Mixture of Length and Pruning Experts for Knowledge Graphs Reasoning

Enjun Du; Siyi Liu; Yongqi Zhang

REVIEW 5 major objections 5 minor 48 references

Mixture of Length and Pruning Experts for Knowledge Graphs Reasoning

T0 review · 5 major / 5 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 MoKGR adapts reasoning depth and pruning per query, beating fixed-depth GNNs on knowledge graph benchmarks.

desk verdict A plausible within-subfield MoE combination for KG reasoning, but the YAGO3-10 headline is confounded by PPR prefiltering and the efficiency table contradicts the text. read the letter →

arxiv 2507.20498 v1 pith:C7XVLYNA submitted 2025-07-28 cs.LG

Enjun Du , Siyi Liu , Yongqi Zhang This is my paper · ORCID

classification cs.LG MSC 68T07

keywords knowledgegraphreasoningmixtureofexpertspathneuralnetworksadaptivelengthpruningpersonalizedpagerankinductive

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 knowledge-graph reasoning is held back by rigid, query-agnostic path exploration: most GNN-based methods propagate messages for a fixed number of hops and prune candidate paths with a single uniform criterion. To fix this, it introduces MoKGR, a mixture-of-experts framework in which a set of length experts selects the path depth that fits each query, and a set of three pruning experts (scoring, attention, semantic similarity) jointly decides which paths to keep. The claimed payoff is two-fold: accuracy improves across transductive and inductive benchmarks, including a large gain on YAGO3-10, and computation drops because the model can stop early on short paths and explore fewer candidates. The paper's central bet is that query-specific path length and query-specific pruning, rather than any single fixed strategy, is what makes reasoning paths informative.

What carries the argument

A mixture-of-experts gating system layered on top of message-passing path encoders. Length experts assign a weight $g_q(\ell)$ to each candidate path length so the final entity score is $\Psi(e_a) = \sum_{\ell\in\mathcal{A}} g_q(\ell)\, s_\ell(q,e_a)$; pruning experts each define a scoring function $\phi^\ell_i(e_a)$ (layer-wise score, max attention over incoming edges, cosine similarity to the query relation), and entities retained at layer $\ell$ are the union over selected experts of their individual top-$K_\ell$ sets. A Gumbel-Sigmoid binary gate on each layer's message update lets the model stop early when the path already carries enough evidence, and a PPR-based pre-filtering step is used for very large graphs.

What would settle it

Run MoKGR on YAGO3-10 two ways: with its PPR subgraph pre-filtering and with a wider or exact subgraph (or the same PPR budget given to NBFNet and AdaProp). If the MRR gap over baselines shrinks or disappears, then the claimed gains are attributable to PPR retention rather than to expert-based length and pruning adaptation.

Watch

Extended reading notes

Core claim

MoKGR's central claim is that personalizing path exploration to the query improves knowledge-graph reasoning. It does this with two complementary mechanisms: length experts, which use a noisy top-k gating to select a subset of path lengths per query and weight each selected length's score; and pruning experts, which score candidate entities from three complementary views, a global scoring view, a local structural attention view, and a semantic-similarity view, then retain the union of the top entities per selected expert. A layer-wise Gumbel-Sigmoid gate encourages early termination when shorter paths suffice, and this is combined with an adaptive sampling curve that widens exploration in middle layers and narrows it later. The experiments report state-of-the-art or near-state-of-the-art numbers across Family, UMLS, WN18RR, FB15k-237, NELL-995, and YAGO3-10 in the transductive setting and across the inductive versions of WN18RR, FB15k-237, and NELL-995.

Load-bearing premise

The largest-scale results on YAGO3-10 assume the pre-computed Personalized PageRank subgraph contains the true answer entity for every test query; if PPR drops the answer, no later expert can recover it.

Editorial extensions

If this is right

If query-adaptive lengths carry the gain, then fixed-depth GNNs like NBFNet and RED-GNN are over-exploring easy queries and under-exploring hard ones; MoKGR's selective depth should dominate whenever query difficulty is heterogeneous.
If pruning experts are complementary, then the union-of-experts criterion should be able to keep high-recall path sets at smaller size than any single score, transferring to other link-prediction and subgraph-retrieval pipelines.
If PPR pre-filtering plus expert pruning handles YAGO3-10 within one 48GB GPU, then the same recipe should scale to larger KGs such as full YAGO or Wikidata slices without the per-layer blow-up of full-exploration GNNs.
The length-selection signal, query relation embedding plus local structure around the query entity, gives a reusable heuristic for when to stop path expansion, beyond this specific model.

