Pith. sign in

REVIEW 4 major objections 6 minor 63 references

Ralts: Robust Aggregation for Enhancing Graph Neural Network Resilience on Bit-flip Errors

T0 review · 4 major / 6 minor · reviewed 2026-08-15 · deepseek-v4-flash

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

Pith's one-line read Three similarity-based aggregation filters recover most of the prediction accuracy that hardware bit flips in weights, embeddings, or adjacency matrices would destroy, at nearly the same computational cost.

desk verdict Genuinely new lightweight robust aggregations for GNN bit-flip robustness, with plausible gains; missing code, error bars, and an unexamined distribution-shift mechanism are the main soft spots. read the letter →

arxiv 2507.18804 v1 pith:ND6BPA47 submitted 2025-07-24 cs.LG

Wencheng Zou , Nan Wu This is my paper · ORCID
classification cs.LG
keywords graphneuralnetworksbit-fliperrorsrobustaggregationhardware-inducedfaultssilentdatacorruptionoutlierfilteringtopologyrecoverymessagepassing
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 claims that graph neural networks, which increasingly guard financial and medical decisions, have an under-examined vulnerability: random bit flips in memory from voltage-scaled or aging chips can silently corrupt model weights, node embeddings, or the adjacency matrix, and that corruption then propagates through message passing. It proposes Ralts, a set of three aggregation functions that filter outliers and repair edges during the aggregation step itself, so any message-passing GNN can adopt them by swapping one component. At a bit error rate of $3 \times 10^{-5}$, the paper reports, these functions improve prediction accuracy by at least 20% when errors are present in model weights or node embeddings, and by at least 10% when errors occur in the adjacency matrix, while staying within about $1.4\times$ the latency of plain mean aggregation. If the claim holds, Ralts is a cheap, model-agnostic defense that lets aggressive system optimizations such as voltage scaling proceed without sacrificing reliability.

What carries the argument

The load-bearing mechanism is the aggregation function itself, rewritten from a plain mean into an outlier filter and topology repairer. Distribution-based aggregation estimates per-dimension mean $\mu$ and standard deviation $\sigma$ of node embeddings during training, then at inference averages only values inside the interval $(\mu - a\sigma, \mu + b\sigma)$, treating the rest as bit-flip outliers. Dynamic-weight aggregation learns a center embedding $m_g$ alongside the model and weights each neighbor by $1/(\|h_u - m_g\|^2 + 1)$, so distant, likely-corrupted embeddings are attenuated rather than discarded. Cosine aggregation computes $\cos(x_i, x_j) = x_i\cdot x_j/(\|x_i\|\|x_j\|)$ for every edge and prunes edges whose similarity falls below a threshold $\alpha$, on the principle that structurally similar nodes are more likely to be connected. A combined variant sums the three with learnable scalars. All forms stay at $O(\ell|E|)$, matching mean aggregation, which is what allows the defense to ride on standard message-passing runtimes with modest latency growth.

What would settle it

Train a GNN on a graph dataset whose first-layer node embeddings have demonstrably non-Gaussian per-dimension distributions (bimodal or heavy-tailed, established by a normality test across nodes), inject bit flips into model weights at a BER of $3\times10^{-5}$, and compare distribution-based aggregation with plain mean aggregation. If the claimed at-least-20% improvement collapses because the fixed $\mu \pm a\sigma$ interval discards healthy embeddings or retains corrupted ones, the Gaussian premise fails; if the gain persists on such a dataset, the method's generality is confirmed.

Watch

Extended reading notes

Core claim

On its own terms, the paper establishes that GNN robustness to hardware bit errors is largely recoverable at the aggregation layer, without retraining and without changing the model architecture. Its central claim is that three robust aggregation functions, each exploiting graph similarity, keep the linear time complexity of mean aggregation while clearly outperforming existing robust alternatives: on average at a BER of $3\times10^{-5}$, distribution-based and dynamic-weight aggregation improve prediction accuracy by 44.1%, 27.4%, 32.6%, 22.2%, and 43.7% relative to mean, median, trimmed-mean, soft-median, and activation-clipping baselines, and cosine aggregation improves accuracy by roughly 9–16% against the same baselines for adjacency-matrix errors. The paper further claims these gains are stable across GCN, GAT, and GIN; across node-level and graph-level tasks; across dense and sparse models; and across small citation and molecular graphs as well as the much larger ogbn-arxiv and ogbn-products graphs. It also reports that combining the three functions with learnable scalars yields additional robustness when errors strike weights, embeddings, and topology simultaneously.

