Fast Track to Winning Tickets: Repowering One-Shot Pruning for Graph Neural Networks

Dawei Cheng; Guibin Zhang; Haoran Yang; Yanwei Yue

REVIEW 4 major objections 6 minor 33 references

Fast Track to Winning Tickets: Repowering One-Shot Pruning for Graph Neural Networks

T0 review · 4 major / 6 minor · reviewed 2026-08-11 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read One-shot pruning plus gradient-degree denoising can match iterative IMP at higher sparsity and up to 44x lower cost.

desk verdict A genuinely new fast-track method for graph lottery tickets with broad experiments, but the headline sparsity advantage rests on an unspecified success criterion and the reporting has several inconsistencies. read the letter →

arxiv 2412.07605 v1 pith:5EXLLIE5 submitted 2024-12-10 cs.LG

Yanwei Yue , Guibin Zhang , Haoran Yang , Dawei Cheng This is my paper · ORCID

classification cs.LG

keywords graphlotteryticketone-shotpruningiterativemagnitudesparsificationGNNgradient-baseddenoisingedgedegreeFastGLT

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

The paper argues that expensive iterative magnitude pruning is not necessary to find graph lottery tickets. It shows that a single one-shot pruning run produces masks already close to IMP masks, with the residual difference following a consistent pattern: one-shot pruning wrongly removes weights with relatively large gradients and edges with relatively low degree. Based on this, FastGLT prunes once to an intermediate sparsity, then repeatedly swaps low-magnitude kept elements for high-gradient weights and low-degree edges from the pruned set. The reported result is a double win: winning subgraphs and subnetworks with higher weight and graph sparsity than IMP-based methods, found in E+D epochs instead of kE, with 1.7-44x speedups and over 95% MAC savings. If right, this reframes one-shot pruning from a cheap fallback into the main route for graph lottery ticket search.

What carries the argument

The load-bearing mechanism is a two-stage mask pipeline. Stage one trains trainable masks $m_g$ and $m_\theta$ end-to-end, then thresholds the lowest-magnitude entries at an intermediate sparsity computed by $\Psi(s)=s-\alpha s^\beta$ to produce one-shot masks $M^\circledcirc_g$ and $M^\circledcirc_\theta$. Stage two, the gradual denoising mechanism, works in $\lceil D/\Delta T\rceil$ intervals: at each interval it labels currently kept elements with the smallest magnitudes as noisy, labels pruned weights with the largest accumulated gradients and pruned edges with the smallest edge degrees as potential, and swaps them via $M^{(\mu+1)}=(M^{(\mu)}\setminus M^{(ns)})\cup M^{(pt)}$, increasing sparsity toward the target by a fixed amount per interval. This swap is what turns a cheap one-shot ticket into a winning ticket.

What would settle it

Take a graph from a different family, such as a heterophilic graph or a protein interaction graph, run FastGLT and an iterative magnitude pruning baseline to the same target sparsity, and compare the identities of restored elements. If the pruned weights that IMP keeps do not have consistently higher accumulated gradients, or the pruned edges IMP keeps do not have consistently lower edge degree, then the denoising signal is not the mechanism and the fast-track claim fails on that graph family.

Watch

Extended reading notes

Core claim

The central discovery is that the gap between one-shot and IMP masks is small and structured rather than random. Across sparsity levels, the Hamming distance between one-shot masks and IMP masks stays far smaller than the distance between random masks and IMP masks, and the mistakes one-shot pruning makes are biased in a specific way: compared with IMP, one-shot pruning cuts weights that have higher gradients and edges that have lower degree. The paper treats these as identifiable structural noise that can be corrected. FastGLT therefore uses one-shot masks as a fast track and denoises them: within each denoising interval it identifies currently kept elements with the smallest magnitudes as noisy, discovers pruned weights with the largest accumulated gradients and pruned edges with the smallest edge degrees as potential, and swaps them, gradually raising sparsity to the target. The paper validates this by finding winning tickets at higher extreme sparsity than UGS and WD-GLT across 4 backbones and 6 datasets, with weight sparsity gains of 1.32%-45.62%, graph sparsity gains of 7.49%-22.71%, and 1.7-44x search speedups.

