Pith. sign in

REVIEW 4 major objections 6 minor 2 cited by

Griffin: Towards a Graph-Centric Relational Database Foundation Model

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 Griffin, a graph-centric foundation model, unifies encoders and decoders so one pretrained checkpoint can handle diverse relational database tasks.

desk verdict A credible and substantial first RDB foundation model attempt whose headline average-rank claim is undercut by an internal regression-metric alignment inconsistency and missing error bars. read the letter →

arxiv 2505.05568 v2 pith:ZWWJYX6Z submitted 2025-05-08 cs.LG cs.AIcs.DB

Yanbo Wang , Xiyuan Wang , Quan Gan , Minjie Wang , Qibin Yang , David Wipf , Muhan Zhang This is my paper · ORCID
classification cs.LGcs.AIcs.DB
keywords relationaldatabasesgraphneuralnetworksfoundationmodelstabulardatapretrainingtransferlearningheterogeneousgraphsfew-shot
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

This paper aims to establish that one pretrained architecture can serve as a foundation model for relational databases (RDBs), replacing the current practice of training a separate graph neural network for each database task. Griffin converts each RDB into a heterogeneous graph whose nodes are rows and whose edges are primary-key/foreign-key links, then uses a unified text encoder for categorical and metadata cells, a pretrained float encoder for numbers, and a task embedding derived from the target column name. The paper evaluates Griffin on 24 tasks from two benchmarks covering over 150 million nodes. It reports that the architecture alone beats task-specific GNN and deep-feature-synthesis baselines in average rank, that pretraining on single tables alone improves performance further, and that supervised fine-tuning on similar or diverse relational datasets gives the largest gains in low-data settings. If these claims hold, a single checkpoint could generalize across schemas, tasks, and domains that currently demand bespoke models.

What carries the argument

The mechanism is Griffin's trio of unified components. A unified feature encoder sends every categorical, textual, and metadata value through the same pretrained text encoder, applies quantile normalization plus a pretrained float encoder/decoder pair to numerical values, and creates a task embedding by encoding the target column's name. The MPNN then alternates cross-attention, in which the node state queries column-name keys and cell-value values, with hierarchical aggregation that averages neighbors within each relation type and takes the max across types. Finally, a unified task decoder predicts classification labels by dot-product similarity to text-embedded label names, and regression targets through the pretrained number decoder.

What would settle it

Rerun all RelBench regression tasks with the original raw-scale targets and the original preprocessing for every baseline, denormalizing Griffin's outputs before computing MAE; if Griffin no longer holds the best average rank, the 'superior or comparable' claim depends on the normalized metric.

Watch

Extended reading notes

Core claim

The central discovery, stated on the paper's own terms, is that relational database prediction can be unified under one graph-centric model without losing to task-specific models. The paper's key evidence is that Griffin-unpretrained, the architecture without any pretraining, already has the best average rank across all compared models, and that Griffin-pretrained improves on it. Pretraining is done in two stages: completion pretraining on over 200 single-table datasets with a masked-cell cosine objective, and joint supervised fine-tuning on selected single-table and RDB tasks. Downstream, Griffin shows strong transferability when the supervised fine-tuning domain is similar to the target or more diverse than it, with the largest benefits in 512-sample few-shot settings.

Load-bearing premise

The central comparison assumes that predicting quantile-normalized target values and reporting those normalized errors is the right yardstick for regression, even though the paper's own alignment table shows original RelBench regression errors on a very different scale, for example study-adverse MAE of 44.473 versus 2.199 after alignment.

Editorial extensions

If this is right

  • A single Griffin checkpoint pretrained only on single-table datasets can transfer to relational database tasks, so foundation-model pretraining does not require a matching target RDB.
  • The architecture alone, without any pretraining, already improves average rank over task-specific GNN and deep-feature-synthesis baselines, so the reported gains are not solely an artifact of pretraining.
  • Joint supervised fine-tuning on relational datasets that are either similar to the downstream domain or more diverse than it produces the largest transfer gains, with the strongest effect in low-data settings with only hundreds of training samples.
  • Because classification uses text-embedded label names and regression shares a single float decoder, the same model can be pointed at a new task by changing only the task embedding and the label embeddings.