Load-bearing premise

The load-bearing premise is that per-dimension node embeddings are approximately Gaussian across nodes, so a fixed interval of a few standard deviations such as $\mu \pm a\sigma$ cleanly separates bit-flip outliers from informative values; embeddings are nonlinear outputs of neighborhood averaging, not sample means, and the interval bounds and similarity threshold are tuned per dataset.

Editorial extensions

If this is right

  • Any message-passing GNN can adopt the defense by swapping its aggregation function; no architectural change is needed, since the distribution statistics, the center embedding, and the combination weights are all obtained during ordinary training.
  • Because unprotected GNNs stay stable below a BER of roughly $10^{-7}$, and Ralts absorbs the damage above that threshold, system designers can justify voltage scaling or lighter error-correction codes in reliability-tolerant regimes.
  • Weight sparsification alone does not make GNNs more robust to bit flips, unlike DNNs, but Ralts improves accuracy consistently at every tested sparsity level of 15%, 30%, and 60%.
  • When errors simultaneously hit weights, node embeddings, and the adjacency matrix, the combined aggregation function outperforms each robust function individually because the three target different error sources.
  • Ralts scales to large, dense graphs: all three functions keep $O(\ell|E|)$ complexity, with profiled latencies on ogbn-arxiv and ogbn-products of roughly $1.7$–$2.1\times$ that of mean aggregation.

Reading between the lines

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

  • A cheap normality check on a clean model's per-dimension embedding statistics would predict, before any bit flips occur, whether distribution-based aggregation will help or hurt on a given dataset; the paper does not explore this diagnostic.
  • The hyperparameters $a$, $b$, and $\alpha$ are tuned per dataset, so a natural testable extension is estimating them automatically from clean validation data, or allowing them to vary per embedding dimension rather than globally.
  • Because the injection framework is fault-model agnostic, the same aggregation filters could plausibly absorb other silent-corruption sources such as timing errors or stuck-at faults, and possibly adversarial perturbations; this extrapolates beyond the paper's random independent bit-flip setup.
  • If the aggregation-level principle holds, the results imply that the efficiency–robustness trade-off in GNN accelerators is best resolved inside message passing rather than by expensive input reconstruction, a system-design consequence the authors leave mostly implicit.
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.

Referee Report

4 major / 6 minor

Summary. This paper studies the robustness of message-passing GNNs to random bit-flip errors in model weights, node embeddings, and adjacency matrices. It first reports an empirical sensitivity analysis across GCN, GAT, and GIN on six medium datasets and two OGB datasets. It then proposes Ralts, a set of three alternative aggregation functions—distribution-based trimming of outlier dimensions, dynamic weight averaging based on distance to a learned center, and cosine-similarity-based edge pruning—that can be used individually or combined. The evaluation compares these functions with mean, max, median, trimmed mean, soft median, and activation-clipping baselines under several BERs, including sparse models, and profiles execution latency. The headline claims are that at BER 3e-5 the robust aggregations improve accuracy by at least 20% for weight/embedding errors and at least 10% for adjacency errors, with execution efficiency comparable to PyG built-in mean aggregation.

Significance. If the empirical claims are taken at face value, Ralts offers a simple, architecture-agnostic, low-overhead defense against an underexplored failure mode, and the sensitivity analysis is a useful reference for hardware/GNN co-design. The paper's strengths are its broad evaluation (three models, eight datasets, three error locations, multiple BERs, dense and sparse models), the linear time complexity of the proposed aggregations, and the direct integration into message-passing frameworks. The efficiency profile is a valuable contribution. However, because the main quantitative claims depend on per-dataset hyperparameters whose selection procedure is not reported, and because the mechanism for weight-error robustness is not demonstrated, the significance as currently evidenced is weaker than the abstract suggests.