Load-bearing premise

The whole method depends on the assumption that one-shot pruning makes the same kind of mistakes on every graph as it does on Cora: it wrongly removes weights that have large gradients and edges whose endpoints have few connections.

Editorial extensions

If this is right

If the fast-track pattern generalizes, graph lottery ticket search no longer needs k rounds of prune-and-retrain; one-shot pruning plus denoising is enough.
FastGLT produces winning tickets with higher extreme sparsity than UGS and WD-GLT: up to 45.62% higher weight sparsity and 22.71% higher graph sparsity on small graphs, and weight sparsity above 70% on large OGB graphs.
The reported efficiency gain is 1.7-44x over IMP-based GLT search with 95.3%-98.6% MAC savings, so sparse GNN training and inference become substantially cheaper.
Because denoising from random tickets loses up to roughly 29% weight sparsity and 24% graph sparsity, the one-shot start is load-bearing: the fast track itself, not just the swapping mechanism, is what makes the method work.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

The same recipe, prune once and then restore elements selected by a measured mismatch signal, could transfer to other structured pruning settings wherever one-shot and iterative masks differ in a measurable way.
A natural next test is whether the gradient and degree signals identify IMP's choices on graphs with very different degree distributions, such as heterophilic graphs or protein interaction graphs; if not, the potential-component metric would need to become dataset-adaptive.
An implicit consequence is that the method's advantage grows with the number of IMP rounds it replaces, so FastGLT is most attractive at very high target sparsity where iterative magnitude pruning becomes prohibitively expensive.
If the pattern is universal, one-shot prune-and-denoise could replace iterative magnitude pruning in other lottery ticket settings, making lottery ticket search feasible on graphs where IMP's cost is prohibitive.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

The paper is worth a look. It makes a real empirical case that one-shot joint graph/weight pruning plus a short denoising phase can find winning tickets nearly as well as IMP, at a fraction of the wall-clock cost. The specific mix—one-shot pruning to an intermediate sparsity, then iterative gradient/degree-based swapping of a small set of pruned elements—is new relative to the GLT literature, and the experiments are broad: 6 datasets, 4 backbones, ablations, and a comparison to UGS and WD-GLT. The ablations showing that starting from one-shot tickets matters (vs. random tickets) support the 'fast track' hypothesis.

The soft spots are real but mostly fixable. The biggest one is the operational definition of 'extreme sparsity.' The paper says it is the maximum sparsity at which a GLT is found, where a GLT requires test accuracy at least the dense baseline, but there is no tolerance and no grid step. The appendix just says an arithmetic sequence is searched. Since the headline claim—higher weight and graph sparsity than IMP-based methods—is computed from these extreme sparsity values, a different reasonable tolerance could shrink or erase the reported gap. This needs to be pinned down before the numbers can be trusted.

There are also internal inconsistencies: the abstract and contributions report different improvement ranges; Table 1 lists UGS on GIN as 28.6× relative time, which does not match the raw times in the same table (average about 9.5×); Algorithm 1 uses the denoising scheduler Υ to compute intermediate sparsity where the text uses Ψ; and Eq. (6) defines ω_θ with graph sparsity instead of weight sparsity. These are minor typos except the Table 1 one, which looks like a reporting error.

One more thing: the gradient/degree motivation is only visualized for Cora. The OGB results suggest the pattern generalizes, but a second dataset in that figure would be more convincing. Some hyperparameters (τ, α, β) are never given explicit defaults.

Who is this for? Researchers working on efficient GNN training or pruning. The method is clearly described and the code is promised. I would send it to peer review, but the authors need to specify the extreme sparsity protocol exactly, correct the Table 1 numbers, and fix the notation errors. I'd cite the method once these are cleaned up.