Reading between the lines

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

  • Editorial inference: because completion pretraining needs no labels, scaling the pretraining corpus with additional unlabeled relational databases should improve generalization further, following the same logic as masked language modeling.
  • Editorial inference: the observed collapse of first-layer cross-attention to mean aggregation suggests a general design rule for column-graph models: run self-attention over cells before conditioning queries on the task.
  • Editorial inference: before deploying Griffin on regression tasks, a practitioner should calibrate the denormalizer per column, because the reported normalized-scale errors are not directly comparable to raw-scale business metrics.
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. The paper proposes Griffin, a graph-centric foundation model for relational databases. An RDB is converted into a heterogeneous temporal graph with rows as nodes and PK-FK relationships as edges. Griffin encodes cells with a frozen text encoder (Nomic) for categorical/text values and a pretrained float encoder for numerical values, augments nodes and edges with metadata embeddings, injects a task embedding derived from the target column name, and processes the sampled rooted subgraph with an MPNN that combines a cross-attention module and hierarchical (per-relation mean, cross-relation max) aggregation. A shared decoder handles classification via inner products with label text embeddings and regression via a pretrained number decoder. Training is multi-stage: completion pretraining on roughly 200 single-table datasets, joint supervised fine-tuning on single-table and/or RDB datasets, and per-task fine-tuning. Experiments span 24 tasks from 4DBInfer and RelBench (over 150 million nodes), and the paper reports three main findings: (Q1) Griffin-unpretrained achieves the best average rank against four GNN and four DFS-based baselines; (Q2) single-table-only pretraining improves average rank; (Q3) transfer benefits correlate with the similarity and diversity of the SFT domain. Code is released at github.com/yanxwb/Griffin.

Significance. If the main claims held as stated, the paper would constitute a meaningful step toward relational-database foundation models: a single pretrained architecture with unified encoders and decoders could replace task-specific GNNs on RDB prediction tasks, and the single-table-to-RDB multi-stage pretraining recipe would be a reusable template. Strengths of the manuscript include the release of code; a large pretraining corpus (over 150 million nodes) assembled from public benchmarks; a self-contained, testable architecture consisting of cross-attention over cells and hierarchical aggregation; the coherence of the pretraining objective in Eq. (6), which uses a fixed text encoder as a target rather than a fitted circular target; and the transferability study, which averages over five random split selections. Two load-bearing issues prevent acceptance of the strongest claims at present: the regression metrics in Table 4 appear to mix incompatible target scales (quantile-normalized versus original raw scale), and the 'only architecture' comparison is confounded by Griffin's different input encoders.