Reading between the lines

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

One testable implication the author leaves implicit: the length experts could be interpreted as an estimator of query difficulty, so their gating weights on a new query could be used to predict which queries are already answerable from a small local subgraph, a cheap pre-filter for any downstream reasoner.
The three pruning experts bear a family resemblance to known complementary signals, global score, local attention, and embedding similarity, and the paper's ablated results suggest that each alone underperforms the mixture; that is evidence for a broader principle that path-quality scoring is multi-dimensional.
A natural falsifying experiment the paper does not run: hold the length experts fixed at a single length and let only pruning experts adapt, versus hold pruning fixed and let only length adapt, on a dataset where true answer distance varies widely like NELL-995, to see which mechanism contributes the majority of the MRR gain.
The reliance on PPR pre-filtering on YAGO3-10 means the headline efficiency numbers should not be read as the cost of the expert mechanism alone; a fair comparison would run a fixed-depth GNN inside the same PPR subgraph.

Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Desk Editor's Note

read the letter

Here's the take. MoKGR is a real, if incremental, combination: it couples hop-dependent length experts with three learned pruning experts and query-level gating. The ablations are consistent with the story, and the non-YAGO results on WN18RR, FB15k-237, and NELL-995 are plausible. That is the good part.

The soft spots are serious. First, the YAGO3-10 headline is not attributable to the expert mechanisms. Appendix D.1 and Algorithm 2 show that for ultra-large datasets the model runs inside a PPR-precomputed subgraph before any length or pruning expert is applied. No PPR recall or subgraph budget is reported. Since one-shot-subgraph is itself PPR-based, the +0.051 MRR margin could just reflect a different PPR construction or budget. Table 2 ablations on YAGO run inside the same PPR subgraph, so they cannot separate the PPR contribution from the expert contribution. This is a load-bearing confound, not a minor omission.

Second, the efficiency claim is contradicted by the paper's own numbers. Table 3b lists MoKGR inference at 358.3 versus 71.9 for one-shot-subgraph and 84.2 for AdaProp, while the text claims "significantly faster inference times." Unless the units differ, that is the opposite. That paragraph should not be in the paper as written.

Third, Appendix F's theorems are assertions, not derivations. Theorem 2 assumes the gating scores learn to encode the error they are supposed to minimize; Theorem 4 calls the compatibility score an "estimate" of mutual information without proof. The Limitations section even admits that theoretical guarantees remain challenging, so the appendix overclaims. It should be dropped or rewritten as a limitations discussion.

Finally, there is no code, no seeds, no error bars, and several key hyperparameters are unspecified. That makes the central empirical claim non-reproducible as submitted.

Who is this for? People working on path-based KG reasoning and MoE-style pruning. They will find the combination interesting and the ablations useful, if code and corrected results appear.

My recommendation: this deserves a serious referee rather than a desk reject, but only because the underlying method and the non-YAGO results are worth checking. Ask the authors to release code with a commit hash, move PPR into the main method with recall numbers, fix the efficiency table, and replace Appendix F with an honest limitations paragraph. Without those changes, I would not accept.

Referee Report

5 major / 5 minor

Summary. MoKGR proposes a mixture-of-experts framework for knowledge-graph reasoning whose two components are (1) a mixture of length experts that adaptively selects and weights reasoning-path lengths per query and (2) a mixture of pruning experts that combines scoring, attention, and semantic similarity to retain per-layer entity sets. The paper reports experiments on six transductive benchmarks and twelve inductive splits, claiming superior accuracy and efficiency over NBFNet, RED-GNN, A*Net, AdaProp, and one-shot-subgraph. Ablations on WN18RR and YAGO3-10 support the contribution of the balancing terms, the noise term, and the individual experts, and a case study on Family illustrates query-adaptive path selection.