Referee Report

4 major / 6 minor

Summary. The paper proposes FastGLT, a one-shot pruning and denoising framework for finding graph lottery tickets. Instead of iterative magnitude pruning, FastGLT first prunes graph and weight masks to an intermediate sparsity in one shot, then gradually swaps low-magnitude retained elements for high-gradient weights and low-degree edges, progressively increasing sparsity to the target. Experiments on six datasets and four GNN backbones compare FastGLT with UGS, WD-GLT, and random pruning, and the authors claim higher weight and graph sparsity, 1.7-44x speedups over IMP-based methods, and 95%+ MAC savings. The central conclusion is that one-shot tickets can be denoised into tickets that are sparser than, and perform comparably to, IMP-based tickets.

Significance. The fast-track hypothesis is interesting and potentially useful: if one-shot tickets are structurally close to IMP tickets, replacing k rounds of IMP with one-shot pruning plus denoising could make graph lottery ticket search substantially cheaper. I credit the authors for releasing code, reporting standard deviations over five trials, and including ablation studies on the initial ticket type, denoising interval, and scheduler decay. If the quantitative claims withstand scrutiny, the practical contribution is solid. However, the two quantitative pillars of the paper—higher sparsity and faster search—are not yet fully substantiated as reported: the extreme-sparsity criterion is not operationally defined, and the wall-clock normalization in Table 1 is internally inconsistent. These issues are fixable and do not by themselves invalidate the method, but they must be resolved before the headline numbers can be trusted.

major comments (4)

[§2.2, Appendix C.5, Tables 2/4/8] The extreme-sparsity measurements that support the headline 'higher sparsity' claim are not operationally defined. Definition 1 requires test accuracy at least that of the dense baseline, but because accuracy is stochastic, a tolerance (e.g., within one standard deviation, within 0.5%, or strictly ≥ the baseline mean) is needed. Appendix C.5 only says that sparsity levels are searched 'in an arithmetic sequence until a lottery ticket can no longer be found,' without specifying the step size or whether UGS and WD-GLT are evaluated under the same search protocol. Since Tables 2, 4, and 8 report extreme sparsities on the basis of this undefined criterion, the reported 1.32%-45.62% weight-sparsity and 7.49%-22.71% graph-sparsity improvements are not falsifiable from the information given; a different tolerance or grid step could materially change them.
[Table 1 and §4.4] The 'Relative Time' column is computed inconsistently, so the 1.7-44x speedup claim is not reproducible as reported. For GCN FastGLT, 1.63x is the mean of the three per-dataset ratios (34.9/21.4, 89.7/41.4, 1366.2/1217.3), while for GIN UGS the reported 28.6x is the sum of the three ratios (39.8/7.3, 61.0/8.6, 141.7/8.8), and for GAT FastGLT 4.8x is again a sum (525.9/333.1, 528.7/284.1, 1270.3/920.7). Obs.4's statement that 'UGS requires 4.0-28.6x' therefore mixes averaged and summed quantities; the table should be recomputed in a single consistent way and the speedup ranges restated accordingly.
[Abstract, §1, §4.2] The headline improvement ranges are mutually inconsistent. The abstract and Obs.1 report weight-sparsity gains of 1.32%-45.62% and graph-sparsity gains of 7.49%-22.71%, while the contribution bullet in §1 reports 5.82%-25.48% and 3.65%-17.48%. Because these ranges are the paper's main quantitative claim, the authors should reconcile them and specify exactly which table and which baseline each range refers to.
[§2.3 and §3.2] The denoising mechanism is motivated by a two-part empirical pattern observed only on Cora: relative to IMP, one-shot pruning removes weights with relatively high gradients and edges with relatively low degree, and Eqs. 5-6 instantiate exactly these signals for every dataset and backbone. The paper should report the same gradient/degree diagnostics for at least one additional small graph and one OGB graph, or explicitly state that transferability is an assumption; otherwise the mechanism's generality is a risk rather than a validated premise.

minor comments (6)

