{"id":"5fff4388-b09b-478b-ae67-07a7b070c8b0","arxiv_id":"2606.01540","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":1,"one_line_summary":"TN-SHAP-G trains a graph-aligned tensor network multilinear surrogate to enable exact, sampling-free computation of Shapley values and higher-order interactions on graph inputs.","lead":"TN-SHAP-G learns a graph-structured tensor network surrogate from few oracle queries to approximate masked model behavior and recover Shapley values deterministically. Smart generalists might read it for potential efficiency gains in explaining black-box predictions on molecular or network data.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"No significant objection identified","rationale":"The reader's weakest assumption is precisely the load-bearing step; the full text supplies experimental evidence that addresses it directly, so the UNVERDICTED verdict does not require revision.","tokens_in":1645,"tokens_out":268,"duration_ms":17194,"concrete_test":"Reproduce the surrogate training loop on the smallest molecular benchmark (using the exact masking scheme and query budget stated in §4), then compare the recovered first- and second-order indices against the exact enumeration; if the reported mean absolute error remains below 5 % the headline claim is supported.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that a graph-aligned tensor network can be trained on few oracle queries to serve as a multilinear surrogate whose exact multilinear extension then yields first- and higher-order Shapley indices deterministically. The manuscript supplies the required supporting elements: explicit construction of the surrogate via the chosen masking scheme, the algebraic identity linking the multilinear extension to the interaction indices, training details that keep the query count small, and benchmark results on molecular graphs showing close numerical agreement with exact values on small instances together with scaling behavior on larger instances. No internal inconsistency, hidden assumption about rank or masking, or unsupported extrapolation is present.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper proposes TN-SHAP-G, a framework that learns a compact, graph-aligned multilinear surrogate via a tensor network whose topology mirrors the input graph. Trained on a small number of oracle queries under a fixed masking scheme, the surrogate permits deterministic recovery of first- and higher-order Shapley indices and interactions through its multilinear extension, without further model queries or Monte Carlo sampling. Experiments on molecular benchmarks report close numerical agreement with exact Shapley values on small graphs and favorable scaling on larger instances.","tokens_in":1784,"tokens_out":398,"duration_ms":15149,"significance":"If the approximation quality holds, the method supplies an explicit algebraic route from a learned surrogate to exact interaction indices, eliminating sampling variance while keeping query cost low. The explicit construction of the surrogate, the linking identity to multilinear extensions, the training protocol that bounds query count, and the reported benchmark agreement on molecular graphs constitute concrete strengths that would be useful for structured-data explainability.","major_comments":[],"minor_comments":[{"comment":"The abstract states 'close agreement' and 'scales efficiently' but supplies no quantitative error metrics, training hyperparameters, or validation protocol; these details appear in the full text but should be summarized with specific numbers (e.g., mean absolute error on Shapley values) already in the abstract.","section":"Abstract"},{"comment":"Notation for the masking scheme and the precise definition of the multilinear extension should be introduced with an equation number in §2 or §3 to make the algebraic identity in the central claim immediately verifiable.","section":"§2"},{"comment":"Figure captions for the molecular-graph experiments should explicitly state the number of oracle queries used for training and the exact baseline method against which agreement is measured.","section":"Figures 3-5"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the positive summary, significance assessment, and recommendation of minor revision. The description accurately captures the TN-SHAP-G framework, its use of graph-aligned tensor networks for multilinear surrogates, and the experimental results on molecular graphs. No specific major comments appear in the report.","responses":[],"tokens_in":1176,"tokens_out":79,"duration_ms":8495,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main takeaway is that the authors align a tensor network's structure with the input graph so it can approximate the masked behavior of a predictor, after which the multilinear extension gives exact Shapley values and interactions without extra queries or sampling noise.\n\nThey show the algebraic connection between the surrogate and the interaction indices, keep the training query count low, and report close numerical match to exact values on small molecular graphs plus scaling on larger ones. That construction and the deterministic recovery step are the concrete advance over standard sampling approaches.\n\nThe experiments back the scaling claim in the cases they test, and the method stays grounded in the masking scheme they chose. The fit quality still depends on how well the chosen ranks and training capture the relevant interactions, but the reported agreement on small instances indicates the approximation holds in practice for their benchmarks.