{"id":"d0512480-b16f-45d5-8e5b-80e8d4c32b71","arxiv_id":"2605.15700","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"AGOP-IxG filters per-sample gradients with a top-K truncated average gradient outer product matrix and outperforms SHAP, Integrated Gradients, InputXGradient, and LIME on Spearman correlation and noise mass across three synthetic tabular tasks while running 350-1650x faster.","lead":"The paper introduces AGOP-IxG, a fast local feature attribution method that multiplies per-sample gradients by a rank-truncated average gradient outer product matrix for tabular classifiers. Practitioners might read it to obtain quicker explanations than SHAP while maintaining competitive fidelity on controlled benchmarks with known ground truth.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Global AGOP computed on training distribution may not align with local per-sample gradients in nonlinear regimes","rationale":"The reader's weakest_assumption correctly isolates the single assumption whose failure would most directly undermine the local-fidelity claims on the nonlinear benchmarks. Because the full manuscript is now accessible, the concern can be stated more precisely in terms of the method's construction (global vs. local outer-product alignment) rather than merely noting missing implementation details. No other internal inconsistency or derivation gap appears more load-bearing than this one.","tokens_in":1817,"tokens_out":395,"duration_ms":34076,"concrete_test":"On the sparse nonlinear synthetic dataset, replace the global AGOP with a locally estimated version obtained by averaging gradient outer products over the 50 nearest training neighbors of each test sample (using Euclidean distance in feature space); recompute Spearman rank correlation for AGOP-IxG. If the metric drops by more than 0.15 relative to the global version while InputXGradient remains unchanged, the global-filter assumption is responsible for the claimed superiority.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"AGOP-IxG pre-multiplies each test-sample gradient by a fixed top-K truncated AGOP matrix derived from the full training distribution. For the reported gains in Spearman rank correlation and top-k precision on the sparse nonlinear and interaction synthetic datasets to be attributable to the method rather than to dataset-specific artifacts, this global matrix must preserve or enhance the locally relevant directions of the per-sample gradient. In a highly nonlinear model the average outer product can cancel opposing gradient signs across different regions of feature space, producing a filter that attenuates rather than amplifies the true local attribution. The paper evaluates on three synthetic tasks that include nonlinearity, yet provides no diagnostic that measures alignment between the global AGOP and a locally estimated outer product (e.g., via neighborhood sampling) at the test points where the headline metrics are computed.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper introduces AGOP-IxG, a local feature attribution method for tabular classifiers that pre-multiplies each per-sample gradient by a top-K rank-truncated Average Gradient Outer Product (AGOP) matrix computed from the training distribution. It constructs three synthetic multi-class tabular tasks (linear, sparse nonlinear, interaction-based) with analytically or numerically derivable ground-truth attributions and compares AGOP-IxG against SHAP (DeepExplainer), Integrated Gradients, InputXGradient, and LIME. AGOP-IxG leads on Spearman rank correlation and noise feature mass across all three tasks and on top-k precision for the interaction task, while being 350x–1650x faster than SHAP. On real datasets (Adult Income, Credit Card Default) it shows comparable global faithfulness under the ROAR protocol.","tokens_in":2001,"tokens_out":669,"duration_ms":29026,"significance":"The controlled benchmark with ground-truth attributions on synthetic tasks is a clear strength that enables rigorous, falsifiable evaluation of local attribution fidelity, addressing a persistent gap in the field. If the performance claims hold, AGOP-IxG supplies a practical, computationally lightweight alternative for per-sample explanations in tabular AutoML pipelines. The reported speed advantage over SHAP is directly relevant for deployment. The work also demonstrates awareness of the distinction between local and global faithfulness by separating the two evaluation regimes.","major_comments":[{"comment":"Part 1, synthetic tasks: the reported gains on the sparse nonlinear and interaction datasets rest on the assumption that a single global AGOP matrix (top-K truncated) computed on the training distribution preserves or enhances the locally relevant directions of per-sample gradients at test points. No diagnostic is supplied that compares this global matrix to a locally estimated outer product (e.g., via neighborhood sampling around the test instances where headline metrics are measured). This is load-bearing because, in nonlinear regimes, sign cancellation across regions can produce a filter that attenuates rather than amplifies true local attributions.","section":"Part 1"},{"comment":"Method description, top-K truncation: the top-K rank threshold is explicitly listed as a free parameter. The manuscript must specify the exact procedure used to choose K for each dataset and confirm that selection was performed without reference to the evaluation metrics (Spearman correlation, top-k precision, noise mass) to eliminate circularity risk in the central performance claims.","section":"Method"}],"minor_comments":[{"comment":"Abstract: the speedup range (350× to 1,650×) should be accompanied by the model architectures, batch sizes, and hardware used for the timing measurements to support reproducibility.","section":"Abstract"},{"comment":"Part 2, ROAR results: the statement that methods cluster within ~1.7% relative AUC would be strengthened by reporting standard deviations across runs or statistical significance tests.","section":"Part 2"}],"recommendation":"major_revision","confidential_remarks":"The manuscript aligns with the scope of interpretability-focused machine-learning venues. The citation pattern appears balanced; no obvious omission of key prior work on gradient-based attribution or synthetic benchmarks is evident from the provided material."