major comments (4)
  1. [§B.2–B.3, Tables 3–4, Figure 2] The regression comparisons that feed the headline average-rank claim are computed on incompatible target scales. Section B.2 states that RelBench regression tasks were redefined via quantile normalization, and Table 3 shows the effect: the rel-trial/study-adverse MAE changes from 44.473 (RelBench original) to 2.199 (aligned), a change of roughly an order of magnitude. Section B.3 then states that for the RelBench datasets, 'the Sage results are taken from the original report' while all other baselines were re-evaluated through the 4DBInfer pipeline. Consequently, in Table 4's regression block Griffin's MAE values (normalized scale) are compared against SAGE values that, if the text is literal, are on the original raw scale; the regression wins (e.g., rel-trial/study-adverse 1.1700 vs. 1.5251 for SAGE) would then be scale artifacts rather than evidence of superiority. The numbers are also internally inconsistent: Table 3 reports both 44.473 and 2.199 for the same task, while Table 4 reports 1.5251 for SAGE, matching neither. Because the MAE/RMSE tasks contribute to the average rank in Figure 2 and to the Section 6.2 claim that 'Griffin-unpretrained outperforms all other models in average rank,' the comparability of the regression metrics must be established before the main claim can be accepted. Please re-evaluate every baseline, including SAGE, under a single explicitly stated target scale (or report both scales for both Griffin and all baselines).
  2. [§4.3, §B.1 (Table 2)] The claim that Griffin-unpretrained 'differs from other GNN baselines only in its architectural design' is confounded by the input encoders. Griffin encodes categorical and text cells with Nomic sentence embeddings, encodes numerical cells with a pretrained float encoder, and additionally injects metadata embeddings (table names, column names, edge types) and a task embedding; the GNN baselines SAGE/GAT/PNA/HGT use one-hot categorical encoding and GloVe text embeddings (Table 2). The Q1 comparison therefore varies architecture and input representation simultaneously, so the superior average rank in Figure 2 cannot be attributed to the cross-attention and hierarchical-aggregation design alone. To support the 'only architecture' claim, add an ablation in which the standard MPNN backbones are given Griffin's encoders (or, equivalently, Griffin's MPNN is given the baselines' one-hot inputs), and qualify the claim in Section 4.3 accordingly.
  3. [§6.2, Tables 4–5] The main results and the ablation are reported without variance or the number of seeds, despite several very small margins: for example, Seznam/charge accuracy is 0.7998 for Griffin-unpretrained versus 0.8053 for GAT, Outbrain/ctr ROC-AUC is 0.6246 versus 0.6239 for SAGE, and in the ablation Virus/wnv is 0.6680 for Griffin versus 0.6648 for Griffin-mean-GNN. The statements in Section 6.2 that Griffin 'demonstrates significant improvements' and that removing the components 'results in a significant performance drop' are not substantiated without repeated runs, error bars, or a significance test; the transfer study in Figure 4 uses five seeds for split selection, so the infrastructure for this exists. Please report means and standard deviations (or per-seed values) over at least three seeds for the main tables, or weaken the 'significant' claims proportionately.
  4. [§6.3, Table 4] The Q2 claim that 'Griffin-Pretrained outperforms its non-pretrained counterpart, validating the universal benefits of pretraining' is contradicted by Table 4 on roughly one third of the tasks: pretraining degrades results on Rel-avito/user-clicks (0.6330 vs. 0.6639), Rel-f1/DNF (0.7091 vs. 0.7052), Rel-f1/top3 (0.7795 vs. 0.7855), Rel-hm/user-churn (0.6804 vs. 0.6847), Rel-f1/position (0.5694 vs. 0.5586), Rel-hm/item-sales (0.8962 vs. 0.8879), Rel-trial/site-success (0.7945 vs. 0.7926), and Rel-trial/study-adverse (1.2148 vs. 1.1700). The average-rank improvement (3.042 vs. 3.708) is real but is not 'universal'; the text should be reworded to claim a net average improvement, and the tasks where pretraining hurts should be discussed, particularly since four of the eight are regression tasks affected by the scale-alignment issue raised above.
minor comments (6)
  1. [Abstract and §1] The phrase 'foundation model attemptation' appears in the abstract and again as 'Graph-centric RelatIonal databaseFoundatIoNmodel attemptation' in Section 1; this is awkward and should be replaced with a standard term such as 'effort' or 'approach.'
  2. [Figure 2 caption] The caption's final sentence, 'All values are positive; higher values indicate better performance for Accuracy and ROC-AUC, while lower values are better for left ones,' is garbled and should state the direction of better performance for each panel explicitly.
  3. [Table 5] Table 5 reports MAE and Logloss values with negative signs (e.g., -0.6593, -0.5684) without explanation; the sign convention (apparently negating losses so that higher is better uniformly) should be stated in the caption.
  4. [§B.2, Table 3] The row label 'Aligned to Griffin Results' is ambiguous about which model produced those numbers; specify the model and state explicitly that the alignment changes the target scale, not merely preprocessing details, so that readers can track which Table 4 entries are comparable.
  5. [§6.4] The sentence 'For transfering between Others-1 and Others-2' contains a typo, and the one-directional transfer benefit (Others-1 to Others-2 but not vice versa) is asserted from figures without reported variance; either add error bars to the transfer figures or soften the asymmetry claim.
  6. [§6.1 and §A] The selection of the 50 labeled single-table datasets used for joint SFT and the assignment of tasks into the Commerce-1/2 and Others-1/2 groups are described qualitatively; provide the full task-to-group mapping and the dataset selection criterion in the appendix to support replication.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity found; the derivation is self-contained, with the main caveat being a benchmark-comparability issue in regression metrics rather than a circular step.

full rationale

The central derivation chain is self-contained. The completion-pretraining loss (Eq. 6) predicts the fixed text-encoder embedding of a masked cell from the remaining row; the same fixed encoder supplies classification label embeddings at inference (Eq. 5), so the target and head are the same external representation, not a fitted function of the output. The float encoder/decoder (Eq. 1) are pretrained on synthetic normal samples and frozen, so regression outputs are decoded by a fixed map plus a denormalization step; no fitted parameter from downstream evaluation is renamed as a prediction. The claim that Griffin-unpretrained outperforms baselines rests on external benchmark numbers from 4DBInfer and RelBench; citation of those benchmarks by overlapping authors is normal and not load-bearing here because the empirical numbers are externally generated and checkable. The one genuine concern is a comparability one, not circularity: Appendix B.2 redefines RelBench regression targets via quantile normalization and Table 3 shows large scale shifts (e.g., rel-trial/study-adverse MAE 44.473 -> 2.199), while B.3 says SAGE numbers are 'taken from the original report'; if Table 4's SAGE regression entries are raw-scale while Griffin's are normalized-scale, the regression rank comparisons are invalid. This is a threat to the headline empirical claim and to fair benchmarking, but it does not make any derivation equivalent to its inputs by construction. Similarly, 'Griffin-unpretrained' still uses externally pretrained text and float encoders, so 'no exposure to external data' is imprecise, but again this is a labeling/fairness issue, not circularity.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The paper's central claim rests on domain assumptions about graph construction, temporal causality, the semantic quality of Nomic embeddings, and the representativeness of pretraining data. The hyperparameters listed above are hand-chosen settings that affect the results but are not fitted to the downstream targets.