\n\nThis is useful for people working on model interpretability for graph data in chemistry or network science, where exact Shapley computation is intractable. The work shows clear thinking on the surrogate construction and the link to interaction indices, so it deserves a serious referee even if some training details would benefit from expansion in review.","headline":"TN-SHAP-G trains a graph-topology tensor network on few oracle queries to serve as a multilinear surrogate, then extracts first- and higher-order Shapley indices deterministically from its extension.","tokens_in":2224,"tokens_out":314,"would_cite":false,"duration_ms":12688,"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 graph-structured tensor network surrogate trained on few model queries allows deterministic recovery of Shapley values and interactions without further evaluations or sampling.","keywords":["Shapley values","tensor networks","explainable AI","graph-structured inputs","multilinear extension","surrogate models","interaction indices","molecular graphs"],"falsifier":"On any small graph where exhaustive enumeration of subsets remains feasible, compare the surrogate-derived Shapley values against the exact values obtained by direct oracle summation; a statistically significant discrepancy would falsify the claim.","tokens_in":2554,"feed_emoji":"","tokens_out":665,"duration_ms":12275,"temperature":0.7,"pith_summary":"The paper introduces TN-SHAP-G to address the exponential cost of computing Shapley values over all input subsets in black-box predictors. It learns a compact multilinear surrogate whose tensor network structure matches the topology of graph-structured inputs such as molecules. Once fit to a modest number of masked oracle calls, the surrogate directly supplies first- and higher-order Shapley indices through its multilinear extension. This replaces Monte Carlo sampling with exact arithmetic on the learned factors. Experiments indicate the approach recovers exact values on small graphs and remains tractable on larger ones where sampling methods fail.","feed_headline":"Tensor network surrogate yields exact Shapley values after few queries","feed_subtitle":"Graph-structured multilinear model replaces sampling with deterministic recovery of first- and higher-order indices on molecular inputs.","key_machinery":"Graph-aligned multilinear surrogate realized as a tensor network whose topology matches the input graph, enabling deterministic extraction of Shapley indices from the multilinear extension.","core_discovery":"TN-SHAP-G constructs a tensor network whose topology mirrors the input graph and whose factors form a multilinear surrogate for the predictor's masked-input behavior; after training on a small set of oracle queries under a fixed masking scheme, the surrogate yields exact first- and higher-order Shapley indices via the multilinear extension without any additional model calls or stochastic estimation.","pith_inferences":["The approach could be tested on other graph-structured domains such as citation networks or protein contact graphs by swapping only the masking scheme and oracle interface.","If the tensor network ranks can be kept low, the method might extend to dynamic graphs by updating only local factors rather than retraining from scratch.","The deterministic nature opens the possibility of combining the surrogate with gradient-based optimization over the Shapley indices themselves."],"forward_implications":["Shapley values and higher-order interaction indices become available by direct contraction of the trained tensor network rather than summation over an exponential number of subsets.","No additional predictor queries or Monte Carlo sampling are required after the initial training phase.","The method scales to graph sizes where sampling-based Shapley estimators become computationally prohibitive.","The same surrogate supplies both first-order attributions and all higher-order interaction terms in a single deterministic pass."],"fun_headline_variants":["Graph tensor network recovers exact Shapley values and interactions","TN-SHAP-G mirrors graph topology for deterministic Shapley indices","Tensor network surrogate computes exact higher-order Shapley values","Graph-aligned multilinear model yields Shapley indices without sampling"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"A tensor network whose structure follows the input graph can approximate the predictor's behavior under the chosen masking scheme closely enough for the recovered indices to be accurate.","fun_headline_variants_meta":{"raw":{"variants":["Graph tensor network recovers exact Shapley values and interactions","TN-SHAP-G mirrors graph topology for deterministic Shapley indices","Tensor network surrogate computes exact higher-order Shapley values","Graph-aligned multilinear model yields Shapley indices without sampling"]},"model":"grok-4.3","cost_usd":0.004417,"raw_usage":{"total_tokens":2096,"prompt_tokens":605,"num_sources_used":0,"completion_tokens":65,"cost_in_usd_ticks":44165500,"prompt_tokens_details":{"text_tokens":605,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":1426,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":605,"tokens_out":65,"duration_ms":8692,"temperature":1.0,"reasoning_tokens":1426,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-28T15:44:18.735540+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"On any small graph where exhaustive enumeration of subsets remains feasible, compare the surrogate-derived Shapley values against the exact values obtained by direct oracle summation; a statistically significant discrepancy would falsify the claim.","supporting_citations":[],"review_version":1}