major comments (4)
  1. [Section IV-A and IV-C, Section VII-B] Hyperparameters a, b, and α are introduced as "adjusted for different datasets or GNNs" (Section IV-A) and as a "tunable threshold parameter" (Section IV-C), but the paper does not describe how these values are chosen, what values are used, or whether a validation set is used. Because the headline improvements in Section VII-B are computed on the same error conditions for which these hyperparameters can be tuned, the reported gains may partly reflect per-condition fitting rather than out-of-sample robustness. Please provide the exact selection protocol (tuning set, grid, criterion) and a table of the chosen hyperparameter values, or a sensitivity analysis showing that the results are stable across a wide range of a, b, and α.
  2. [Section IV-A, Figs. 6 and 9] The mechanism for weight-error robustness is not demonstrated. Section IV-A estimates µ and σ from clean training embeddings and then applies the fixed interval (µ−aσ, µ+bσ) at inference; Section IV-B similarly uses a center m_g. When bit flips occur in model weights, the corrupted network computes embeddings through systematically altered weights, so the whole per-dimension distribution can shift rather than only producing sparse outliers. The paper provides no measurement of how much the embedding distribution shifts under weight errors and no ablation with re-estimated, recentered, or adaptive thresholds. The CLT argument in Section IV-A addresses the shape of clean embeddings, not the train-test distribution shift. In the absence of such evidence, the strong improvement reported for weight errors cannot be attributed specifically to principled outlier removal. Please add this analysis or qualify the weight-error claim.
  3. [Section V (Implementation details), Figs. 6–14] The manuscript reports that results are averaged over five different random seeds with 10 independent runs per seed, but no figure or table presents error bars, standard deviations, confidence intervals, or significance tests. In several comparisons in Figs. 9–12 the curves lie close together, so the quantitative claims (e.g., 44.1% average improvement in Section VII-B) are not supported by any measure of variability. Please report per-condition means with variance or confidence intervals and, where feasible, paired tests across seeds for the main comparisons.
  4. [Section IV-B] The dynamic weight aggregation uses a "learnable center embedding" m_g trained via backpropagation, but the paper does not state whether m_g is a model parameter subject to bit flips in the weight-error scenario, nor does it give initialization, loss-term, or update details. If m_g is a stored parameter, it can be corrupted directly; if it is instead a statistic of clean embeddings, it faces the same distribution-shift problem as µ and σ. The manuscript should clarify the status of m_g and evaluate the sensitivity of dynamic-weight aggregation to corrupted or shifted centers.
minor comments (6)
  1. [Abstract and Section VII-B] The abstract's "at least 20%" improvement claim is not tied to a specific baseline; Section VII-B reports different numbers for different baselines (44.1% vs mean, 27.4% vs median, etc.). Please specify the baseline and conditions for the abstract claim.
  2. [Fig. 5] The Gaussian-supporting figure shows only the first element of first-layer embeddings for four dataset-model pairs, while the text claims that values in each dimension of node embeddings exhibit near-Gaussian behavior. Please show more dimensions/layers or add quantitative normality checks.
  3. [Eq. (2)] The weight formula Weight_u = 1/(d_u^2+1) is not invariant to the scale of node embeddings; the manuscript should discuss how the scale of d_u^2 affects the weighting and whether any normalization is needed.
  4. [Section IV-D and Fig. 11] The combination of aggregation functions with learnable scalars is mentioned, but no details are given for how the scalars are initialized or trained, and the Combined curve in Fig. 11 is not described in terms of this mechanism. Please specify the combination procedure.
  5. [Section VI and Fig. 6] The statement that GNNs maintain stable performance for BERs below 10^-7 to 10^-6 is not directly visible in Fig. 6, whose x-axis starts at 1e-6; please either extend the plots to these BERs or rephrase the claim.
  6. [Section VII-D and Fig. 13] The geometric-mean latency values quoted in the text (1.42x, 1.08x, 1.37x) do not clearly match the values shown in Fig. 13; please align the text and figure so the reported numbers can be verified.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: Ralts' robustness and efficiency claims are empirical evaluations against external baselines, not derivations that reduce to their own definitions.