[Algorithm 1, line 6] Line 6 computes the intermediate sparsity as sinm ← Υ(stgt), but Υ is defined in §3.2 as the denoising scheduler; the intermediate sparsity should be computed with Ψ(s)=s−αs^β from §3.1.
[Eq. 6 and surrounding text] The formula for ωθ uses stgt_g in both ωg and ωθ, but the weight-sparsity increment should presumably use stgt_θ; this appears to be a typo and should be corrected.
[Table 7] In the GCN block of Table 7, FastGLT reports accuracy 79.11 for both Citeseer and PubMed; the Citeseer value duplicates the PubMed value and is implausible relative to the 70.51 Citeseer baseline. This data error should be corrected.
[§3.2 and Table 3] The denoising scheduler Υ(µ)=τ(1−µ/µend)^κ contains an initial ratio τ that is never specified in Table 3 or in the sensitivity analysis in Appendix C.3; please report the value of τ used in all experiments.
[Table 1 and Figures 1/5] The text uses 'wall-lock time' in the Figure 1 caption and 'wall-lock' in Figure 5; this should read 'wall-clock.' In addition, Table 1's header 'Relative Time (s)' is a unitless ratio and should be labeled accordingly.
[§3.1] The phrase 'task-itrelevant loss function' should read 'task-irrelevant loss function.'

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical method study whose central claims are benchmarked against external baselines, and the self-citations present are not load-bearing.

full rationale

FastGLT is an empirical pruning method, not a derivation from first principles, and its core claims do not reduce to its inputs. The denoising mechanism in Sec. 3.2 is motivated by the Cora-based observation (Sec. 2.3) that one-shot masks differ from IMP masks in gradient and degree statistics, but the paper tests this hypothesis rather than assuming it: Table 7 compares FastGLT with plain one-shot pruning, Fig. 6 and Table 5 compare denoising from one-shot versus random tickets, and the headline sparsity/speed results are measured against the external UGS and WD-GLT baselines. No equation defines the output in terms of the claimed result: Eqs. 5-8 describe a mask-swapping schedule whose target sparsity is a user input, and the 'extreme sparsity' reported in Tables 2 and 4 is an empirical search outcome, not an identity. The self-citations (e.g., Zhang et al. 2024a,b,c; Cheng et al. 2023; Wang et al. 2024) appear in the introduction and related work as examples of existing research lines and are not used to justify the method's correctness; the load-bearing prior-work citations (Wang et al. 2023d for information loss, Zhu and Gupta 2017 and Evci et al. 2020 for the scheduler, Hui et al. 2023 for performance collapse) are external. The lack of a fully specified accuracy tolerance for the GLT criterion is a reproducibility concern, not a circularity, because the reported advantage is still an empirical comparison against baselines under the same criterion.

Assumptions & free parameters 4 free parameters · 4 assumptions · 0 invented entities

The method's claims depend on several hand-set hyperparameters (α, β, κ, τ, ΔT) and on an empirical assumption that the one-shot/IMP gap is characterized by gradients and edge degrees. The paper ablates κ and ΔT but not α, β, or the gradient/degree pattern itself.

free parameters (4)

α and β in Ψ(s) = s - αs^β = α=0.01, β=1.2
Used to compute intermediate sparsity before one-shot pruning; no sensitivity analysis reported for these coefficients.
κ decay factor in denoising scheduler Υ = κ=1
Selected because κ=1 outperformed κ=2 and κ=3 in Table 6, then applied uniformly.
τ initial ratio in denoising scheduler Υ = not reported
Controls how many noisy elements are identified in the first interval; the value is not listed in Table 3 or the main text.
ΔT denoising interval = 3, 10, or 20 depending on dataset/backbone
Set per dataset/backbone in Table 3; sensitivity analysis shows modest impact, but the choices affect the running time.

