{"id":"136c7a02-dcdf-4f2c-a3b5-510dca78caee","arxiv_id":"2601.21309","paper_version":4,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":5,"one_line_summary":"A graph-condensation framework that aligns condensed graphs with low-frequency 'causal' structure via spectral contrastive learning transfers better across tasks and datasets than existing condensation methods, with up to 13.41% relative gains in cross-task benchmarks.","lead":"Huge graph datasets are expensive to train on, so this paper compresses them into tiny synthetic graphs that keep the 'causal' (stable, low-frequency) structure, letting models trained on the tiny graphs still work on new tasks and new datasets. The method reports accuracy gains up to 13.41% over prior graph-condensation methods across six benchmarks, including a new financial dataset.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Low-frequency = causal assumption is unvalidated; if inverted-spectral ablation works equally, TGCC's causal mechanism claim collapses.","rationale":"The reader's weakest_assumption identifies exactly the spectral-causality identification as the load-bearing premise. My stress-test confirms this by tracing the logic: the paper's novelty is 'causal-invariance-based' graph condensation, and the entire theoretical framing depends on low-frequency = causal. Without this identification, the method reduces to a combination of spectral regularization and contrastive learning, which could still work empirically but would not justify the causal claim. The proposed test—an inverted-spectral ablation—directly aims at this premise. If the inverted variant matches TGCC, the 'causal' explanation is vacuous; if it fails, the premise is supported on the tested datasets. Either outcome would settle the concern. The verdict remains CONDITIONAL because this test is not run in the paper, and the missing ST-GCond baseline and error bars also matter, but the spectral assumption is the deeper theoretical gap. I agree with the reader's assessment, so no change to the verdict is needed.","tokens_in":14549,"tokens_out":9725,"duration_ms":104243,"concrete_test":"Run TGCC with the spectral selection inverted: treat high-frequency components as causal (preserve them) and low-frequency as non-causal (perturb them). Keep the rest of the pipeline identical. Evaluate on the same cross-task/cross-dataset protocols (Tables 2-4). If the inverted variant achieves comparable or better transfer accuracy, the low-frequency=causal identification is not the source of TGCC's performance and the causal mechanism claim is unsupported. If the inverted variant degrades significantly, the assumption gains empirical support on these benchmarks.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central mechanism rests on an untested identification: 'we consider the low-frequency components in the graph as causal content, while the high-frequency components are treated as non-causal content' (Section 'Causal Invariant Feature Extraction'). Every subsequent step—intervention V (Eq 1), invariance objective (Eq 5), negative sample construction (Eq 13), and InfoNCE (Eq 14)—is a mechanism for preserving low-frequency structure, not for discovering or validating causal structure. The only support cited (Liu et al. 2022; Mo et al. 2024) shows that spectral augmentations correlate with invariance in graph contrastive learning, not that low-frequency content is causal for arbitrary downstream tasks or domains. Theorem 1 bounds L_causal in terms of eigenvalue differences but presupposes the spectral labeling; it does not test whether low-frequency content transfers causal knowledge. If this proxy fails for a downstream task/domain, the condensed graph may simply retain generic low-frequency structure, and the observed gains (e.g., 13.41% on Reddit) would be attributable to spectral smoothing or contrastive regularization rather than causal invariance. The paper's own exploratory experiments (Appendix F) are not available in the provided text, and even if they show GCL invariance is low-frequency, they do not establish causality. Thus the assumption is load-bearing and unvalidated.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes TGCC, a graph condensation method designed to produce condensed graphs that transfer across tasks and datasets. The method assumes that low-frequency spectral components of a graph constitute causal, invariant content while high-frequency components are non-causal. It constructs an augmented view by perturbing high-frequency structure (Eq. 1-2), enforces dimension-level invariance and independence objectives (Eq. 5-9), performs gradient-matching condensation on the original and augmented graphs (Eq. 12), and injects causal-invariant information via an InfoNCE loss with negative samples constructed by perturbing low-frequency components (Eq. 13-14). Empirical results on five public datasets and the new FinReport dataset are reported for cross-task link prediction (Table 2), cross-dataset node classification (Table 3), and combined cross-task/cross-dataset link prediction (Table 4), together with ablations, generalizability, and efficiency comparisons.","tokens_in":14793,"tokens_out":3732,"duration_ms":41888,"significance":"If the empirical results