full rationale

The paper makes no first-principles prediction that is defined in terms of its own outcome. Section IV defines three aggregation mechanisms procedurally: distribution-based aggregation fits per-dimension mean and standard deviation on clean training embeddings and trims values outside the fitted interval; dynamic-weight aggregation learns a center embedding and reweights by inverse squared distance; cosine aggregation prunes edges below a tunable similarity threshold. None of these definitions presupposes the accuracy improvements reported in Sections VII-B and VII-D. The reported gains are measured against external PyG baselines (mean, max, median, trimmed mean, soft median, and activation clipping) under injected bit flips, and they could in principle have gone the other way, as the paper itself observes for graph-level tasks where the gains are described as slightly diminished. The tunable hyperparameters a, b, and alpha are selected per dataset, but ordinary hyperparameter selection does not make the outcome equivalent to the input, and the paper does not define the target accuracy in terms of these parameters. The Gaussian/CLT motivation in Section IV-A is a heuristic whose failure would be a correctness or robustness risk, not a circular step, and the paper explicitly concedes that not all embedding distributions strictly follow Gaussian distributions. No load-bearing self-citation chain or imported uniqueness theorem appears, and the efficiency claims are profiled against PyG implementations. The paper is self-contained as an empirical robustness study, so no circularity is present.

Assumptions & free parameters 3 free parameters · 4 assumptions · 0 invented entities

The central robustness claims rest on three types of inputs: hand-tuned thresholds (a,b,α) that are not disclosed, empirical distributional assumptions about embeddings and graph homophily, and a random independent bit-flip fault model. The learnable center m_g in dynamic weight aggregation is a trainable parameter rather than a free hyperparameter, but it is an extra capacity not present in standard GNNs.

free parameters (3)
  • a (distribution lower threshold multiplier) = not reported
    Hand-tuned per dataset/GNN to set the lower bound of the accepted embedding value interval; no values disclosed.
  • b (distribution upper threshold multiplier) = not reported
    Hand-tuned per dataset/GNN to set the upper bound of the accepted embedding value interval; no values disclosed.
  • α (cosine similarity threshold) = not reported
    Hand-tuned per dataset/GNN to prune edges whose cosine similarity falls below α; no values disclosed.
assumptions (4)
  • domain assumption Per-dimension node embeddings are approximately Gaussian across nodes, so a fixed multiple of standard deviations separates bit-flip outliers from clean values.
    Invoked in Section IV-A and Fig. 5; justified by CLT on neighborhood averaging, but embeddings are nonlinear outputs, not sample means, so the Gaussian shape is an empirical assumption that may fail.
  • domain assumption Nodes with high feature similarity are more likely to be connected (homophily), so low-similarity edges are likely spurious and can be safely pruned.
    Stated in Section IV introduction; underpins cosine aggregation, but many real graphs are heterophilous.
  • domain assumption Bit-flip errors occur independently with probability p (BER) in memory bits, and each flip corrupts the stored value.
    Fault model in Section III, adopted from silicon characterization literature; reasonable but only one fault model, not covering correlated or stuck-at faults.
  • domain assumption Bit flips in weights/embeddings produce extreme values that can be identified by statistical deviation.
    Implicit in distribution-based and dynamic weight aggregations; a flip in the mantissa may produce small perturbations rather than extreme outliers, which these methods would miss.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Ralts: Robust Aggregation for Enhancing Graph Neural Network Resilience on Bit-flip Errors." pith.science (2026). https://pith.science/paper/ND6BPA47

@misc{pith2026250718804,
  author       = {Pith},
  title        = {Pith review of: Ralts: Robust Aggregation for Enhancing Graph Neural Network Resilience on Bit-flip Errors},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ND6BPA47}},
  note         = {Machine review of arXiv:2507.18804}
}
abstract

Graph neural networks (GNNs) have been widely applied in safety-critical applications, such as financial and medical networks, in which compromised predictions may cause catastrophic consequences. While existing research on GNN robustness has primarily focused on software-level threats, hardware-induced faults and errors remain largely underexplored. As hardware systems progress toward advanced technology nodes to meet high-performance and energy efficiency demands, they become increasingly susceptible to transient faults, which can cause bit flips and silent data corruption, a prominent issue observed by major technology companies (e.g., Meta and Google). In response, we first present a comprehensive analysis of GNN robustness against bit-flip errors, aiming to reveal system-level optimization opportunities for future reliable and efficient GNN systems. Second, we propose Ralts, a generalizable and lightweight solution to bolster GNN resilience to bit-flip errors. Specifically, Ralts exploits various graph similarity metrics to filter out outliers and recover compromised graph topology, and incorporates these protective techniques directly into aggregation functions to support any message-passing GNNs. Evaluation results demonstrate that Ralts effectively enhances GNN robustness across a range of GNN models, graph datasets, error patterns, and both dense and sparse architectures. On average, under a BER of $3\times10^{-5}$, these robust aggregation functions improve prediction accuracy by at least 20\% when errors present in model weights or node embeddings, and by at least 10\% when errors occur in adjacency matrices. Ralts is also optimized to deliver execution efficiency comparable to built-in aggregation functions in PyTorch Geometric.