free parameters (4)
  • MPNN depth = 4 layers
    Number of message passing layers fixed across all experiments (Appendix C.2).
  • sampling fanout = 20 per layer
    Temporal neighbor sampling fanout, fixed across tasks (Appendix C.2).
  • hidden dimension = 512
    Model width, uniform across components (Appendix C.2).
  • limited-sample SFT size = 4096 samples
    Cutoff for the low-data transfer experiments (Appendix D.1).
assumptions (5)
  • domain assumption Rows as nodes and PK-FK pairs as edges preserve task-relevant information for the predictive tasks considered.
    The whole graph-centric formulation in Section 2 depends on this equivalence.
  • domain assumption Temporal causality: only rows with earlier timestamps may be used to predict a target row.
    Defines the rooted temporal subgraph in Section 2 and excludes future information by construction.
  • domain assumption Nomic text embeddings provide a shared semantic space over column names, category values, and labels across different databases.
    The unified encoder and the classification head of Section 3.3 both rely on this space.
  • domain assumption Quantile normalization makes numeric columns from all tasks comparable through a single pretrained float encoder.
    Numerical processing in Section 3.1 and baseline alignment in Appendix B.2.
  • domain assumption Pretraining datasets are sufficiently representative for transfer to the downstream RDB tasks.
    The foundation model and transfer claims in Section 4 rest on this.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Griffin: Towards a Graph-Centric Relational Database Foundation Model." pith.science (2026). https://pith.science/paper/ZWWJYX6Z

@misc{pith2026250505568,
  author       = {Pith},
  title        = {Pith review of: Griffin: Towards a Graph-Centric Relational Database Foundation Model},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/ZWWJYX6Z}},
  note         = {Machine review of arXiv:2505.05568}
}
read the original abstract

We introduce Griffin, the first foundation model attemptation designed specifically for Relational Databases (RDBs). Unlike previous smaller models focused on single RDB tasks, Griffin unifies the data encoder and task decoder to handle diverse tasks. Additionally, we enhance the architecture by incorporating a cross-attention module and a novel aggregator. Griffin utilizes pretraining on both single-table and RDB datasets, employing advanced encoders for categorical, numerical, and metadata features, along with innovative components such as cross-attention modules and enhanced message-passing neural networks (MPNNs) to capture the complexities of relational data. Evaluated on large-scale, heterogeneous, and temporal graphs extracted from RDBs across various domains (spanning over 150 million nodes), Griffin demonstrates superior or comparable performance to individually trained models, excels in low-data scenarios, and shows strong transferability with similarity and diversity in pretraining across new datasets and tasks, highlighting its potential as a universally applicable foundation model for RDBs. Code available at https://github.com/yanxwb/Griffin.

Figures

Figures reproduced from arXiv: 2505.05568 by the authors.

