Pith. sign in

REVIEW 4 major objections 6 minor 49 references

The paper claims that a single vision-language model, OMG-VLM, learns node classification and link prediction across text-attributed, image-attributed, and multimodal-attributed graphs at once, outperforming specialized GNN- and LLM-based b

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

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

T0 review · deepseek-v4-flash

2026-08-01 13:21 UTC pith:PDZ4H6ZX

load-bearing objection Genuinely new unified-modality graph learning framework with strong ablations, but single-run numbers and unexplained baseline shifts between tables leave 'consistently outperforms' under-supported. the 4 major comments →

arxiv 2607.19128 v1 pith:PDZ4H6ZX submitted 2026-07-21 cs.LG

One Model, Many Graphs: Learning over Attributed Graphs across Heterogeneous Modalities with Vision-Language Models

Jiayi Yang , Yifang Chen , Yuanfu Sun , Jiajin Liu , Qiaoyu Tan This is my paper
classification cs.LG
keywords heterogeneous modality schemasattributed graph learningvision-language modelsgraph adapterstarget-aware textual aggregationgraph-aware visual representation learningnode classificationlink prediction
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

OMG-VLM tries to establish that one pretrained vision-language model can serve as a shared backbone for attributed graph learning no matter whether nodes carry text, images, or both. The paper argues that by converting textual and visual neighborhoods into compact, VLM-native token sequences — through target-conditioned cross-attention over neighbor token banks — a single model can replace the usual per-modality architectures. On node classification and link prediction across e-commerce, citation, and social graphs, the model beats state-of-the-art GNN- and LLM-based baselines on every in-domain benchmark and on most unseen-graph transfer benchmarks. If true, practitioners would no longer need separate models for text-attributed, image-attributed, and multimodal-attributed graphs, and a single model could be trained once and reused across domains and modalities.

Core claim

In the paper's own telling, the central discovery is that the native embedding space of a frozen vision-language model can host graph context directly. Textual neighbor tokens are embedded with the backbone's own tokenizer, and image neighbor tokens pass through the frozen visual encoder, with two small adapters compressing and aggregating them through cross-attention conditioned on the target node. This design lets a single generative model optimize the same next-token objective over text-attributed, image-attributed, and multimodal-attributed graphs jointly. The empirical claim is that this unified model outperforms every category of baseline — GNNs, text-only LLM methods, multimodal LLM m

What carries the argument

The load-bearing components are the two structure-aware graph adapters. The target-aware textual aggregation adapter (ΦT) uses learnable query tokens attended to the target text, then retrieves relevant neighbor text via multi-head cross-attention over a bank of neighbor-token embeddings produced by the VLM's own tokenizer. The graph-aware visual representation learning adapter (ΦI) compresses each neighbor image into a small set of tokens with learnable visual queries, then refines the target image tokens by cross-attending to the compressed neighbor bank across L aggregation layers. Both adapters output token sequences that plug directly into the VLM's input and are trained end-to-end with

Load-bearing premise

The load-bearing assumption is that a fixed-size bag of neighbor attribute tokens drawn from mixed hop distances — with no edge index, hop distance, or path information — supplies enough graph context for node classification and link prediction; the near-random Cora link-prediction result suggests that when structure matters beyond the attributes, the model has no way to see it.