hold, TGCC is a practically useful contribution: transferable graph condensation would let users train once on a condensed graph and apply it to multiple downstream tasks and datasets, avoiding repeated condensation. The paper ships code and a new FinReport dataset, and Table 2 shows consistent, standard-deviation-reported gains of up to 13.4% over strong baselines in the cross-task setting. However, the central causal claim is not established: the identification of 'low-frequency = causal' is imported from prior work and is load-bearing for the method's stated mechanism. The missing comparison to ST-GCond, the closest transferable baseline, and the absence of error bars in Tables 3-4 also weaken the evidence for state-of-the-art transfer performance. The contribution is therefore potentially significant, but the causal framing and some experimental omissions require attention before the claims can be accepted.","major_comments":[{"comment":"The premise that low-frequency graph components are causal content and high-frequency components are non-causal content is stated as fact but is an unvalidated assumption inherited from Liu et al. 2022 and Mo et al. 2024. All subsequent steps—the intervention V (Eq. 1), the invariance objective (Eq. 5), the negative-sample construction (Eq. 13), and the InfoNCE injection (Eq. 14)—are mechanisms for preserving low-frequency structure, not for discovering or validating causal structure. The paper cites its own 'exploratory experiments (Appendix F)' as confirmation, but that appendix is absent; even GCL invariance at low frequencies would not establish causality for arbitrary downstream tasks. Because the abstract and title claim a causal mechanism, this identification is load-bearing. Please either provide direct evidence for the spectral proxy (e.g., an inverted-spectral ablation where hi","section":"Causal Invariant Feature Extraction (Eq. 1-9)"},{"comment":"ST-GCond is identified in Related Work as 'the method most similar to ours' and 'costly,' but it is never included as a baseline in any experiment. Since ST-GCond is also a transferable graph condensation method, omitting it makes the claim of state-of-the-art cross-task/cross-dataset performance unsubstantiated. The manuscript should add ST-GCond to Tables 2-4 (and the ablation table if feasible) and discuss the comparison.","section":"Experiments (Tables 2-6)"},{"comment":"Tables 3 and 4 report only point estimates with no standard deviations or confidence intervals, unlike Table 2. Several TGCC improvements over baselines are 1-2 percentage points (e.g., Table 3: 45.3 vs 44.9 on Flickr at 0.05%; Table 4: 60.5 vs 55.6 AUC on Cora at 0.1%). Without variance or significance testing, these gains cannot be distinguished from noise. Please report standard deviations for all reported transfer results and, where possible, statistical tests across seeds.","section":"Tables 3 and 4"},{"comment":"Theorem 1 is the paper's theoretical justification for causal invariance, but the proof is deferred to 'Appendix A' and is not included in the manuscript. Moreover, the statement mixes notation: λ_i and γ_i are described as amplitudes of the i-th frequency of A and V, while the bound uses λ_A_j and γ_V_j without definition; θ_i is introduced but not specified. A reader cannot verify the theorem or its role in the method. Include the complete proof and state the assumptions precisely, or remove the theoretical claim and rely on the empirical evaluation.","section":"Theoretical Analysis (Theorem 1)"}],"minor_comments":[{"comment":"The notation for Θ_L, Θ_f, Θ_l is undefined or ambiguous; 'Θ is a parameter updated during training' does not clarify what Θ_L is in the matching term. Please define all symbols and the optimization constraints.","section":"Eq. 2"},{"comment":"The expression for the negative-sample Laplacian is garbled (e.g., 'λ(1−κ)∗N/2u(1−κ)∗N/2u⊤(1−κ)∗N/2'). It should be written as a proper sum over eigenvalues with clear index ranges.","section":"Eq. 13"},{"comment":"The text says 'The si represents the standard deviation of the i-th dimension before normalization' but the equation uses sA_i and sV_i without definition. Also, the equivalence between the normalization of Z and the inner-product form should be stated more carefully.","section":"Eq. 9 / text"},{"comment":"The ablation table reports node-classification accuracy on Flickr and Citeseer. However, Table 2 reports link prediction accuracy for the same datasets, and the absolute numbers differ substantially (e.g., Flickr 50.2 vs 70.6). Please clarify whether the settings differ (task, split, condensation ratio) and, if so, state this explicitly.","section":"Table 6"},{"comment":"The quantitative claims '3.2% lower than the ground truth' and 'average performance loss of 9.8%' are not tied to any table or experimental description in the text. Please provide the corresponding setup and results.","section":"Introduction (Challenges 1-2)"},{"comment":"Several items are deferred to appendices that are not included in the manuscript text: Appendix A (proof of Theorem 1), Appendix F (exploratory experiments), Appendix G (additional results). Since the main text refers to these as supporting evidence, they should be made available for verification.","section":"Appendices"}],"recommendation":"major_revision","confidential_remarks":"The core empirical finding—consistent gains in cross-task link prediction with standard deviations—is credible and worth pursuing. The major concern is the gap between the causal narrative and the actual mechanism: the method preserves low-frequency structure, and the causal interpretation is an untested assumption. This is fixable by either validating the spectral proxy or reframing the claims. Also, the omission of ST-GCond and the absence of error bars in Tables 3-4 are significant for a transferable-condensation paper. I recommend major revision rather than rejection, as the empirical method appears sound after addressing these issues."