Figures

Figures reproduced from arXiv: 2507.18804 by the authors.

Figure 1
Figure 1. Bit errors in node embeddings (top subfigure) and adjacency [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Bit errors and energy consumption in memory with voltage [PITH_FULL_IMAGE:figures/full_fig_p002_2.png] view at source ↗
Figure 3
Figure 3. Bit error injection for analyzing GNN robustness, where bit [PITH_FULL_IMAGE:figures/full_fig_p002_3.png] view at source ↗
Figures from the paper (8 more)
Figure 4
Figure 4. Figure 4: Robust aggregation in GNNs: (a) distribution-based aggregation trims extreme values using distribution analysis; (b) dynamic weight [PITH_FULL_IMAGE:figures/full_fig_p003_4.png]
Figure 5
Figure 5. Figure 5: Distribution of the first element in the first-layer node [PITH_FULL_IMAGE:figures/full_fig_p003_5.png]
Figure 6
Figure 6. Figure 6: GNN prediction accuracy under different intensities of bit-flip [PITH_FULL_IMAGE:figures/full_fig_p005_6.png]
Figure 7
Figure 7. Figure 7: GNN prediction accuracy under different intensities of bit-flip [PITH_FULL_IMAGE:figures/full_fig_p005_7.png]
Figure 9
Figure 9. Figure 9: Comparison of GNN prediction accuracy using robust ag [PITH_FULL_IMAGE:figures/full_fig_p006_9.png]
Figure 11
Figure 11. Figure 11: Comparison of GNN prediction accuracy using different [PITH_FULL_IMAGE:figures/full_fig_p007_11.png]
Figure 13
Figure 13. Figure 13: Profiled latency of different aggregation functions across [PITH_FULL_IMAGE:figures/full_fig_p007_13.png]
Figure 14
Figure 14. Figure 14: Comparison of GIN prediction accuracy and latency using [PITH_FULL_IMAGE:figures/full_fig_p008_14.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

63 extracted references · 55 canonical work pages

  1. [1]

    Semi-supervised classification with graph convolutional networks,

    T. N. Kipf and M. Welling, “Semi-supervised classification with graph convolutional networks,” arXiv preprint arXiv:1609.02907 , 2016

    arXiv 2016

  2. [2]

    Graph embedding on biomedical networks: methods, applications and evaluations,

    X. Yue et al. , “Graph embedding on biomedical networks: methods, applications and evaluations,” Bioinformatics, 2020

    work page 2020

  3. [3]

    A comprehensive survey on graph neural networks,

    Z. Wu et al., “A comprehensive survey on graph neural networks,” IEEE TNNLS, 2020

    work page 2020

  4. [4]

    Ironman: Gnn-assisted design space exploration in high- level synthesis via reinforcement learning,

    N. Wu et al., “Ironman: Gnn-assisted design space exploration in high- level synthesis via reinforcement learning,” in Proc. GLSVLSI, 2021

    work page 2021

  5. [5]

    Build a gnn-based real-time fraud detection solution using amazon sagemaker, amazon neptune, and the deep graph library

    J. Zhang et al. , “Build a gnn-based real-time fraud detection solution using amazon sagemaker, amazon neptune, and the deep graph library.” https://aws.amazon.com/blogs/machine-learning/build-a-gnn-based- real-time-fraud-detection-solution-using-amazon-sagemaker-amazon- neptune-and-the-deep-graph-library/, [AWS Machine Learning Blog, Online; accessed 01-...

    work page 2024

  6. [6]

    Financial fraud detection using graph neural networks: A systematic review,

    S. Motie and B. Raahemi, “Financial fraud detection using graph neural networks: A systematic review,”Expert Systems With Applications, 2023

    work page 2023

  7. [7]

    Cat-gnn: Enhancing credit card fraud detection via causal temporal graph neural networks,

    Y . Duan et al., “Cat-gnn: Enhancing credit card fraud detection via causal temporal graph neural networks,” arXiv preprint arXiv:2402.14708 , 2024

    arXiv 2024

  8. [8]

