{"id":"24fb8ee1-bed2-4e23-a86a-e843fc18d9d3","arxiv_id":"2605.01310","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"GraphSculptor builds efficient pre-training coresets for graph self-supervised learning using combined structural and semantic diversity metrics, achieving 99.6% performance with 10% of the data.","lead":"GraphSculptor selects a small coreset of graphs for pre-training self-supervised models by measuring both structural diversity from graph statistics and semantic diversity from text descriptions encoded by language models. This approach could make large graph learning tasks much more practical by cutting pre-training costs dramatically while keeping most of the performance.","discovery_kind":"new_method","skeptic_critique":{"model":"grok-4.3","headline":"Diversity preservation in graph stats + LM text encodings may not align with SSL pre-training loss landscape","rationale":"The reader's weakest assumption matches the point where the argument is least secured: the sufficiency of the chosen diversity signals for the SSL loss. Full-text experiments would need to show that ablating either view degrades results and that the theoretical bound is tight under the actual SSL objective; without that, the 99.6% claim rests on an unverified alignment between selection space and training dynamics.","tokens_in":1719,"tokens_out":345,"duration_ms":34693,"concrete_test":"Recompute the 10% coreset using only the structural feature vectors (no LM component) and only the semantic embeddings (no graph stats), then pre-train the same SSL model on each and measure downstream performance gap to the joint GraphSculptor coreset; if either single-view version closes >80% of the gap to full data, the joint metric is not load-bearing.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that cluster-aware selection on intrinsic graph statistics plus LM-encoded graph-to-text descriptions yields a coreset whose pre-training loss is close to the full set (supported by the derived bound). This holds only if the chosen metric space correlates with the SSL objective (contrastive/generative losses on augmentations). If the statistics and text summaries miss local motifs or augmentation-sensitive substructures that dominate the loss, the bound does not guarantee downstream fidelity even if diversity is maximized. The 50% uniform subsample result shows redundancy exists, but does not confirm the specific features used are the right ones for 10% selection.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper claims to introduce GraphSculptor, a label-free method for pre-training coreset construction in graph self-supervised learning. It quantifies structural diversity using intrinsic graph statistics and semantic diversity using pre-trained language models on graph-to-text descriptions, then performs cluster-aware selection in a unified metric space. A theoretical bound on the loss gap is derived, and experiments show that a 10% coreset retains 99.6% of full-data downstream performance while cutting pre-training time by nearly 90%.","tokens_in":1836,"tokens_out":579,"duration_ms":42116,"significance":"If the results hold, this work is significant for providing a scalable solution to the high computational costs of graph SSL by exploiting redundancy in unlabeled data. The combination of empirical performance gains and a theoretical loss-gap bound offers both practical utility and theoretical insight into data-efficient pre-training.","major_comments":[{"comment":"§4 (Theoretical Analysis): the derived bound on the loss gap between coreset and full-data pre-training relies on the assumption that the unified structural-semantic metric space correlates with the SSL objective (contrastive/generative losses on augmentations); no analysis or empirical correlation is shown between the chosen graph statistics/LM encodings and augmentation-sensitive substructures that dominate the loss, which is load-bearing for the bound to support the downstream fidelity claim.","section":"§4 (Theoretical Analysis)"},{"comment":"§5 (Experiments): the 10% coreset achieving 99.6% performance is the central empirical claim, but the setup reports only aggregate results without ablations isolating structural vs. semantic contributions or direct comparison to 10% uniform random sampling (beyond the 50% uniform result of 96%); this leaves open whether the cluster-aware selection on the specific features is necessary or if the bound holds only for the chosen metric.","section":"§5 (Experiments)"},{"comment":"§3.3 (Cluster-aware Selection): the integration into a unified metric space for preserving joint diversity is presented as sufficient for maintaining pre-training effectiveness, but without validation that these features capture motifs relevant to the SSL loss landscape (as opposed to general diversity), the selection may not guarantee the observed performance even if the bound is mathematically correct.","section":"§3.3 (Cluster-aware Selection)"}],"minor_comments":[{"comment":"Abstract: the time reduction is stated as 'nearly 90%'; report the precise measured reduction (with standard deviation if applicable) in the main experimental table for reproducibility.","section":"Abstract"},{"comment":"Notation throughout: ensure the structural feature vector and semantic embedding are denoted consistently when combined into the unified metric (e.g., avoid switching between f_s and e_sem without explicit definition).","section":"Notation"}],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the detailed and constructive review. The comments highlight important aspects of the theoretical motivation, experimental validation, and feature relevance that we will address in the revision to strengthen the manuscript.","responses":[{"response":"We agree that the bound in §4 is derived under the assumption that the unified metric serves as a proxy for the SSL objective. The mathematical derivation holds under this premise to motivate the selection. To strengthen the connection to empirical results, we will add an empirical analysis in the revised manuscript that computes correlations between the structural-semantic features and the actual contrastive/generative losses on augmentations for selected graphs versus the full set. This will provide direct support for the assumption without changing the bound itself.","revision_made":"yes","referee_comment":"[§4 (Theoretical Analysis)] the derived bound on the loss gap between coreset and full-data pre-training relies on the assumption that the unified structural-semantic metric space correlates with the SSL objective (contrastive/generative losses on augmentations); no analysis or empirical correlation is shown between the chosen graph statistics/LM encodings and augmentation-sensitive substructures that dominate the loss, which is load-bearing for the bound to support the downstream fidelity claim."