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a plausible empirical contribution in a crowded area, and the causal framing is mostly borrowed and untested. The one thing I'd want before trusting the headline numbers is a direct comparison to ST-GCond, which the paper itself names as the closest method and then never runs.\n\nWhat's actually new: the specific pipeline — GCIL's dimension-level invariance objective (Eq 5-9), GCond-style gradient matching on both the original and an augmented view, and spectral InfoNCE with inverted-spectrum negatives (Eq 13-14) — hasn't been put together before, and the FinReport financial dataset is a useful addition. The cross-task link prediction results in Table 2 are reported with standard deviations and show consistent gains, up to 13.4% relative over GCond on Reddit. The ablations in Table 6 suggest all three modules contribute. That's real evidence, not a toy setup.\n\nWhere it gets soft. The central claim is that TGCC preserves 'causal invariant features,' but 'causal' is defined as low-frequency spectral content, a label inherited from Mo et al. 2024 and Liu et al. 2022. The paper never tests whether low-frequency structure is actually causal for the downstream tasks. Theorem 1 is a bound that presupposes that identification. If the spectral proxy is wrong, the method still might work — the gains could come from spectral smoothing and contrastive regularization — but the stated mechanism collapses. That's a major framing issue, not a fatal one, because the empirical results are external and non-circular.\n\nMore addressable problems: ST-GCond is never included in Tables 2-4, which is hard to justify for a transferability claim. Tables 3 and 4 have no error bars, so 1-2 point improvements over baselines are unquantified. The '13.41%' headline is a relative improvement (8.7 absolute points on Reddit); the abstract doesn't say that. And the FinReport dataset has no URL and the code no commit hash, so the resource isn't actually public yet.\n\nWho this is for: people working on graph condensation or graph transfer learning. It deserves a serious referee — the problem matters and the evidence is substantive — but it needs a revision that adds ST-GCond, error bars, and a more honest framing of the causal assumption. If the authors can do that, it's a decent paper.","headline":"Real empirical gains in transferable graph condensation, but the causal theory is borrowed and untested; missing ST-GCond comparison and error bars keep it from being a clean accept.","tokens_in":15423,"tokens_out":2585,"would_cite":false,"duration_ms":27366,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"TGCC claims that preserving low-frequency (causal-invariant) graph structure during condensation yields transferable condensed graphs, reporting up to 13.41% gains over prior methods.","keywords":["graph dataset condensation","causal invariance","graph contrastive learning","spectral graph theory","transfer learning","cross-task adaptation","graph neural networks","domain-invariant features"],"falsifier":"Run TGCC on a heterophilic graph whose labels correlate with high-frequency structure (or construct such a graph by relabeling a homophilic dataset), and compare against a variant that preserves high-frequency content as the 'causal' signal. If the high-frequency variant matches or outperforms TGCC on cross-task accuracy, the low-frequency causal assumption is refuted and the spectral proxy does not carry the transfer benefit.","tokens_in":14311,"feed_emoji":"🕸️","tokens_out":7949,"duration_ms":69941,"temperature":0.7,"pith_summary":"TGCC argues that graph condensation can be made transferable if the condensed graph keeps the causal-invariant content of the original graph, which the paper identifies with low-frequency spectral structure. The method combines causal intervention on high-frequency structure, gradient-matching condensation on original and augmented graphs, and spectral-domain contrastive learning that injects low-frequency invariants into the condensed graph. Experiments report up to 13.41% accuracy gains over prior methods in cross-task and cross-domain settings, and state-of-the-art results on 5 of 6 datasets when task and dataset match. If correct, this makes condensed graph datasets reusable assets rather than single-use approximations, and opens a causal-invariance route to transferable graph learning. The paper also releases FinReport, a financial graph dataset pairing corporate financial reports with analyst research