},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the constructive and detailed comments. The points raised are substantive and we address each one directly below, outlining the revisions we will make to the manuscript.","responses":[{"response":"We agree that an explicit diagnostic comparing the global AGOP to local estimates would strengthen the paper. In the revision we will add a new subsection that computes local AGOP matrices via neighborhood sampling (k-nearest neighbors in feature space) around each test point, compares the leading eigenvectors to those of the global matrix, and reports the resulting attribution metrics when the local filter is substituted for the global one. This will directly test whether sign cancellation attenuates performance on the nonlinear and interaction tasks. The current strong results with ground-truth labels provide supporting evidence, but we accept that the requested diagnostic is a valuable addition.","revision_made":"yes","referee_comment":"Part 1, synthetic tasks: the reported gains on the sparse nonlinear and interaction datasets rest on the assumption that a single global AGOP matrix (top-K truncated) computed on the training distribution preserves or enhances the locally relevant directions of per-sample gradients at test points. No diagnostic is supplied that compares this global matrix to a locally estimated outer product (e.g., via neighborhood sampling around the test instances where headline metrics are measured). This is load-bearing because, in nonlinear regimes, sign cancellation across regions can produce a filter that attenuates rather than amplifies true local attributions."},{"response":"We will expand the method section to state the precise selection rule: K is the smallest integer such that the sum of the top-K eigenvalues of the training-set AGOP accounts for at least 95 % of the trace. This threshold is computed once on the training distribution before any test-set evaluation or metric computation. We confirm that K was never tuned against Spearman correlation, top-k precision, or noise-mass values. The revision will also include a sensitivity table showing metric variation for K values bracketing the chosen threshold.","revision_made":"yes","referee_comment":"Method description, top-K truncation: the top-K rank threshold is explicitly listed as a free parameter. The manuscript must specify the exact procedure used to choose K for each dataset and confirm that selection was performed without reference to the evaluation metrics (Spearman correlation, top-k precision, noise mass) to eliminate circularity risk in the central performance claims."}],"tokens_in":1609,"tokens_out":509,"duration_ms":25388,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The paper introduces AGOP-IxG, which pre-multiplies per-sample gradients by a top-K truncated average gradient outer product matrix from the training distribution. It reports leading Spearman correlation and noise feature mass on three synthetic tabular tasks plus large speed gains over SHAP.","headline":"AGOP-IxG gives a fast local attribution filter for tabular data with clear wins on synthetic ground-truth tasks, but the global training-derived matrix needs checks for nonlinear alignment.","tokens_in":2490,"tokens_out":134,"would_cite":false,"duration_ms":25047,"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":"AGOP-IxG ... pre-multiplies the per-sample gradient by a top-K rank-truncated Average Gradient Outer Product matrix"},{"relation":"unclear","rs_module":"IndisputableMonolith/Foundation/AbsoluteFloorClosure.lean","rs_theorem":"absolute_floor_iff_bare_distinguishability","paper_passage":"M = 1/n ∑ g_i g_i^T ... eigendecompose M = V Λ V^T"}],"headline":"AGOP gradient-covariance filter for tabular attribution operates in ML explainability domain with no structural overlap to RS distinction-forcing or J-cost machinery","alignment":"orthogonal","rationale":"The paper's core construction (training-set AGOP matrix M, top-K eigendecomposition, per-sample gradient pre-multiplication g M_K followed by InputXGradient) is a second-moment filter on gradients for local feature scores. RS derives J(x)=½(x+x⁻¹)−1, φ-ladders, 8-tick periodicity and spacetime from a single distinction via machine-checked forcing (reality_from_one_distinction, AbsoluteFloorClosure, Cost.FunctionalEquation). No shared primitives, cost functions, ratio symmetry or periodicity appear; the domains (cs.LG attribution vs. parameter-free physics) are disjoint.","tokens_in":48221,"confidence":"high","tokens_out":342,"duration_ms":15321,"cache_read_input_tokens":32896,"cache_creation_input_tokens":0},"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"AGOP-IxG filters per-sample gradients with a training-derived covariance matrix to improve local feature attributions on tabular