gradient ascent yet preserves roughly 30 percent more model utility on average.","pith_inferences":["Deployed models could satisfy data-deletion requests with less accuracy loss if influence-guided updates replace uniform ones.","Extending the influence signal past the final classifier layer could improve results on very deep networks.","Heterogeneous update strengths may help other model-editing tasks where only a subset of knowledge must change."],"forward_implications":["Unlearning becomes feasible when no retain set can be stored or accessed.","The utility-forgetting trade-off seen in uniform methods is reduced without weakening the removal of target data.","The same reweighting idea can be tested on other gradient-based unlearning procedures.","Performance gains appear consistently on both vision and language benchmarks."],"fun_headline_variants":["Influence scores enable 30% utility gain in retain-free unlearning","IMU matches unlearning depth with 30% higher utility using influence","Same forget depth 30% more utility with influence-guided IMU","IMU weights forget samples by influence for 30% better utility"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The classifier-level influence scores give a sufficiently accurate ordering of which forget samples warrant the strongest updates.","fun_headline_variants_meta":{"raw":{"variants":["Influence scores enable 30% utility gain in retain-free unlearning","IMU matches unlearning depth with 30% higher utility using influence","Same forget depth 30% more utility with influence-guided IMU","IMU weights forget samples by influence for 30% better utility"]},"model":"grok-4.3","cost_usd":0.01214,"raw_usage":{"total_tokens":5291,"prompt_tokens":655,"num_sources_used":0,"completion_tokens":71,"cost_in_usd_ticks":121399500,"prompt_tokens_details":{"text_tokens":655,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4565,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":655,"tokens_out":71,"duration_ms":53946,"temperature":1.0,"reasoning_tokens":4565,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-19T01:14:29.481049+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Replace the influence-derived weights with random weights of the same distribution and measure whether the utility gain over uniform gradient ascent disappears while forgetting metrics remain unchanged.","supporting_citations":[],"review_version":1}