Figure 1
Figure 1. Overview of the Griffin Model Framework. The framework first transforms RDBs into a graph structure by representing each row as a node and using primary key–foreign key relationships as edges. Given a target column, a temporally constrained subgraph is sampled and processed using a unified encoder module before being passed to a MPNN. Finally, the unified task decoders generate predictions based on whether the task … view at source ↗
Figure 2
Figure 2. Performance Comparison of Fully Fine-Tuned Models on Individual Tasks. This figure compares the performance of four GNN baselines, four single-table baselines with DFS, and two Griffin variants, each fine-tuned on individual tasks. The leftmost subfigure presents the average rank across all tasks. The remaining subfigures group tasks by evaluation metric, with results averaged accordingly. All values are positive; h… view at source ↗
Figure 3
Figure 3. Ablation Study on Different Model Design Choices. This figure compares the performance of Griffin￾unpretrained and two ablated variants, with cross-attention and max-aggregation removed, respectively. The leftmost subfigure presents the average rank across all tasks. The remaining subfigures group tasks by evaluation metric, with results averaged accordingly. All values are positive; higher values indicate better pe… view at source ↗
Figures from the paper (11 more)
Figure 4
Figure 4. Figure 4: Evaluating Transferability Across Different SFT Domains. This figure compares the impact of different SFT strategies on transferability. Each subfigure presents four models: a no-pretraining baseline and three models pretrained on single-table data followed by SFT on d…
Figure 5
Figure 5. Figure 5: Histogram of row counts of TPBerta 0-1 1-2 2-3 3-4 4-5 5-6 6-7 7-8 8-9 9-10 10+ 0 10 20 24 15 3 2 2 0 2 1 0 0 2 Row Count Ranges (×104 ) Frequency Histogram of Row Counts of CARTE [PITH_FULL_IMAGE:figures/full_fig_p013_5.png]
Figure 6
Figure 6. Figure 6: , while the column count distribution is shown in [PITH_FULL_IMAGE:figures/full_fig_p013_6.png]
Figure 7
Figure 7. Figure 7: Histogram of column counts of TPBerta 0-5 5-10 10-15 15-20 20-25 25-30 30-35 35-40 40-45 0 10 20 0 26 17 6 1 1 0 0 0 Column Count Ranges Frequency Histogram of Column Counts of CARTE [PITH_FULL_IMAGE:figures/full_fig_p014_7.png]
Figure 8
Figure 8. Figure 8: Histogram of column counts of CARTE Dataset Tables Columns Rows Seznam 4 14 2681983 Airbnb 4 34 10800000 Amazon 3 15 24291489 Diginetica 5 28 3672396 Outbrain 8 31 4778954 Retailrocket 3 11 23033676 Stackexchange 7 49 5399818 Virus 3 38 145000 Telstra 5 12 136000 Talki…
Figure 9
Figure 9. Figure 9: Cross-attention weight visualization across four layers. The heatmap shows that in the first layer, query and [PITH_FULL_IMAGE:figures/full_fig_p017_9.png]
Figure 10
Figure 10. Figure 10: Performance comparison of different SFT strategies. The figure presents the results for five baselines, including [PITH_FULL_IMAGE:figures/full_fig_p018_10.png]
Figure 11
Figure 11. Figure 11: Impact of domain transferability under limited-sample SFT. The figure compares transfer performance across [PITH_FULL_IMAGE:figures/full_fig_p019_11.png]
Figure 12
Figure 12. Figure 12: Impact of domain transferability under mixed SFT with single-tabular datasets. The figure evaluates whether [PITH_FULL_IMAGE:figures/full_fig_p019_12.png]
Figure 13
Figure 13. Figure 13: Critical difference ranking of domain transferability across different SFT strategies. The three subfigures [PITH_FULL_IMAGE:figures/full_fig_p020_13.png]
Figure 14
Figure 14. Figure 14: Evaluating Few-shot Performance: Comparison Between Griffin and TabPFNv2. This figure compares the [PITH_FULL_IMAGE:figures/full_fig_p020_14.png]

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. Incremental Evaluation and Training in Relational Deep Learning

    cs.LG 2026-08 conditional novelty 5.0 of 10

    Relational deep learning models can be evaluated and updated incrementally over time, and incremental fine-tuning matches or beats from-scratch retraining on most standard benchmark tasks.

  2. PluRel-to-RDB-PFN: Schema-Guided Synthetic Relational Pretraining

    cs.LG 2026-07 conditional novelty 5.0 of 10

    External synthetic relational data from PluRel, when curated with a real-world-schema-first curriculum, recovers 87.6-93.8% of RDB-PFN's performance using ~33K tasks instead of ~1.8M.

Reference graph

Works this paper leans on

43 extracted references · 6 canonical work pages · cited by 2 Pith papers

  1. [1]

    14 Griffin: Towards a Graph-Centric Relational Database Foundation Model B

    13 Griffin: Towards a Graph-Centric Relational Database Foundation Model 0-5 5-10 10-15 15-20 20-25 25-30 30-35 35-40 40-450 20 40 60 4 59 59 29 16 1 0 1 1 Column Count Ranges Frequency Histogram of Column Counts of TPBerta Figure 7: Histogram of column counts of TPBerta 0-5 5-10 10-15 15-20 20-25 25-30 30-35 35-40 40-450 10 20 0 26 17 6 1 1 0 0 0 Column ...

    work page 2014

  2. [8]

    The impact of data set similarity and diver- sity on transfer learning success in time series forecasting

    Ehrig, C., Sonnleitner, B., Neumann, U., Cleophas, C., and Forestier, G. The impact of data set similarity and diver- sity on transfer learning success in time series forecasting. arXiv preprint arXiv:2404.06198,

    arXiv

  3. [11]

    Tabgnn: Multiplex graph neural network for tabular data prediction.arXiv preprint arXiv:2108.09127,

    Guo, X., Quan, Y ., Zhao, H., Yao, Q., Li, Y ., and Tu, W. Tabgnn: Multiplex graph neural network for tabular data prediction.arXiv preprint arXiv:2108.09127,

    arXiv

  4. [12]

    Harnessing explanations: Llm-to-lm interpreter for enhanced text-attributed graph representation learning