    Scalable graph learning for anti-money laundering: A first look,

    M. Weber et al., “Scalable graph learning for anti-money laundering: A first look,” arXiv preprint arXiv:1812.00076 , 2018

    arXiv 2018

Show all 63 references
  1. [9]

    Enhancing graph neural network-based fraud detectors against camouflaged fraudsters,

    Y . Dou et al. , “Enhancing graph neural network-based fraud detectors against camouflaged fraudsters,” in Proc. CIKM, 2020

    2020

  2. [10]

    αcyber: Enhancing robustness of android malware detection system against adversarial attacks on heterogeneous graph based model,

    S. Hou et al. , “αcyber: Enhancing robustness of android malware detection system against adversarial attacks on heterogeneous graph based model,” in Proc. CIKM, 2019

    2019

  3. [11]

    Graph neural networks for hardware vulnerability analysis—can you trust your gnn?,

    L. Alrahis and O. Sinanoglu, “Graph neural networks for hardware vulnerability analysis—can you trust your gnn?,” in Proc. VTS, IEEE, 2023

    2023

  4. [12]

    Graph neural networks for vulnerability detection: A counterfactual explanation,

    Z. Chu et al. , “Graph neural networks for vulnerability detection: A counterfactual explanation,” in Proceedings of the 33rd ACM SIGSOFT International Symposium on Software Testing and Analysis , 2024

    2024

  5. [13]

    Fraud Reports by Federal Trade Commission

    “Fraud Reports by Federal Trade Commission.” https: //www.ftc.gov/news-events/news/press-releases/2024/02/nationwide- fraud-losses-top-10-billion-2023-ftc-steps-efforts-protect-public, [Online; accessed 01-Nov-2024]

    2024

  6. [14]

    Value of fraud loss in the United States in 2022, by payment method

    “Value of fraud loss in the United States in 2022, by payment method.” https://www.statista.com/statistics/958997/fraud-loss-usa-by- payment-method/, [Online; accessed 01-Nov-2024]

    2022

  7. [15]

    Robust machine learning systems: Challenges, current trends, perspectives, and the road ahead,

    M. Shafique et al. , “Robust machine learning systems: Challenges, current trends, perspectives, and the road ahead,” IEEE D & T , 2020

    2020

  8. [16]

    Neutron soft error rate measurements in a 90-nm cmos process and scaling trends in sram from 0.25-/spl mu/m to 90-nm generation,

    P. Hazucha et al. , “Neutron soft error rate measurements in a 90-nm cmos process and scaling trends in sram from 0.25-/spl mu/m to 90-nm generation,” in IEEE IEDM, 2003

    2003

  9. [17]

    Razor: circuit-level correction of timing errors for low- power operation,

    D. Ernst et al., “Razor: circuit-level correction of timing errors for low- power operation,” IEEE Micro, 2004

    2004

  10. [18]

    Impact of technology and voltage scaling on the soft error susceptibility in nanoscale cmos,

    V . Chandra and R. Aitken, “Impact of technology and voltage scaling on the soft error susceptibility in nanoscale cmos,” in IEEE International Symposium on Defect and Fault Tolerance of VLSI Systems , 2008

    2008

  11. [19]

    Thundervolt: enabling aggressive voltage underscaling and timing error resilience for energy efficient deep learning accelera- tors,

    J. Zhang et al., “Thundervolt: enabling aggressive voltage underscaling and timing error resilience for energy efficient deep learning accelera- tors,” in Proc. DAC, 2018

    2018

  12. [20]

    Detecting silent data corruptions in the wild,

    H. D. Dixit et al., “Detecting silent data corruptions in the wild,” arXiv preprint arXiv:2203.08989, 2022

    2022 arXiv

  13. [21]

    Detection and prevention of silent data corruption in an exabyte-scale database system,

    D. F. Bacon, “Detection and prevention of silent data corruption in an exabyte-scale database system,” in The 18th IEEE Workshop on Silicon Errors in Logic-System Effects , 2022

    2022

  14. [22]

    An assessment of vulnerability of hardware neural networks to dynamic voltage and temperature variations,

    X. Jiao et al. , “An assessment of vulnerability of hardware neural networks to dynamic voltage and temperature variations,” in Proc. ICCAD, 2017

    2017

  15. [23]

    Dnn engine: A 28-nm timing-error tolerant sparse deep neural network processor for iot applications,