What would settle it

On a link-prediction benchmark where node attributes are randomly permuted across nodes while the edge set is held fixed (so attributes carry no signal about edges), OMG-VLM should drop toward chance accuracy if its neighborhood context is purely attribute-based, while a GNN using only adjacency would retain structural signal; the Cora LP result (51.80) already points in this direction.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • A single model can be trained once on a mix of text-, image-, and multimodal-attributed graphs and then applied to unseen graphs in any of those schemas without dataset-specific fine-tuning; the transfer experiments show consistent gains (up to +20.15 points) over the best baseline.
  • Graph context can be compressed to a bounded token budget — about 8 text tokens per target and 32 tokens per image neighbor — keeping inference fast and prompt lengths roughly three times shorter than naively serializing the neighborhood into a VLM prompt.
  • The gains come from the trained adapters rather than from the VLM's pretrained knowledge alone: zero-shot prompting with serialized neighborhoods performs far worse (e.g., 8.50 vs. 67.85 on the arXiv node classification task), and the framework remains effective when the backbone is swapped across three different VLM families and scales.
  • Per-neighbor image compression cuts computational cost by about 75% while keeping accuracy within roughly 1–2 points, so visual neighborhood aggregation scales to larger neighborhood sizes without prohibitive FLOPs.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The fixed-size neighbor-token-bag design has no explicit edge index, hop-distance, or path encoding, so the model cannot represent structure that is not inferable from the attributes of the collected neighbors; the near-chance Cora link-prediction result (51.80) is consistent with this limitation on sparse citation topologies, and a natural extension would add a lightweight positional or edge-inde
  • If the transfer gains reflect reusable neighborhood semantics rather than memorized structure, then attribute-shuffling experiments — permuting node attributes while keeping edges fixed — should sharply degrade performance; that experiment would pin down what the adapters actually learn.
  • The empirical support is strongest on graphs where attributes are highly self-descriptive (Amazon products, Reddit images); a stress test on anonymized graphs with uninformative or randomly permuted attributes would reveal whether the method learns graph structure at all or mostly aggregates semantic similarity.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces OMG-VLM, a unified vision-language-model framework for node classification and link prediction on attributed graphs with heterogeneous modality schemas (text-only, image-only, and multimodal). The method uses a frozen VLM backbone, LoRA adaptation, and two adapters: target-aware textual aggregation and graph-aware visual representation learning. The authors claim consistent state-of-the-art performance over GNN- and LLM-based baselines on in-domain and transfer benchmarks, with ablations and efficiency analyses.

Significance. If the empirical claims hold, OMG-VLM would be a useful contribution: it is one of the first systematic attempts to use a single VLM backbone for heterogeneous attributed graphs, avoids external modality encoders by operating in the VLM's native embedding space, and is accompanied by code, multi-backbone ablations, zero-shot controls, and computational-cost comparisons. However, the central comparative claim currently rests on single-run numbers and on baseline tables that are internally inconsistent at load-bearing points. The architectural 'structure-aware' claim is also not fully supported by the design as written. These issues must be resolved before the paper can be accepted.