Significance. If the accuracy and efficiency claims held, MoKGR would be a useful contribution to query-adaptive reasoning over knowledge graphs, and the expert-wise ablations are well-motivated. The paper's strengths include evaluation on standard transductive and inductive benchmarks, a thorough ablation of the proposed components, and a case study that concretely demonstrates personalized path selection. However, the attribution of the headline YAGO3-10 gains to the expert mechanisms is not established because PPR pre-filtering bounds every later decision, the reported inference numbers in Table 3b appear to contradict the efficiency claim, and the theoretical appendix does not contain valid derivations. These issues are load-bearing for the central claims and must be resolved before the results can be fully credited.

major comments (5)

[§4.2 / Appendix D.1 / Algorithm 2] The YAGO3-10 results in Table 1 cannot be attributed to MoKGR's length and pruning experts because Algorithm 2 constructs G_sub from the PPR cache before any expert mechanism is applied, and Appendix D.1 states that PPR pre-filtering is performed in advance for YAGO3-10. Thus every later score is upper-bounded by PPR recall. The paper does not report PPR recall, the subgraph budget, the PPR threshold, or the number of retained entities, and Table 2's YAGO3-10 ablations are all conducted inside the same PPR subgraph, so they cannot separate the PPR contribution from the expert contribution. Since one-shot-subgraph is itself PPR-based, the reported +0.051 MRR margin could reflect a different PPR construction or budget rather than the proposed experts. Please report PPR recall on YAGO3-10, include an ablation with and without PPR, and control the PPR budget across baselines if a comparative efficiency or accuracy claim is retained.
[§4.2 / Table 3b / Figure 3] The efficiency claim in Section 4.2 that MoKGR achieves 'significantly faster inference times' is contradicted by Table 3b, which reports MoKGR inference time of 358.3 versus 71.9 for one-shot-subgraph and 84.2 for AdaProp. Unless the units or evaluation protocol differ between the rows, the paper's own table undermines the inference-efficiency claim. Please clarify the units and protocol, provide a unified timing setup, and correct the claim if it is not supported by the measurements.
[Appendix F, Theorems 2–4] The theoretical analysis in Appendix F is not a valid derivation. Theorem 2 assumes without proof that the learned compatibility score [Q(c_q)]_l correlates with negative expected reasoning error. Theorem 3 assumes that at least one of the three pruning experts ranks every optimal-path entity in its top-K set and that expert failures are independent, which is exactly the property needing proof. Theorem 4 assumes that the compatibility score estimates mutual information between the answer and path length and cites Huang et al. 2025, an unrelated code-debugging paper. Section 6 correctly states that 'developing theoretical guarantees for the optimality of the selected paths remains challenging,' which is consistent with my assessment that these theorems should be removed or explicitly labeled as conjectures or empirical motivations rather than presented as proofs.
[Eq. (18), Eq. (24), Appendix D.1] Several hyperparameters that directly control the method's behavior are not reported: the truncation threshold T in Eq. (18), the adaptive sampling parameters K_s, K_h, K_l, l_i, and a in Eq. (24), and the PPR exploration scope in Appendix D.1. These parameters influence early stopping, exploration breadth, and candidate retention at every layer, so the reported results cannot be reproduced without them. Please include a complete hyperparameter table with final values for each dataset in Appendix A.1.
[§4.2 / Tables 1 and 6] The central empirical claim is that MoKGR consistently outperforms strong baselines, but no error bars, multiple seeds, or significance tests are reported. Several margins in Tables 1 and 6 are small relative to the reported precision, so the claim of 'exceptional performance across all benchmark datasets' is stronger than the evidence. Please report standard deviations over at least three seeds or another appropriate uncertainty measure, and temper the wording of the performance claim accordingly.

minor comments (5)

[Abstract] 'On diverse benchmark' should be 'on diverse benchmarks'.
[Eq. (3) / Table 2] Eq. (3) defines epsilon as Gaussian noise epsilon ~ N(0,1), but the ablation in Table 2 treats epsilon as a scalar (epsilon = 0 or epsilon = 0.2). Please clarify how the scalar epsilon is used in the fixed-noise setting and how the Gaussian sample is drawn in the dynamic setting.
[Throughout] Use consistent names: 'Adaprop' appears in Section 4.1 while 'AdaProp' is used elsewhere, and 'Y AGO3-10' contains inconsistent spacing in Tables 1 and 3.
[Figure 5] The captions of Figure 5 contain garbled tokens such as 'Ours/uni' and '/uni00000017/uni00000018'; these should be repaired.
[§4.4 / Figure 6] The case study states that MoKGR 'emphasizes aunt for niece queries' and 'brother, nephew, and uncle for brother queries,' but the heatmaps plot relation-type ratios and need a short explanation of how the reported emphasis is quantified.