    He, X., Bresson, X., Laurent, T., Perold, A., LeCun, Y ., and Hooi, B. Harnessing explanations: Llm-to-lm interpreter for enhanced text-attributed graph representation learning. arXiv preprint arXiv:2305.19523,

    arXiv

  5. [13]

    and Hooi, B

    He, Y . and Hooi, B. Unigraph: Learning a cross-domain graph foundation model from natural language.arXiv preprint arXiv:2402.13630,

    arXiv

  6. [14]

    K., M¨uller, T., Piccinno, F., and Eisen- schlos, J

    Herzig, J., Nowak, P. K., M¨uller, T., Piccinno, F., and Eisen- schlos, J. M. Tapas: Weakly supervised table parsing via pre-training.arXiv preprint arXiv:2004.02349,

    arXiv 2004

  7. [15]

    Tabpfn: A transformer that solves small tabu- lar classification problems in a second.arXiv preprint arXiv:2207.01848,

    Hollmann, N., M ¨uller, S., Eggensperger, K., and Hut- ter, F. Tabpfn: A transformer that solves small tabu- lar classification problems in a second.arXiv preprint arXiv:2207.01848,

    arXiv

  8. [16]

    Pytorch frame: A modular framework for multi-modal tabular learning

    Hu, W., Yuan, Y ., Zhang, Z., Nitta, A., Cao, K., Kocijan, V ., Sunil, J., Leskovec, J., and Fey, M. Pytorch frame: A modular framework for multi-modal tabular learning. arXiv preprint arXiv:2404.00776,

    arXiv

Show all 43 references
  1. [17]

    Tab- transformer: Tabular data modeling using contextual em- beddings.arXiv preprint arXiv:2012.06678,

    Huang, X., Khetan, A., Cvitkovic, M., and Karnin, Z. Tab- transformer: Tabular data modeling using contextual em- beddings.arXiv preprint arXiv:2012.06678,

    2012 arXiv

  2. [18]

    Om- nitab: Pretraining with natural and synthetic data for few-shot table-based question answering.arXiv preprint arXiv:2207.03637,

    Jiang, Z., Mao, Y ., He, P., Neubig, G., and Chen, W. Om- nitab: Pretraining with natural and synthetic data for few-shot table-based question answering.arXiv preprint arXiv:2207.03637,

    arXiv

  3. [19]

    Kanter, J. M. and Veeramachaneni, K. Deep feature syn- thesis: Towards automating data science endeavors. In 2015 IEEE international conference on data science and advanced analytics (DSAA), pp. 1–10. IEEE,

    2015

  4. [21]

    Spider 2.0: Evaluat- ing language models on real-world enterprise text-to-sql workflows.arXiv preprint arXiv:2411.07763,

    Lei, F., Chen, J., Ye, Y ., Cao, R., Shin, D., Su, H., Suo, Z., Gao, H., Hu, W., Yin, P., et al. Spider 2.0: Evaluat- ing language models on real-world enterprise text-to-sql workflows.arXiv preprint arXiv:2411.07763,

    arXiv

  5. [22]

    One for all: Towards training one graph model for all classification tasks.arXiv preprint arXiv:2310.00149,

    Liu, H., Feng, J., Kong, L., Liang, N., Tao, D., Chen, Y ., and Zhang, M. One for all: Towards training one graph model for all classification tasks.arXiv preprint arXiv:2310.00149,

    arXiv

  6. [23]

    Tapex: Table pre-training via learning a neural sql executor.arXiv preprint arXiv:2107.07653,

    Liu, Q., Chen, B., Guo, J., Ziyadi, M., Lin, Z., Chen, W., and Lou, J.-G. Tapex: Table pre-training via learning a neural sql executor.arXiv preprint arXiv:2107.07653,

    arXiv

  7. [24]

    X., Duderstadt, B., and Mulyar, A

    Nussbaum, Z., Morris, J. X., Duderstadt, B., and Mulyar, A. Nomic embed: Training a reproducible long context text embedder.arXiv preprint arXiv:2402.01613,

    arXiv

  8. [25]

    Pennington, J., Socher, R., and Manning, C. D. Glove: Global vectors for word representation. InProceedings of the 2014 conference on empirical methods in natural language processing (EMNLP), pp. 1532–1543,

    2014

  9. [27]

    and Paulheim, H

    Ristoski, P. and Paulheim, H. Rdf2vec: Rdf graph em- beddings for data mining. InThe Semantic Web–ISWC 2016: 15th International Semantic Web Conference, Kobe, Japan, October 17–21, 2016, Proceedings, Part I 15, pp. 498–514. Springer,

    2016

  10. [28]