major comments (4)
  1. [Appendix C] All results are reported from single runs, with no seed variance, confidence intervals, or significance tests (Appendix C: 'All results are reported from single runs'). Several headline margins in Table 1 are small: CoraLP +0.70 over UniGraph2, CDsNC +1.85 over GraphPrompter, and arXivNC +3.26 over LLaGA. Without variance estimates or repeated trials, the claim that OMG-VLM 'consistently outperforms' cannot be distinguished from noise. Please provide multi-seed results with mean±std and, where relevant, significance tests.
  2. [Tables 1 and 3] Baseline numbers are internally inconsistent between Table 1 and Table 3. GraphVLM on RedditS LP changes from 55.70 to 90.81, GraphPrompter on PubMed NC from 56.92 to 11.59 and on CDs NC from 30.71 to 11.83, and Graph4MM on Cora LP from 14.90 to 52.60. Appendix C.2 explains only LLaGA and MLaGA; it does not explain why these other baselines shift by up to 35 points when the stated protocol change (full ogbn-arXiv) should alter training-set composition, not the evaluation of transfer graphs. The paper must clarify which configuration supports the headline claim and report both configurations consistently, ideally with error bars.
  3. [Eqs. (2), (6), and (9)] The input representation in Eq. (2) concatenates neighbor attribute token banks but contains no edge index, hop-distance, path, or any topological encoding. RoPE in Eq. (6) only encodes token order inside the text adapter, not graph structure. Thus the model is a fixed-size attribute-bag aggregator, not a structure-aware graph learner as claimed. This explains why CoraLP in Table 1 is near random (51.80) and why increasing K to 70 changes it to 61.90 (Table 9): the model only sees more attributes, not more structure. Either add explicit structural encoding or substantially temper the 'structure-aware' claim.
  4. [Section 5.1 / Table 2] The 'diverse domains' claim for transfer is overstated. The held-out graphs are close variants of the training domains: PubMed is another citation graph like arXiv; RedditM is another social graph like RedditS; Amazon VideoGames/CDs are from the same Amazon platform as Arts/Movies. True cross-domain generalization (e.g., citation to e-commerce to social) is not demonstrated. Please qualify the generalization claim to within-domain transfer or add genuinely unseen domains.
minor comments (6)
  1. [Appendix E.3] Typo: 'effecti veness' should be 'effectiveness'.
  2. [References] Reference 'Orbifold, 2019. The Cora dataset' is nonstandard; please use the canonical citation for Cora (e.g., Sen et al., 2008) or clearly identify the version used.
  3. [Table 3 caption] The caption should explicitly state that all training and evaluation settings other than the ogbn-arXiv sampling are identical to Table 1, and should explain why the listed transfer baselines change so dramatically. Currently the reader cannot tell whether these are two valid configurations or one configuration is erroneous.
  4. [Figure 3] The legend 'Best Mean Best Attn. Ours' is confusing because the bars are not labeled per variant. Please clarify which adapter replacement each bar corresponds to.
  5. [Appendix B] The sentence 'we first randomly include all available 1-hop neighbors of the center node' is contradictory: if all 1-hop neighbors are included, no randomness applies at that step. Clarify the sampling procedure.
  6. [Section 5.2] The phrase 'with consistent gains on all other transfer datasets' is too strong given the small CoraLP margin and the baseline inconsistency; please report margins and confidence information alongside such claims.

Circularity Check

1 steps flagged

Link prediction is scored from neighborhood contexts that already contain the queried edge's endpoints, making the answer readable from the input by construction.

specific steps
  1. self definitional [Sec. 3 (Eq. 1, Eq. 2), Appendix B, Table 13 (link-prediction prompt templates)]
    "Given a target node v, ... Xv = ΦT (Tv,NT(v))∥ΦI(Iv,NI(v)) ... For link prediction, the query is a node pair (vi, vj) and Y denotes whether an edge exists between them. ... we select neighbors from predefined graph edges during both training and evaluation. ... Given two nodes ... Image neighborhood: <image neighborhood> ... are these two central nodes connected?"

    For a positive test pair (vi, vj), the input neighbor banks NT/NI are built from the graph edges, so vj is a 1-hop neighbor of vi and vi is a 1-hop neighbor of vj. The query node's own attributes therefore already appear in the other node's neighborhood context. For a negative pair, the queried node is absent from those banks. The link-prediction answer can thus be recovered by checking whether the other endpoint's tokens occur in the supplied neighbor context, i.e., the target Y is encoded in the input by construction rather than predicted from a masked graph. The paper does not state that the query edge is removed before neighborhood construction, so the reported LP gains (e.g., RedditS 99.10, Movies 95.30, RedditM 92.45) partly reduce to a membership test on the input.

full rationale

The only substantive circularity I can substantiate from the paper's text is in the link-prediction evaluation: the neighbor samplers used to build Eq. (2)'s input are defined over graph edges, and the link-prediction label is precisely whether the queried edge exists. Unless the queried edge is masked — which the paper never states — the model's input already contains the other endpoint for positive pairs, so the claimed link-prediction performance is partly an artifact of the input construction rather than a structural prediction. This is a self-definitional reduction under the task's pattern 1. I do not count the self-citations to GraphVLM/MLaGA as load-bearing circularity: those are baselines, and the paper also compares against external methods and reports its own ablations. The single-run and Table 1-vs-Table 3 inconsistencies are serious reproducibility/correctness concerns, but they are not circularity under the provided rules. Node classification, adapter ablations, and zero-shot controls are not constructed from the predicted label and appear to have independent content. Overall, the link-prediction shortcut affects several headline numbers (including the largest transfer gains), so a partial-circularity score of 6 is appropriate; the node-classification results keep the paper from being fully circular.