assumptions (4)

domain assumption Trainable binary masks mg and mθ optimized by gradient descent yield useful one-shot masks
Sec. 3.1 co-trains the masks with the network; no formal guarantee that the resulting masks isolate important structures.
domain assumption The gradient/degree pattern observed on Cora in Sec. 2.3 holds across the six datasets and four backbones
The potential-component discovery in Eq. 6 uses accumulated gradients for weights and edge degree for edges based on one motivating plot.
ad hoc to paper The exponential decay Ψ(s)=s-αs^β prevents performance collapse at high target sparsity
Introduced in Sec. 3.1 to pre-calculate intermediate sparsity; no independent justification for the functional form.
domain assumption Swapping a small number of noisy and potential elements each interval keeps the model trainable
Sec. 3.2 assumes the mask update in Eq. 8 preserves stable training until the target sparsity is reached.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Fast Track to Winning Tickets: Repowering One-Shot Pruning for Graph Neural Networks." pith.science (2026). https://pith.science/paper/5EXLLIE5

@misc{pith2026241207605,
  author       = {Pith},
  title        = {Pith review of: Fast Track to Winning Tickets: Repowering One-Shot Pruning for Graph Neural Networks},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/5EXLLIE5}},
  note         = {Machine review of arXiv:2412.07605}
}

abstract

Graph Neural Networks (GNNs) demonstrate superior performance in various graph learning tasks, yet their wider real-world application is hindered by the computational overhead when applied to large-scale graphs. To address the issue, the Graph Lottery Hypothesis (GLT) has been proposed, advocating the identification of subgraphs and subnetworks, \textit{i.e.}, winning tickets, without compromising performance. The effectiveness of current GLT methods largely stems from the use of iterative magnitude pruning (IMP), which offers higher stability and better performance than one-shot pruning. However, identifying GLTs is highly computationally expensive, due to the iterative pruning and retraining required by IMP. In this paper, we reevaluate the correlation between one-shot pruning and IMP: while one-shot tickets are suboptimal compared to IMP, they offer a \textit{fast track} to tickets with a stronger performance. We introduce a one-shot pruning and denoising framework to validate the efficacy of the \textit{fast track}. Compared to current IMP-based GLT methods, our framework achieves a double-win situation of graph lottery tickets with \textbf{higher sparsity} and \textbf{faster speeds}. Through extensive experiments across 4 backbones and 6 datasets, our method demonstrates $1.32\% - 45.62\%$ improvement in weight sparsity and a $7.49\% - 22.71\%$ increase in graph sparsity, along with a $1.7-44 \times$ speedup over IMP-based methods and $95.3\%-98.6\%$ MAC savings.

Figures

Figures reproduced from arXiv: 2412.07605 by the authors.

**Figure 1.** (Left) Accuracy (↑) of UGS and FastGLT on Cora+GAT, with fixed weight sparsity sθ = 90% and graph sparsity sg ∈ {10%, 20%, · · · , 80%} (Right) Relative wallclock time (↓) compared to a single baseline training for searching GLTs. Note that FastGLT requires far less walllock time to obtain subnetwork/subgraph with better performance than multiple rounds of IMP employed in UGS. processes of GNNs (Jin et al. 2021; … view at source ↗

**Figure 2.** (Left) Hamming distance between masks generated by IMP, one-shot, and random sparsification methods on Cora with various graph sparsity levels sg ∈ {5.0%, 9.8%, . . . , 64.2%}. Notably, as sparsity increases, the distance between random and IMP masks rapidly grows, whereas one-shot masks retain greater similarity to IMP masks. (Right) Comparison of gradient magnitude/edge degree for weights/edges pruned by IMP or… view at source ↗

**Figure 3.** The detailed illustration of FastGLT compared to conventional IMP-based GLT. FastGLT replaces most of the timeconsuming iterative stages with one-shot pruning as a fast track, and leverages a gradual denoising module to fine-tune the one-shot tickets to the target sparsity with performance in no way inferior to that of IMP. is therefore assigned to the ⌈d/∆T⌉ th interval. At the end of interval µ(1 ≤ µ ≤ ⌈D/∆T⌉ = µ… view at source ↗

Figures from the paper (3 more)

**Figure 4.** Figure 4: Results of node classification over Cora/Citeseer/PubMed with GCN/GIN/GAT backbones. Black dash lines represent [PITH_FULL_IMAGE:figures/full_fig_p006_4.png]