reports.","feed_headline":"Low-frequency structure lifts transfer accuracy by 13.41%","feed_subtitle":"Keeping low-frequency causal structure lets one condensed graph train for new tasks, beating methods on 5/6.","key_machinery":"The load-bearing object is the graph Laplacian's low-frequency eigenspace, treated as a proxy for causal-invariant graph content. The intervention step generates an augmented graph by adding and deleting edges to corrupt high-frequency structure while keeping low frequencies unchanged; the invariance objective then aligns the representations of original and augmented graphs dimension-by-dimension with equalized standard deviations; the independence objective decorrelates dimensions via a covariance/HSIC surrogate; and spectral contrastive learning injects the low-frequency invariant signal into the condensed graph using negative samples built from a truncated low-frequency reconstruction of","core_discovery":"The paper's central claim is that graph condensation can be made transferable by forcing the synthetic graph to preserve 'domain causal-invariant features,' which the paper identifies with the low-frequency components of the graph spectrum. High-frequency components are treated as non-causal, task-specific noise and are deliberately perturbed during training. The method, TGCC, combines three objectives: a causal-invariant extraction module that aligns representations of the original and an intervened graph in each dimension while enforcing pairwise independence (Equations 5-9); a gradient-matching condensation step that mimics training on both the original and augmented graphs (Equation 12);","pith_inferences":["If the low-frequency-as-causal proxy generalizes, similar spectral causal regularizers could be dropped into other graph self-supervised methods and dataset distillation frameworks, not just condensation.","The paper's own Flickr result at r=1% (a relative failure) hints that confounders and latent variables can break the invariance extraction; identifying when the proxy fails is a direct next test.","A controlled experiment swapping the spectral negative-sample construction for random edge perturbation would separate the causal-preservation effect from generic contrastive regularization.","The release of FinReport, a financial-report-to-research-report matching graph, adds a domain benchmark for transferable graph learning beyond the paper's own experiments."],"forward_implications":["One condensed graph can serve multiple downstream tasks: a graph condensed for node classification transfers to link prediction, with accuracy gains up to 13.41% over prior methods (Table 2).","Condensed graphs can transfer across datasets: models trained on Ogbn-arxiv condensation and evaluated on Cora, Citeseer, Flickr, Reddit, and FinReport outperform baselines in most settings (Table 3).","The method is computationally practical: TGCC is roughly 3x faster than SFGC and 2x faster than GEOM on Ogbn-arxiv and FinReport at comparable or better accuracy (Figure 4).","Each of the three modules contributes: ablations show removing causal invariant extraction, contrastive condensation, or spectral contrastive learning degrades node classification performance (Table 6)."],"fun_headline_variants":["Low-frequency causal structure makes graph condensation transferable","Transferable graph condensation keeps causal low frequencies","Graph condensation: preserve causal spectrum for new tasks","Causal graph condensation boosts cross-domain transfer up to 13.41%","One condensed graph for many tasks: low-freq is causal"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The paper assumes that low-frequency spectral content of a graph is causal, invariant knowledge and high-frequency content is non-causal noise; if this spectral proxy is wrong for a task or domain, the causal interpretation of TGCC's gains collapses and the method's advantage reduces to generic contrastive regularization.","fun_headline_variants_meta":{"raw":{"variants":["Low-frequency causal structure makes graph condensation transferable","Transferable graph condensation keeps causal low frequencies","Graph condensation: preserve causal spectrum for new tasks","Causal graph condensation boosts cross-domain transfer up to 13.41%","One condensed graph for many tasks: low-freq is causal"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000737,"raw_usage":{"total_tokens":3126,"prompt_tokens":735,"completion_tokens":2391,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":479,"completion_tokens_details":{"reasoning_tokens":2312}},"tokens_in":479,"tokens_out":2391,"duration_ms":19588,"temperature":1.0,"reasoning_tokens":2312,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-03T07:00:56.228859+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run TGCC on a heterophilic graph whose labels correlate with high-frequency structure (or construct such a graph by relabeling a homophilic dataset), and compare against a variant that preserves high-frequency content as the 'causal' signal. If the high-frequency variant matches or outperforms TGCC on cross-task accuracy, the low-frequency causal assumption is refuted and the spectral proxy does not carry the transfer benefit.","supporting_citations":[],"review_version":1}