Axiom & Free-Parameter Ledger

5 free parameters · 5 axioms · 0 invented entities

The paper introduces no new physical entities; the only numbers fitted/selected are architectural hyperparameters (MT, MI, K, L, LoRA rank, learning rate) tuned via in-domain sensitivity experiments. The substantive 'axioms' are that the VLM's embedding space is cross-modally compatible and that an unordered bag of neighbor attributes is sufficient graph context.

free parameters (5)
  • MT (text query count) = 8
    Chosen from sensitivity Table 9; increasing from 1 to 8 improves RedditM LP from 79.10 to 92.45, but larger values do not help consistently.
  • MI (per-neighbor visual token budget) = 32
    Chosen as balance; MI=8 removes useful detail, MI=64 gives no consistent gain (Table 9).
  • K (neighborhood budget) = 10
    Default choice materially affects transfer: CoraLP improves from 51.80 at K=10 to 61.90 at K=70, but the paper keeps K=10 as default.
  • L (visual aggregation layers) = 1
    Figure 7 shows L=1 best on ArtsNC; deeper aggregation does not improve reliably, so the paper fixed L=1.
  • Training hyperparameters (LoRA r=64, alpha=16, lr=1e-5, 3 epochs) = r=64, a=16, lr=1e-5, 3 epochs
    Hand-set optimization choices from Appendix B; not derived from first principles and no sensitivity reported for them.
axioms (5)
  • domain assumption Pretrained VLM text-token and ViT visual-patch embeddings are sufficiently aligned that concatenating adapter outputs in the VLM's native embedding space supports cross-modal reasoning without explicit alignment (Sec. 4.1, Eq. 2).
    Central to the 'avoids explicit cross-modal realignment' claim; if the spaces are misaligned, the unified objective Eq. (15) has no reason to transfer.
  • domain assumption A bag of up to K neighbor attributes, retrieved by cross-attention, captures the graph structure needed for node classification and link prediction (Eqs. 8, 12).
    No edge/path/topology features are fed to the model; all structure is encoded as neighbor text/image tokens. The near-random Cora transfer result suggests this assumption can fail.
  • standard math Standard Transformer self-/cross-attention, RoPE, and autoregressive LM objective provide a valid trainable backend (Eqs. 3, 6, 8, 11).
    Background machinery assumed correct; not derived in the paper.
  • domain assumption Frozen VLM visual encoders (Qwen-VL, LLaVA, Qwen3-VL) provide useful visual features for graph reasoning without fine-tuning (Sec. 4.3, Table 10).
    The paper keeps the visual encoder frozen; if frozen ViT features are not discriminative for the graph tasks, the image adapter cannot compensate.
  • domain assumption Downsampling ogbn-arXiv and the exact positive/negative edge split for link prediction preserve a fair comparison (Appendices A and C).
    Baselines are reproduced under this modified protocol; reported differences from original papers are attributed to it, but no ablation isolates the effect of split construction.

pith-pipeline@v1.3.0-alltime-deepseek · 20976 in / 14011 out tokens · 132530 ms · 2026-08-01T13:21:19.353113+00:00 · methodology