**Figure 5.** Figure 5: The wall-lock time of UGS, WD-GLT, FastGLT to [PITH_FULL_IMAGE:figures/full_fig_p007_5.png]

**Figure 6.** Figure 6: (Left) Ablation study on ∆T. We vary ∆T ∈ {3, 5, 10, 20, 30, 50} on Ogbn-Arxiv+ResGCN and Citeseer+GAT with fixed sparsity {sθ = 80%, sg = 40%}; (Right) Test accuracy curves showcasing the denoising process from randomly initialized tickets and one-shot tickets on Ci…

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

33 extracted references · 30 canonical work pages

[1]

Includes a conceptual outline and/or pseudocode descrip- tion of AI methods introduced (yes)

work page
[2]

Clearly delineates statements that are opinions, hypothe- sis, and speculation from objective facts and results (yes)

work page
[3]

Hamilton, W.; Ying, Z.; and Leskovec, J

Pruning neural networks at initialization: Why are we missing the mark? arXiv preprint arXiv:2009.08576. Hamilton, W.; Ying, Z.; and Leskovec, J. 2017. Inductive representation learning on large graphs. In Proceedings of NIPS. Hu, W.; Fey, M.; Zitnik, M.; Dong, Y .; Ren, H.; Liu, B.; Catasta, M.; and Leskovec, J. 2020. Open graph benchmark: Datasets for m...

arXiv 2009
[4]

(yes/partial/no)

Proof sketches or intuitions are given for complex and/or novel results. (yes/partial/no)

work page
[5]

(yes/partial/no)

Appropriate citations to theoretical tools used are given. (yes/partial/no)

work page
[6]

(yes/partial/no/NA)

All theoretical claims are demonstrated empirically to hold. (yes/partial/no/NA)

work page
[7]

Provides well marked pedagogical references for less- familiar readers to gain background necessary to repli- cate the paper (yes) Does this paper make theoretical contributions? (no) If yes, please complete the list below

work page
[8]

(yes/partial/no)

All assumptions and restrictions are stated clearly and formally. (yes/partial/no)

work page

Show all 33 references

[9]

(yes/partial/no)

All novel claims are stated formally (e.g., in theorem statements). (yes/partial/no)
[10]

(yes/partial/no)

Proofs of all novel claims are included. (yes/partial/no)
[11]

This paper lists all final (hyper-)parameters used for each model/algorithm in the paper’s experiments. (yes)
[12]

(yes) A Algorithm Workflow We conclude the overall workflow of our FastGLT in Algo

This paper states the number and range of values tried per (hyper-)parameter during development of the paper, along with the criterion used for selecting the final pa- rameter setting. (yes) A Algorithm Workflow We conclude the overall workflow of our FastGLT in Algo. 1. Algor...

2023
[14]

(yes/no/NA) Does this paper rely on one or more datasets? (yes) If yes, please complete the list below

All experimental code used to eliminate or disprove claims is included. (yes/no/NA) Does this paper rely on one or more datasets? (yes) If yes, please complete the list below
[15]

A motivation is given for why the experiments are con- ducted on the selected datasets (yes/)
[16]

All novel datasets introduced in this paper are included in a data appendix. (NA)
[17]

All novel datasets introduced in this paper will be made publicly available upon publication of the paper with a license that allows free usage for research purposes. (NA)
[18]

All datasets drawn from the existing literature (poten- tially including authors’ own previously published work) are accompanied by appropriate citations. (yes)
[19]

All datasets drawn from the existing literature (poten- tially including authors’ own previously published work) are publicly available. (yes)
[20]

(NA) Does this paper include computational experiments? (yes) If yes, please complete the list below

All datasets that are not publicly available are described in detail, with explanation why publicly available alter- natives are not scientifically satisficing. (NA) Does this paper include computational experiments? (yes) If yes, please complete the list below
[21]

Any code required for pre-processing data is included in the appendix. (yes)
[22]

All source code required for conducting and analyzing the experiments is included in a code appendix. (yes)
[23]

All source code required for conducting and analyzing the experiments will be made publicly available upon publication of the paper with a license that allows free usage for research purposes. (yes)
[24]