},{"response":"We thank the referee for this suggestion. The 50% uniform result was included to illustrate data redundancy, but we acknowledge that a 10% random baseline and component-wise ablations are needed for completeness. In the revised version, we will add (i) direct comparison of GraphSculptor at 10% against 10% uniform random sampling across all datasets and (ii) ablations that separately evaluate structural-only, semantic-only, and combined selection. These will clarify the contribution of the unified metric and cluster-aware approach.","revision_made":"yes","referee_comment":"[§5 (Experiments)] the 10% coreset achieving 99.6% performance is the central empirical claim, but the setup reports only aggregate results without ablations isolating structural vs. semantic contributions or direct comparison to 10% uniform random sampling (beyond the 50% uniform result of 96%); this leaves open whether the cluster-aware selection on the specific features is necessary or if the bound holds only for the chosen metric."},{"response":"The structural statistics (e.g., degree, clustering) and LM-encoded semantics were selected because they capture properties preserved under typical graph augmentations used in SSL. We recognize that explicit validation linking them to loss-relevant motifs would be beneficial. We will revise §3.3 to include a discussion of this alignment and add a targeted analysis (e.g., motif preservation statistics on selected coresets) to demonstrate relevance to the SSL objective, thereby supporting the effectiveness of the selection.","revision_made":"partial","referee_comment":"[§3.3 (Cluster-aware Selection)] the integration into a unified metric space for preserving joint diversity is presented as sufficient for maintaining pre-training effectiveness, but without validation that these features capture motifs relevant to the SSL loss landscape (as opposed to general diversity), the selection may not guarantee the observed performance even if the bound is mathematically correct."}],"tokens_in":1505,"tokens_out":674,"duration_ms":28799,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"The main thing to know is that their 10% coreset keeps 99.6% of full-data downstream performance after pretraining and cuts time by nearly 90% on the datasets they tested. They start from the observation that even a random 50% subsample already retains over 96% performance, which shows real redundancy in the unlabeled graphs.","headline":"GraphSculptor cuts graph SSL pretraining to a 10% coreset via graph stats plus LM text encodings and claims near-full performance, but the features may not track the actual loss landscape.","tokens_in":2324,"tokens_out":158,"would_cite":false,"duration_ms":30847,"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 10% coreset of graphs retains 99.6% of full pre-training performance for graph self-supervised learning while cutting training time by nearly 90%.","keywords":["graph self-supervised learning","coreset selection","data efficiency","pre-training","structural diversity","semantic diversity","cluster selection"],"falsifier":"Selecting a 10% coreset with GraphSculptor on a fresh large graph dataset and measuring downstream performance below 95% of the full-data baseline after identical pre-training would falsify the effectiveness claim.","tokens_in":2628,"feed_emoji":"📉","tokens_out":535,"duration_ms":23195,"temperature":0.7,"pith_summary":"Large unlabeled graph datasets contain substantial redundancy, as shown by the fact that uniform subsampling of half the graphs still keeps over 96% of downstream performance. GraphSculptor exploits this by constructing a coreset that measures structural diversity from intrinsic graph statistics and semantic diversity from language-model encodings of graph-to-text descriptions. These two signals are combined in one metric space, after which cluster-aware selection keeps the joint diversity intact. A theoretical bound on the pre-training loss gap between coreset and full data supports the approach. If the claim holds, practitioners can run graph self-supervised pre-training on far smaller subsets with almost no drop in later task accuracy.","feed_headline":"10% coreset keeps 99.6% graph pre-training performance","feed_subtitle":"GraphSculptor blends graph statistics with language-model text encodings to cut pre-training time nearly 90% with minimal accuracy loss.","key_machinery":"GraphSculptor, a label-free coreset builder that fuses structural feature vectors from graph statistics with semantic encodings from language models on graph-to-text descriptions, then applies cluster-aware selection in the combined space.","core_discovery":"GraphSculptor constructs pre-training coresets by quantifying structural diversity with intrinsic graph statistics to form feature vectors and semantic diversity by encoding graph-to-text descriptions with a pre-trained language model. These signals are fused into a unified metric space where cluster-aware selection preserves joint structural-semantic diversity. The method supplies a label-free solution and derives a theoretical bound on the loss gap to full-data pre-training, with experiments showing that a 10% coreset reaches 99.6% of full performance and reduces pre-training time by nearly 90%.","pith_inferences":[],"forward_implications":[],"fun_headline_variants":["GraphSculptor crafts 10% coreset for 99.6% graph pre-training performance","Structural and semantic fusion yields 10% coreset with 99.6% retention","GraphSculptor cuts graph pre-training time nearly 90% at 10% coreset size","Graph stats and language model encodings enable 10% coresets for graph pre-training"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"That preserving joint structural-semantic diversity via cluster-aware selection on graph statistics and language-model graph-to-text encodings is sufficient to maintain pre-training effectiveness for downstream tasks without labels.","fun_headline_variants_meta":{"raw":{"variants":["GraphSculptor crafts 10% coreset for 99.6% graph pre-training performance","Structural and semantic fusion yields 10% coreset with 99.6% retention","GraphSculptor cuts graph pre-training time nearly 90% at 10% coreset size","Graph stats and language model encodings enable 10% coresets for graph pre-training"]},"model":"grok-4.3","cost_usd":0.01442,"raw_usage":{"total_tokens":6158,"prompt_tokens":726,"num_sources_used":0,"completion_tokens":96,"cost_in_usd_ticks":144203000,"prompt_tokens_details":{"text_tokens":726,"audio_tokens":0,"image_tokens":0,"cached_tokens":64},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":5336,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":726,"tokens_out":96,"duration_ms":75668,"temperature":1.0,"reasoning_tokens":5336,"cache_read_input_tokens":64,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-05-09T14:32:18.522349+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"Selecting a 10% coreset with GraphSculptor on a fresh large graph dataset and measuring downstream performance below 95% of the full-data baseline after identical pre-training would falsify the effectiveness claim.","supporting_citations":[],"review_version":1}