0 comments
read the original abstract

Vision-language models (VLMs) provide a unified representation space for textual and visual information, yet their potential as general-purpose backbones for graph-structured data remains largely unexplored. In practice, attributed graphs exhibit substantial modality heterogeneity: some graphs contain only textual node attributes, others only visual attributes, while still others provide both. Existing graph learning approaches are typically designed for fixed modality schemas, requiring separate models for different settings and limiting scalability and cross-graph generalization. To bridge this gap, we present OMG-VLM (One Model, Many Graphs with Vision-Language Models), a unified framework for learning over attributed graphs across heterogeneous modality schemas. OMG-VLM leverages a pretrained VLM as a shared backbone and introduces structure-aware graph adapters that integrate neighborhood information while remaining compatible with the VLM's native embedding space. This design enables effective learning over text-attributed, image-attributed, and multi-attributed graphs within a single model. Extensive experiments across diverse domains show that OMG-VLM consistently outperforms state-of-the-art GNN- and LLM-based baselines on attributed graph learning tasks such as node classification and link prediction, while exhibiting strong generalization to unseen graphs and varying modality schemas. The source code is available at https://github.com/Jo-eyang/OMG-VLM.

Figures

Figures reproduced from arXiv: 2607.19128 by Jiajin Liu, Jiayi Yang, Qiaoyu Tan, Yifang Chen, Yuanfu Sun.

Figure 1
Figure 1. Figure 1: Heterogeneous modalities over attributed [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the OMG-VLM framework. OMG-VLM uses a pretrained VLM as a shared backbone [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Effectiveness of graph adapters. Bars com [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 6
Figure 6. Figure 6: Training-strategy ablation. Bars show accu [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 5
Figure 5. Figure 5: Effect of per-neighbor compression as the [PITH_FULL_IMAGE:figures/full_fig_p008_5.png] view at source ↗
Figure 7
Figure 7. Figure 7: Sensitivity to the number of visual aggregation [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Qualitative example for MAG Amazon-Arts node classification. Full text information is omitted for [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Qualitative example for TAG ogbn-arXiv node classification. Full text information is omitted for brevity. [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Qualitative example for IAG RedditS link prediction. [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

49 extracted references · 6 canonical work pages · 4 internal anchors

  1. [1]

    Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V

    When graph meets multimodal: benchmarking and meditating on multimodal attributed graph learning , author=. Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V. 2 , pages=

  2. [2]

    arXiv preprint arXiv:2407.12860 , year=

    Stage: Simplified text-attributed graph embeddings using pre-trained llms , author=. arXiv preprint arXiv:2407.12860 , year=

    Pith/arXiv arXiv

  3. [3]

    Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

    Mosaic of modalities: A comprehensive benchmark for multimodal graph learning , author=. Proceedings of the Computer Vision and Pattern Recognition Conference , pages=

  4. [4]

    ACM SIGKDD Explorations Newsletter , volume=

    Exploring the potential of large language models (llms) in learning on graphs , author=. ACM SIGKDD Explorations Newsletter , volume=. 2024 , publisher=

    2024

  5. [5]

    ACM Transactions on Intelligent Systems and Technology , volume=

    Graph machine learning in the era of large language models (llms) , author=. ACM Transactions on Intelligent Systems and Technology , volume=. 2025 , publisher=

    2025

  6. [6]

    LLaVA-NeXT: Improved reasoning, OCR, and world knowledge , url=

    Liu, Haotian and Li, Chunyuan and Li, Yuheng and Li, Bo and Zhang, Yuanhan and Shen, Sheng and Lee, Yong Jae , month=. LLaVA-NeXT: Improved reasoning, OCR, and world knowledge , url=

  7. [7]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    Mario: Multimodal graph reasoning with large language models , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  8. [8]

    Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

    GraphVLM: Benchmarking Vision Language Models for Multimodal Graph Learning , author=. Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition , pages=

  9. [9]

    Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 , year=

    MLaGA: Multimodal Large Language and Graph Assistant , author=. Proceedings of the 32nd ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2 , year=

  10. [10]

    arXiv.org , author =

  11. [11]

    Sun, Yuanfu and Li, Kang and Guo, Pengkang and Liu, Jiajin and Tan, Qiaoyu , year =. Mario:. doi:10.48550/ARXIV.2603.05181 , abstract =

    doi:10.48550/arxiv.2603.05181

  12. [12]

    Bai, Shuai and Cai, Yuxuan and Chen, Ruizhe and Chen, Keqin and Chen, Xionghui and Cheng, Zesen and Deng, Lianghao and Ding, Wei and Gao, Chang and Ge, Chunjiang and Ge, Wenbin and Guo, Zhifang and Huang, Qidong and Huang, Jie and Huang, Fei and Hui, Binyuan and Jiang, Shutong and Li, Zhaohai and Li, Mingsheng and Li, Mei and Li, Kaixin and Lin, Zicheng a...