    P. N. Whatmough et al. , “Dnn engine: A 28-nm timing-error tolerant sparse deep neural network processor for iot applications,” IEEE JSSC, 2018

    2018

  16. [24]

    Ares: A framework for quantifying the resilience of deep neural networks,

    B. Reagen et al., “Ares: A framework for quantifying the resilience of deep neural networks,” in Proc. DAC, 2018

    2018

  17. [25]

    Bit error robustness for energy-efficient dnn accelera- tors,

    D. Stutz et al. , “Bit error robustness for energy-efficient dnn accelera- tors,” Proc. MLSys, 2021

    2021

  18. [26]

    Adversarial attacks and defenses on graphs,

    W. Jin et al. , “Adversarial attacks and defenses on graphs,” ACM SIGKDD Explorations Newsletter , 2021

    2021

  19. [27]

    Adversarial attack and defense on graph data: A survey,

    L. Sun et al., “Adversarial attack and defense on graph data: A survey,” IEEE Transactions on Knowledge and Data Engineering , 2022

    2022

  20. [28]

    Pygfi: Analyzing and enhancing robustness of graph neural networks against hardware errors,

    R. Wang et al. , “Pygfi: Analyzing and enhancing robustness of graph neural networks against hardware errors,” arXiv preprint arXiv:2212.03475, 2022

    2022 arXiv

  21. [29]

    Graph structure learning for robust graph neural networks,

    W. Jin et al., “Graph structure learning for robust graph neural networks,” in Proc. SIGKDD, 2020

    2020

  22. [30]

    Topology attack and defense for graph neural networks: an optimization perspective,

    K. Xu et al., “Topology attack and defense for graph neural networks: an optimization perspective,” in Proc. IJCAI, 2019

    2019

  23. [31]

    Robustness of graph neural networks at scale,

    S. Geisler et al., “Robustness of graph neural networks at scale,” Proc. NeurIPS, 2021

    2021

  24. [32]

    Fast graph representation learning with PyTorch Geometric,

    M. Fey and J. E. Lenssen, “Fast graph representation learning with PyTorch Geometric,” in ICLR Workshop on Representation Learning on Graphs and Manifolds , 2019

    2019

  25. [33]

    How powerful are graph neural networks?,

    K. Xu et al., “How powerful are graph neural networks?,” in Proc. ICLR, 2018

    2018

  26. [34]

    Computing graph neural networks: A survey from algorithms to accelerators,

    S. Abadal et al. , “Computing graph neural networks: A survey from algorithms to accelerators,” ACM CSUR, 2021

    2021

  27. [35]

    Empowering graph representation learning with test-time graph transformation,

    W. Jin et al., “Empowering graph representation learning with test-time graph transformation,” in Proc. ICLR, 2023

    2023

  28. [36]

    Resilient low voltage accelerators for high energy efficiency,

    N. Chandramoorthy et al. , “Resilient low voltage accelerators for high energy efficiency,” in Proc. HPCA, IEEE, 2019

    2019

  29. [37]

    On characterizing near-threshold sram failures in finfet technology,

    S. Ganapathy et al., “On characterizing near-threshold sram failures in finfet technology,” in Proc. DAC, 2017

    2017

  30. [38]

    Matic: Learning around errors for efficient low-voltage neural network accelerators,

    S. Kim et al. , “Matic: Learning around errors for efficient low-voltage neural network accelerators,” in Proc. DATE, IEEE, 2018

    2018

  31. [39]

    The theory of error-correcting codes,

    F. MacWilliams, “The theory of error-correcting codes,” Elsevier Science Publishers BV google schola , 1977

    1977

  32. [40]

    Chapter 5 - error coding techniques,

    “Chapter 5 - error coding techniques,” in Architecture Design for Soft Errors (S. Mukherjee, ed.), pp. 161–206, Morgan Kaufmann, 2008

    2008

  33. [41]

    Berry: Bit error robustness for energy-efficient reinforce- ment learning-based autonomous systems,

    Z. Wan et al., “Berry: Bit error robustness for energy-efficient reinforce- ment learning-based autonomous systems,” in Proc. DAC, IEEE, 2023

    2023

  34. [42]

    Mulberry: Enabling bit-error robustness for energy- efficient multi-agent autonomous systems,

    Z. Wan et al. , “Mulberry: Enabling bit-error robustness for energy- efficient multi-agent autonomous systems,” in Proc. ASPLOS, 2024