All source code implementing new methods have com- ments detailing the implementation, with references to the paper where each step comes from (yes)
[25]

If an algorithm depends on randomness, then the method used for setting seeds is described in a way sufficient to allow replication of results. (NA)
[26]

This paper specifies the computing infrastructure used for running experiments (hardware and software), includ- ing GPU/CPU models; amount of memory; operating system; names and versions of relevant software libraries and frameworks. (yes)
[27]

This paper formally describes evaluation metrics used and explains the motivation for choosing these metrics. (yes)
[28]

This paper states the number of algorithm runs used to compute each reported result. (yes)
[29]

Analysis of experiments goes beyond single-dimensional summaries of performance (e.g., average; median) to in- clude measures of variation, confidence, or other distri- butional information. (yes)
[30]

(partial)

The significance of any improvement or decrease in performance is judged using appropriate statistical tests (e.g., Wilcoxon signed-rank). (partial)
[2018]

In Proceedings of the 2018 World Wide Web Conference, 449–458

Provable and practical approximations for the degree distribution using sublinear graph samples. In Proceedings of the 2018 World Wide Web Conference, 449–458. Evci, U.; Gale, T.; Menick, J.; Castro, P. S.; and Elsen, E

2018
[2019]

For Ogbl-Collab, we employed collaborations un- til 2017 as training edges, those in 2018 as validation edges, and those in 2019 as test edges

For Ogbn-Proteins, protein nodes were segregated into training, validation, and test sets based on their species of origin. For Ogbl-Collab, we employed collaborations un- til 2017 as training edges, those in 2018 as validation edges, and those in 2019 as test edges. E.5 More ...

2017
[2020]

In In- ternational Conference on Machine Learning , 2943–2952

Rigging the lottery: Making all tickets winners. In In- ternational Conference on Machine Learning , 2943–2952. PMLR. Fang, J.; Zhang, G.; Wang, K.; Du, W.; Duan, Y .; Wu, Y .; Zimmermann, R.; Chu, X.; and Liang, Y . 2024. On regular- ization for explaining graph neural networ...

2024 arXiv
[2021]

Reproducibility Checklist

Graph condensation for graph neural networks. arXiv preprint arXiv:2110.07580. Kipf, T. N.; and Welling, M. 2016. Semi-supervised classi- fication with graph convolutional networks. arXiv preprint arXiv:1609.02907. Kipf, T. N.; and Welling, M. 2017. Semi-supervised classi- fic...

2016 arXiv

Pith tools

Reviewed August 11, 2026 · model on record in the stance chip above.

[1] [1]

Includes a conceptual outline and/or pseudocode descrip- tion of AI methods introduced (yes)

work page

[2] [2]

Clearly delineates statements that are opinions, hypothe- sis, and speculation from objective facts and results (yes)

work page

[3] [3]

Hamilton, W.; Ying, Z.; and Leskovec, J

Pruning neural networks at initialization: Why are we missing the mark? arXiv preprint arXiv:2009.08576. Hamilton, W.; Ying, Z.; and Leskovec, J. 2017. Inductive representation learning on large graphs. In Proceedings of NIPS. Hu, W.; Fey, M.; Zitnik, M.; Dong, Y .; Ren, H.; Liu, B.; Catasta, M.; and Leskovec, J. 2020. Open graph benchmark: Datasets for m...

arXiv 2009

[4] [4]

(yes/partial/no)

Proof sketches or intuitions are given for complex and/or novel results. (yes/partial/no)

work page

[5] [5]

(yes/partial/no)

Appropriate citations to theoretical tools used are given. (yes/partial/no)

work page

[6] [6]

(yes/partial/no/NA)

All theoretical claims are demonstrated empirically to hold. (yes/partial/no/NA)

work page

[7] [7]

Provides well marked pedagogical references for less- familiar readers to gain background necessary to repli- cate the paper (yes) Does this paper make theoretical contributions? (no) If yes, please complete the list below

work page

[8] [8]

(yes/partial/no)

All assumptions and restrictions are stated clearly and formally. (yes/partial/no)

work page