    Pith/arXiv arXiv doi:10.48550/arxiv.2511.21631

  13. [13]

    IEEE Potentials , author =

    Feed-forward neural networks , volume =. IEEE Potentials , author =. 1994 , keywords =. doi:10.1109/45.329294 , abstract =

    doi:10.1109/45.329294 1994

  14. [14]

    and Hashimoto, Tatsunori and Vinyals, Oriol and Liang, Percy and Dean, Jeff and Fedus, William , month = oct, year =

    Wei, Jason and Tay, Yi and Bommasani, Rishi and Raffel, Colin and Zoph, Barret and Borgeaud, Sebastian and Yogatama, Dani and Bosma, Maarten and Zhou, Denny and Metzler, Donald and Chi, Ed H. and Hashimoto, Tatsunori and Vinyals, Oriol and Liang, Percy and Dean, Jeff and Fedus, William , month = oct, year =. Emergent. doi:10.48550/arXiv.2206.07682 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2206.07682

  15. [15]

    Flamingo: a

    Alayrac, Jean-Baptiste and Donahue, Jeff and Luc, Pauline and Miech, Antoine and Barr, Iain and Hasson, Yana and Lenc, Karel and Mensch, Arthur and Millican, Katie and Reynolds, Malcolm and Ring, Roman and Rutherford, Eliza and Cabi, Serkan and Han, Tengda and Gong, Zhitao and Samangooei, Sina and Monteiro, Marianne and Menick, Jacob and Borgeaud, Sebasti...

    Pith/arXiv arXiv doi:10.48550/arxiv.2204.14198

  16. [16]

    Zhang, Jingyi and Huang, Jiaxing and Jin, Sheng and Lu, Shijian , month = feb, year =. Vision-. doi:10.48550/arXiv.2304.00685 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2304.00685

  17. [17]

    doi:10.48550/arXiv.2104.09864 , abstract =

    Su, Jianlin and Lu, Yu and Pan, Shengfeng and Murtadha, Ahmed and Wen, Bo and Liu, Yunfeng , month = nov, year =. doi:10.48550/arXiv.2104.09864 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2104.09864

  18. [18]

    Findings of the

    Sun, Yuanfu and Ma, Zhengnan and Fang, Yi and Ma, Jing and Tan, Qiaoyu , editor =. Findings of the. 2025 , pages =. doi:10.18653/v1/2025.findings-naacl.131 , abstract =

    work page doi:10.18653/v1/2025.findings-naacl.131 2025

  19. [19]

    Proceedings of the 2019

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina , editor =. Proceedings of the 2019. 2019 , pages =. doi:10.18653/v1/N19-1423 , abstract =

    doi:10.18653/v1/n19-1423 2019

  20. [20]

    APoLLo: Unified Adapter and Prompt Learning for Vision Language Models

    Chowdhury, Sanjoy and Nag, Sayan and Manocha, Dinesh , month = dec, year =. doi:10.48550/arXiv.2312.01564 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2312.01564

  21. [21]

    and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , month = oct, year =

    Hu, Edward J. and Shen, Yelong and Wallis, Phillip and Allen-Zhu, Zeyuan and Li, Yuanzhi and Wang, Shean and Wang, Lu and Chen, Weizhu , month = oct, year =. doi:10.48550/arXiv.2106.09685 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2106.09685

  22. [22]