classifiers.","keywords":["feature attribution","tabular data","gradient outer product","local explanations","synthetic benchmarks","model interpretability","AutoML","SHAP comparison"],"falsifier":"Compute ground-truth attributions on a new synthetic dataset that deliberately places test samples in a region of feature space distant from the training distribution and measure whether AGOP-IxG's Spearman correlation drops below that of plain InputXGradient.","tokens_in":2698,"feed_emoji":"📊","tokens_out":831,"duration_ms":43376,"temperature":0.7,"pith_summary":"The paper introduces AGOP-IxG as a method that pre-multiplies each input sample's gradient by a low-rank version of the average gradient outer product matrix computed over the training set. This produces feature attributions for tabular models that are compared against SHAP, Integrated Gradients, InputXGradient, and LIME on three synthetic datasets where ground-truth attributions can be calculated exactly. AGOP-IxG records the highest Spearman rank correlation with true attributions and assigns the least mass to noise features across linear, sparse nonlinear, and interaction-based tasks, while running several hundred times faster than SHAP. On two real tabular datasets the method yields global faithfulness scores within 1.7 percent of the other techniques when measured by the ROAR protocol. The design targets local per-sample explanations rather than global feature ranking.","feed_headline":"AGOP-IxG beats SHAP on tabular attribution accuracy and speed","feed_subtitle":"On synthetic tasks with known ground truth the gradient-covariance filter shows higher rank correlation and lower noise mass while running 1","key_machinery":"The top-K rank-truncated Average Gradient Outer Product (AGOP) matrix, which encodes the dominant directions of gradient variation across the training distribution and serves as a fixed linear filter applied to each test sample's gradient.","core_discovery":"Pre-multiplying the per-sample gradient vector by the top-K rank-truncated Average Gradient Outer Product matrix produces attributions whose ordering and magnitude align more closely with analytically known ground-truth feature contributions on synthetic tabular classification tasks than the attributions returned by SHAP, Integrated Gradients, InputXGradient, or LIME, while requiring orders-of-magnitude less computation.","pith_inferences":["The same fixed AGOP matrix could be reused across an entire model family or across successive retrainings without recomputation, further reducing cost in production pipelines.","Because the filter is derived from gradient statistics rather than from the loss surface directly, the approach may extend naturally to regression or ranking models where per-sample gradients are still available.","If the AGOP matrix is recomputed periodically on a sliding window of recent data, the method could adapt to mild distribution drift while retaining its speed advantage.","The observed clustering of global faithfulness scores suggests that local attribution quality and global feature ranking are partially orthogonal objectives that may require separate evaluation protocols."],"forward_implications":["On the linear synthetic task AGOP-IxG attains the highest Spearman correlation with the true linear coefficients.","Across all three synthetic tasks the method assigns lower total attribution mass to the injected noise features than any baseline.","On the interaction dataset AGOP-IxG records the best top-k precision for recovering the known interacting features.","Wall-clock time for a full test-set explanation is between 350 and 1,650 times lower than DeepExplainer SHAP under identical hardware.","Global ROAR AUC on Adult Income and Credit Card Default stays within roughly 1.7 percent relative difference of the other four methods."],"fun_headline_variants":["AGOP-IxG leads SHAP on synthetic tabular attribution rank correlation","AGOP-IxG covariance filter outperforms baselines on ground truth tasks","Gradient outer product method runs hundreds of times faster than SHAP","AGOP-IxG shows higher fidelity local attributions on controlled benchmarks"],"cache_read_input_tokens":64,"weakest_assumption_plain":"The average gradient outer product computed once on the training distribution remains a useful filter even for test points that lie far from the training distribution or inside highly nonlinear regions of the model.","fun_headline_variants_meta":{"raw":{"variants":["AGOP-IxG leads SHAP on synthetic tabular attribution rank correlation","AGOP-IxG covariance filter outperforms baselines on ground truth tasks","Gradient outer product method runs hundreds of times faster than SHAP","AGOP-IxG shows higher fidelity local attributions on controlled benchmarks"]},"model":"grok-4.3","cost_usd":0.011091,"raw_usage":{"total_tokens":4916,"prompt_tokens":744,"num_sources_used":0,"completion_tokens":73,"cost_in_usd_ticks":110912000,"prompt_tokens_details":{"text_tokens":744,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":4099,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":744,"tokens_out":73,"duration_ms":45340,"temperature":1.0,"reasoning_tokens":4099,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-20T19:58:01.280957+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Compute ground-truth attributions on a new synthetic dataset that deliberately places test samples in a region of feature space distant from the training distribution and measure whether AGOP-IxG's Spearman correlation drops below that of plain InputXGradient.","supporting_citations":[],"review_version":1}