    2024

  35. [43]

    Greentpu: Improving timing error resilience of a near- threshold tensor processing unit,

    P. Pandey et al., “Greentpu: Improving timing error resilience of a near- threshold tensor processing unit,” in Proc. DAC, 2019

    2019

  36. [44]

    Read: Reliability-enhanced accelerator dataflow opti- mization using critical input pattern reduction,

    Z. Zhang et al. , “Read: Reliability-enhanced accelerator dataflow opti- mization using critical input pattern reduction,” in Proc. ICCAD, 2023

    2023

  37. [45]

    Sparse dnns with improved adversarial robustness,

    Y . Guo et al., “Sparse dnns with improved adversarial robustness,” Proc. NeurIPS, 2018

    2018

  38. [46]

    Reliability evaluation of pruned neural networks against errors on parameters,

    Z. Gao et al., “Reliability evaluation of pruned neural networks against errors on parameters,” in Proc. DFT, 2020

    2020

  39. [47]

    Understanding structural vulnerability in graph convo- lutional networks,

    L. Chen et al., “Understanding structural vulnerability in graph convo- lutional networks,” in Proc. IJCAI, 2021

    2021

  40. [48]

    Link prediction based on graph neural net- works,

    M. Zhang and Y . Chen, “Link prediction based on graph neural net- works,” Proc. NeurIPS, 2018

    2018

  41. [49]

    Evaluating graph neural networks for link prediction: Current pitfalls and new benchmarking,

    J. Li et al. , “Evaluating graph neural networks for link prediction: Current pitfalls and new benchmarking,” Proc. NeurIPS, 2024

    2024

  42. [50]

    Birds of a feather: Homophily in social networks,

    M. McPherson et al., “Birds of a feather: Homophily in social networks,” Annual review of sociology , 2001

    2001

  43. [51]

    Principal neighbourhood aggregation for graph nets,

    G. Corso et al. , “Principal neighbourhood aggregation for graph nets,” Proc. NeurIPS, 2020

    2020

  44. [52]

    Graph attention networks,

    P. Veli ˇckovi´c et al. , “Graph attention networks,” arXiv preprint arXiv:1710.10903, 2017

    2017 arXiv

  45. [53]

    Revisiting semi-supervised learning with graph embed- dings,

    Z. Yang et al., “Revisiting semi-supervised learning with graph embed- dings,” in Proc. ICML, 2016

    2016

  46. [54]

    Pubmed dataset

    “Pubmed dataset.” https://pubmed.ncbi.nlm.nih.gov/download/, [Online; accessed 01-Oct-2024]

    2024

  47. [55]

    Subgraph matching kernels for attributed graphs,

    N. Kriege and P. Mutzel, “Subgraph matching kernels for attributed graphs,” in Proc. ICML, 2012

    2012

  48. [56]

    Protein function prediction via graph kernels,

    K. M. Borgwardt et al., “Protein function prediction via graph kernels,” Bioinformatics, pp. i47–i56, 06 2005

    2005

  49. [57]

    The network data repository with interactive graph analytics and visualization,

    R. A. Rossi and N. K. Ahmed, “The network data repository with interactive graph analytics and visualization,” in AAAI, 2015

    2015

  50. [58]

    Microsoft academic graph: When experts are not enough,

    K. Wang et al. , “Microsoft academic graph: When experts are not enough,” Quantitative Science Studies , 2020

    2020

  51. [59]

    The extreme classification repository: Multi-label datasets and code,

    K. Bhatia et al. , “The extreme classification repository: Multi-label datasets and code,” 2016

    2016

  52. [60]

    Towards sparsification of graph neural networks,

    H. Peng et al. , “Towards sparsification of graph neural networks,” in Proc. ICCD, 2022

    2022

  53. [61]

    Hydra: Pruning adversarially robust neural networks,

    V . Sehwag et al., “Hydra: Pruning adversarially robust neural networks,” Proc. NeurIPS, 2020

    2020

  54. [62]

    Sparsity winning twice: Better robust generalization from more efficient training,

    T. Chen et al. , “Sparsity winning twice: Better robust generalization from more efficient training,” in Proc. ICLR, 2022

    2022

  55. [63]

    Time and space complexity of graph convolutional networks,

    D. Blakely et al. , “Time and space complexity of graph convolutional networks,” 2021

    2021

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.