    Shu, Dong and Zhao, Haiyan and Hu, Jingyu and Liu, Weiru and Payani, Ali and Cheng, Lu and Du, Mengnan , month = sep, year =. Large. doi:10.48550/arXiv.2501.01346 , abstract =

    doi:10.48550/arxiv.2501.01346

  23. [23]

    Tokenized Graph Transformer with Neighborhood Augmentation for Node Classification in Large Graphs

    Chen, Jinsong and Liu, Chang and Gao, Kaiyuan and Li, Gaichao and He, Kun , month = may, year =. Tokenized. doi:10.48550/arXiv.2305.12677 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2305.12677

  24. [24]

    Learning

    Radford, Alec and Kim, Jong Wook and Hallacy, Chris and Ramesh, Aditya and Goh, Gabriel and Agarwal, Sandhini and Sastry, Girish and Askell, Amanda and Mishkin, Pamela and Clark, Jack and Krueger, Gretchen and Sutskever, Ilya , month = feb, year =. Learning. doi:10.48550/arXiv.2103.00020 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2103.00020

  25. [25]

    doi:10.48550/arXiv.2310.13023 , abstract =

    Tang, Jiabin and Yang, Yuhao and Wei, Wei and Shi, Lei and Su, Lixin and Cheng, Suqi and Yin, Dawei and Huang, Chao , month = may, year =. doi:10.48550/arXiv.2310.13023 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2310.13023

  26. [26]

    He, Ruining and McAuley, Julian , month = apr, year =. Ups and. Proceedings of the 25th. doi:10.1145/2872427.2883037 , abstract =

    arXiv

  27. [27]

    Amazon review data , url =

  28. [28]

    Veličković, Petar and Cucurull, Guillem and Casanova, Arantxa and Romero, Adriana and Liò, Pietro and Bengio, Yoshua , month = feb, year =. Graph. doi:10.48550/arXiv.1710.10903 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1710.10903

  29. [29]

    Graph Consulting , author =

    The. Graph Consulting , author =

  30. [30]

    Training Graph Neural Networks with 1000 Layers

    Li, Guohao and Müller, Matthias and Ghanem, Bernard and Koltun, Vladlen , month = apr, year =. Training. doi:10.48550/arXiv.2106.07476 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2106.07476

  31. [31]

    Liu, Haotian and Li, Chunyuan and Wu, Qingyang and Lee, Yong Jae , month = dec, year =. Visual. doi:10.48550/arXiv.2304.08485 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2304.08485

  32. [32]

    doi:10.48550/arXiv.2301.12597 , abstract =

    Li, Junnan and Li, Dongxu and Savarese, Silvio and Hoi, Steven , month = jun, year =. doi:10.48550/arXiv.2301.12597 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2301.12597

  33. [33]

    Trans. Multi. , author =. 2023 , pages =. doi:10.1109/TMM.2021.3138298 , abstract =

    arXiv 2023

  34. [34]

    Proceedings of the 27th

    Wei, Yinwei and Wang, Xiang and Nie, Liqiang and He, Xiangnan and Hong, Richang and Chua, Tat-Seng , month = oct, year =. Proceedings of the 27th. doi:10.1145/3343031.3351034 , abstract =

    arXiv

  35. [35]

    Advances in Neural Information Processing Systems , author =

    A. Advances in Neural Information Processing Systems , author =. 2023 , pages =

    2023

  36. [36]

    Harnessing

    He, Xiaoxin and Bresson, Xavier and Laurent, Thomas and Perold, Adam and LeCun, Yann and Hooi, Bryan , month = mar, year =. Harnessing. doi:10.48550/arXiv.2305.19523 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2305.19523

  37. [37]

    , year =

    Rosenblatt, F. , year =. The perceptron:. Psychological Review , publisher =. doi:10.1037/h0042519 , abstract =

    doi:10.1037/h0042519

  38. [38]

    and Ying, Rex and Leskovec, Jure , month = sep, year =

    Hamilton, William L. and Ying, Rex and Leskovec, Jure , month = sep, year =. Inductive. doi:10.48550/arXiv.1706.02216 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1706.02216