    E., Yuan, Y ., Zhang, Z., et al

    11 Griffin: Towards a Graph-Centric Relational Database Foundation Model Robinson, J., Ranjan, R., Hu, W., Huang, K., Han, J., Dobles, A., Fey, M., Lenssen, J. E., Yuan, Y ., Zhang, Z., et al. Relbench: A benchmark for deep learning on relational databases.arXiv preprint arXiv...

    arXiv

  11. [29]

    B., and Goldstein, T

    Somepalli, G., Goldblum, M., Schwarzschild, A., Bruss, C. B., and Goldstein, T. Saint: Improved neural networks for tabular data via row attention and contrastive pre- training.arXiv preprint arXiv:2106.01342,

    arXiv

  12. [30]

    Tablegpt2: A large multimodal model with tabular data integration

    Su, A., Wang, A., Ye, C., Zhou, C., Zhang, G., Chen, G., Zhu, G., Wang, H., Xu, H., Chen, H., et al. Tablegpt2: A large multimodal model with tabular data integration. arXiv preprint arXiv:2411.02059,

    arXiv

  13. [31]

    Llama: Open and efficient foundation lan- guage models.arXiv preprint arXiv:2302.13971,

    Touvron, H., Lavril, T., Izacard, G., Martinet, X., Lachaux, M.-A., Lacroix, T., Rozi`ere, B., Goyal, N., Hambro, E., Azhar, F., et al. Llama: Open and efficient foundation lan- guage models.arXiv preprint arXiv:2302.13971,

    arXiv

  14. [32]

    4dbinfer: A 4d benchmarking toolbox for graph-centric predictive mod- eling on relational dbs.arXiv preprint arXiv:2404.18209,

    Wang, M., Gan, Q., Wipf, D., Cai, Z., Li, N., Tang, J., Zhang, Y ., Zhang, Z., Mao, Z., Song, Y ., et al. 4dbinfer: A 4d benchmarking toolbox for graph-centric predictive mod- eling on relational dbs.arXiv preprint arXiv:2404.18209,

    arXiv

  15. [33]

    Tablebench: A comprehensive and complex benchmark for table ques- tion answering.arXiv preprint arXiv:2408.09174,

    Wu, X., Yang, J., Chai, L., Zhang, G., Liu, J., Du, X., Liang, D., Shu, D., Cheng, X., Sun, T., et al. Tablebench: A comprehensive and complex benchmark for table ques- tion answering.arXiv preprint arXiv:2408.09174,

    arXiv

  16. [34]

    Z., Sun, J., Wu, J., and Chen, J

    Yan, J., Zheng, B., Xu, H., Zhu, Y ., Chen, D. Z., Sun, J., Wu, J., and Chen, J. Making pre-trained language models great on tabular prediction.arXiv preprint arXiv:2403.01841,

    arXiv

  17. [35]

    Tabert: Pre- training for joint understanding of textual and tabular data

    Yin, P., Neubig, G., Yih, W.-t., and Riedel, S. Tabert: Pre- training for joint understanding of textual and tabular data. arXiv preprint arXiv:2005.08314,

    2005 arXiv

  18. [36]

    Spider: A large-scale human-labeled dataset for complex and cross-domain semantic parsing and text-to-sql task.arXiv preprint arXiv:1809.08887,

    Yu, T., Zhang, R., Yang, K., Yasunaga, M., Wang, D., Li, Z., Ma, J., Li, I., Yao, Q., Roman, S., et al. Spider: A large-scale human-labeled dataset for complex and cross-domain semantic parsing and text-to-sql task.arXiv preprint arXiv:1809.08887,

    arXiv

  19. [37]

    Florence: A new foundation model for computer vision.arXiv preprint arXiv:2111.11432,

    Yuan, L., Chen, D., Chen, Y .-L., Codella, N., Dai, X., Gao, J., Hu, H., Huang, X., Li, B., Li, C., et al. Florence: A new foundation model for computer vision.arXiv preprint arXiv:2111.11432,

    arXiv

  20. [38]

    Contextgnn: Beyond two-tower recommendation systems

    Yuan, Y ., Zhang, Z., He, X., Nitta, A., Hu, W., Wang, D., Shah, M., Huang, S., Stojanoviˇc, B., Krumholz, A., et al. Contextgnn: Beyond two-tower recommendation systems. arXiv preprint arXiv:2411.19513,

    arXiv

  21. [39]

    Gfs: Graph- based feature synthesis for prediction over relational database.Proceedings of the VLDB Endowment