  39. [39]

    and Welling, Max , month = feb, year =

    Kipf, Thomas N. and Welling, Max , month = feb, year =. Semi-. doi:10.48550/arXiv.1609.02907 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.1609.02907

  40. [40]

    Hu, Weihua and Fey, Matthias and Zitnik, Marinka and Dong, Yuxiao and Ren, Hongyu and Liu, Bowen and Catasta, Michele and Leskovec, Jure , month = feb, year =. Open. doi:10.48550/arXiv.2005.00687 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2005.00687 2005

  41. [41]

    Interview:

    Majumder, Bodhisattwa Prasad and Li, Shuyang and Ni, Jianmo and McAuley, Julian , editor =. Interview:. Proceedings of the 2020. 2020 , pages =. doi:10.18653/v1/2020.emnlp-main.653 , abstract =

    work page doi:10.18653/v1/2020.emnlp-main.653 2020

  42. [42]

    doi:10.48550/arXiv.2402.08170 , abstract =

    Chen, Runjin and Zhao, Tong and Jaiswal, Ajay and Shah, Neil and Wang, Zhangyang , month = apr, year =. doi:10.48550/arXiv.2402.08170 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2402.08170

  43. [43]

    doi:10.48550/arXiv.2502.00806 , abstract =

    He, Yufei and Sui, Yuan and He, Xiaoxin and Liu, Yue and Sun, Yifei and Hooi, Bryan , month = apr, year =. doi:10.48550/arXiv.2502.00806 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2502.00806

  44. [44]

    GraphPrompter: Multi-stage Adaptive Prompt Optimization for Graph In-Context Learning

    Lv, Rui and Zhang, Zaixi and Zhang, Kai and Liu, Qi and Gao, Weibo and Liu, Jiawei and Yan, Jiaxia and Yue, Linan and Yao, Fangzhou , month = may, year =. doi:10.48550/arXiv.2505.02027 , abstract =

    work page internal anchor Pith review Pith/arXiv arXiv doi:10.48550/arxiv.2505.02027

  45. [45]

    Dosovitskiy, Alexey and Beyer, Lucas and Kolesnikov, Alexander and Weissenborn, Dirk and Zhai, Xiaohua and Unterthiner, Thomas and Dehghani, Mostafa and Minderer, Matthias and Heigold, Georg and Gelly, Sylvain and Uszkoreit, Jakob and Houlsby, Neil , month = jun, year =. An. doi:10.48550/arXiv.2010.11929 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2010.11929 2010

  46. [46]

    Bai, Jinze and Bai, Shuai and Yang, Shusheng and Wang, Shijie and Tan, Sinan and Wang, Peng and Lin, Junyang and Zhou, Chang and Zhou, Jingren , month = oct, year =. Qwen-. doi:10.48550/arXiv.2308.12966 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2308.12966

  47. [47]

    doi:10.48550/arXiv.2510.16990 , abstract =

    Ning, Xuying and Fu, Dongqi and Wei, Tianxin and Xu, Wujiang and He, Jingrui , month = oct, year =. doi:10.48550/arXiv.2510.16990 , abstract =

    doi:10.48550/arxiv.2510.16990

  48. [48]

    doi:10.48550/arXiv.2506.02568 , abstract =

    Fan, Dongzhe and Fang, Yi and Liu, Jiajin and Difallah, Djellel and Tan, Qiaoyu , month = jun, year =. doi:10.48550/arXiv.2506.02568 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.02568

  49. [49]

    Liu, Jiajin and Fan, Dongzhe and Shen, Jiacheng and Ji, Chuanhao and Zha, Daochen and Tan, Qiaoyu , month = jun, year =. Graph-. doi:10.48550/arXiv.2506.10282 , abstract =

    Pith/arXiv arXiv doi:10.48550/arxiv.2506.10282