    Zhang, H., Gan, Q., Wipf, D., and Zhang, W. Gfs: Graph- based feature synthesis for prediction over relational database.Proceedings of the VLDB Endowment. ISSN, 2150:8097, 2023a. Zhang, H., Wen, X., Zheng, S., Xu, W., and Bian, J. Towards foundation models for learning on tabu...

    arXiv

  22. [40]

    Graphany: A foundation model for node clas- sification on any graph.arXiv preprint arXiv:2405.20445,

    Zhao, J., Mostafa, H., Galkin, M., Bronstein, M., Zhu, Z., and Tang, J. Graphany: A foundation model for node clas- sification on any graph.arXiv preprint arXiv:2405.20445,

    arXiv

  23. [41]

    Xtab: Cross-table pretraining for tabular transformers.arXiv preprint arXiv:2305.06090,

    Zhu, B., Shi, X., Erickson, N., Li, M., Karypis, G., and Shoaran, M. Xtab: Cross-table pretraining for tabular transformers.arXiv preprint arXiv:2305.06090,

    arXiv

  24. [42]

    The detailed information is shown in Table

    and RelBench (Robinson et al., 2024), covering a wide range of domains, scales, and tasks. The detailed information is shown in Table

    2024

  25. [2003]

    Brown, T. B. Language models are few-shot learners.arXiv preprint arXiv:2005.14165,

    2005 arXiv

  26. [2016]

    Relgnn: Com- posite message passing for relational deep learning.arXiv preprint arXiv:2502.06784,

    Chen, T., Kanatsoulis, C., and Leskovec, J. Relgnn: Com- posite message passing for relational deep learning.arXiv preprint arXiv:2502.06784,

    arXiv

  27. [2017]

    J., Grinsztajn, L., and Varoquaux, G

    Kim, M. J., Grinsztajn, L., and Varoquaux, G. Carte: pre- training and transfer for tabular learning.arXiv preprint arXiv:2402.16785,

    arXiv

  28. [2018]

    Tqa-bench: Evaluating llms for multi-table question answering with scalable context and symbolic extension.arXiv preprint arXiv:2411.19504,

    Qiu, Z., Peng, Y ., He, G., Yuan, B., and Wang, C. Tqa-bench: Evaluating llms for multi-table question answering with scalable context and symbolic extension.arXiv preprint arXiv:2411.19504,

    arXiv

  29. [2019]

    C., Kraska, T., and Karger, D

    Chepurko, N., Marcus, R., Zgraggen, E., Fernandez, R. C., Kraska, T., and Karger, D. Arda: automatic relational data augmentation for machine learning.arXiv preprint arXiv:2003.09758,

    2003 arXiv

  30. [2020]

    Cre- ating embeddings of heterogeneous relational datasets for data integration tasks

    Cappuzzo, R., Papotti, P., and Thirumuruganathan, S. Cre- ating embeddings of heterogeneous relational datasets for data integration tasks. InProceedings of the 2020 ACM SIGMOD international conference on management of data, pp. 1335–1349,

    2020

  31. [2021]

    Atj-net: Auto-table-join network for automatic learning on relational databases

    Bai, J., Wang, J., Li, Z., Ding, D., Zhang, J., and Gao, J. Atj-net: Auto-table-join network for automatic learning on relational databases. InProceedings of the Web Con- ference 2021, pp. 1540–1551,

    2021

  32. [2022]

    Tabr: Tabular deep learning meets nearest neighbors in 2023.arXiv preprint arXiv:2307.14338,

    10 Griffin: Towards a Graph-Centric Relational Database Foundation Model Gorishniy, Y ., Rubachev, I., Kartashev, N., Shlenskii, D., Kotelnikov, A., and Babenko, A. Tabr: Tabular deep learning meets nearest neighbors in 2023.arXiv preprint arXiv:2307.14338,

    2023 arXiv

  33. [2023]

    Supervised learning on relational databases with graph neural networks.arXiv preprint arXiv:2002.02046,

    Cvitkovic, M. Supervised learning on relational databases with graph neural networks.arXiv preprint arXiv:2002.02046,

    2002 arXiv

  34. [2024]

    E., Ranjan, R., Robinson, J., Ying, R., You, J., and Leskovec, J

    Fey, M., Hu, W., Huang, K., Lenssen, J. E., Ranjan, R., Robinson, J., Ying, R., You, J., and Leskovec, J. Rela- tional deep learning: Graph representation learning on relational databases.arXiv preprint arXiv:2312.04615,

    arXiv

  35. [2025]

    Chen, W., Wang, H., Chen, J., Zhang, Y ., Wang, H., Li, S., Zhou, X., and Wang, W. Y . Tabfact: A large-scale dataset for table-based fact verification.arXiv preprint arXiv:1909.02164,

    1909 arXiv

Pith tools

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