Circularity Check

3 steps flagged · score 3.0 of 10

Central benchmark claims are externally grounded, but the Appendix F optimality, information-gain, and path-preservation theorems each assume the learned gating scores already encode the quantity they claim to derive.

self definitional [Appendix F.2, Theorem 2 (Optimality of Path Length Selection)]
"The key insight is that the compatibility score [Q(cq)]l learns to correlate with the negative expected error −E(l,q) through training. This occurs because queries that benefit more from specific path lengths will have higher accuracy when those lengths are selected, leading to lower task loss. ... Therefore, the adaptive path length selection mechanism converges to the optimal weighting that minimizes the expected reasoning error given the computational constraints."

The theorem's conclusion is that adaptive length selection minimizes expected reasoning error. The proof's only substantive step is to assert that the learned compatibility score Q(cq) 'learns to correlate' with −E(l,q); that correlation is exactly the optimality property at issue. No derivation ties the gating network's training objective to the expected error E(l,q), so the result is an assumption restated as a convergence claim.
self definitional [Appendix F.4, Theorem 4 (Information Gain of Adaptive Path Selection)]
"The compatibility score [Q(cq)]l in MoKGR can be interpreted as an estimate of the information gain I(Ea;Pl|eq,rq). By learning to assign higher weights to path lengths with higher information gain, MoKGR effectively solves the optimization problem ... Therefore, the adaptive path length selection mechanism in MoKGR maximizes the expected information gain about the answer entity while respecting computational constraints."

Maximizing a weighted sum of compatibility scores is 'maximizing information gain' only because the proof first defines Q(cq) to be an estimate of the information gain I(Ea;Pl|eq,rq). The theorem never computes or bounds mutual information; it labels the existing gating score as that quantity and then concludes the optimization objective is solved. This is a definitional renaming, not a derivation.

1 more flagged steps

self definitional [Appendix F.3, Theorem 3 (Preservation of Optimal Paths)]
"For the optimal path, at least one of the experts should rank e∗l highly. Let's say that for the best-matched expert i∗, we have ρi∗≤ϵ for some small ϵ>0. Then ... This proves that the mixture of pruning experts preserves the optimal reasoning path with high probability."

The proof assumes that the optimal path's entities are ranked highly by at least one selected pruning expert (ρi∗≤ϵ), which is the preservation property the theorem is meant to demonstrate. The subsequent union-bound algebra only propagates this assumption; without it, nothing in the scoring definitions ϕSco, ϕAtt, ϕSem implies such a bound. The word 'should' marks the assumption as a hope rather than a consequence.

full rationale

The central accuracy claims (Section 4.2, Tables 1 and 6) rest on filtered MRR/Hit@k comparisons over six standard benchmarks against externally published baselines; no fitted parameter is renamed as a prediction, and the benchmark numbers are not reconstructible from the method's inputs by construction. The YAGO3-10 story is weakened by the Appendix D.1 PPR pre-filter, which bounds attainable recall and is not ablated, and by Table 3b's inference times (MoKGR 358.3 vs one-shot-subgraph 71.9) that contradict the Section 4.2 claim of 'significantly faster inference times'; these are attribution and consistency problems, not construction-circularity. Appendix A.2's use of test-query shortest-path distances to justify the short-path inductive bias is a mild leakage concern, but it does not make the model's entity predictions equivalent to the length statistics. The genuine circularity is confined to Appendix F: Theorems 2, 3, and 4 assume the learned gating scores Q(cq), or the pruning experts' rankings, already encode the expected error, information gain, or optimal-path preservation that they claim to prove; Section 6 even admits theoretical guarantees 'remain challenging,' which corroborates that these appendix proofs are post-hoc assertions. Because the main benchmark evidence is external and self-contained, the paper does not reduce to its inputs; the circularity is partial and supplementary, so score 3.

Assumptions & free parameters 10 free parameters · 4 assumptions · 0 invented entities

The method rests on trained gating networks, hand-tuned hyperparameters, and several unverified assumptions in the theoretical appendix. No new physical or conceptual entities are postulated; the length experts and pruning experts are trained model components, not independent entities with external falsifiable handles.

free parameters (10)

k1 = tuned in (3, L-Lmin); exact values not reported
Controls how many path lengths are scored per query; a core design choice in Eqs. (3)-(5).
k2 = fixed to 2 (Appendix A.1)
Controls which pruning experts vote at each layer in Eq. (6).
Lmin = tuned from 1 to L-2; exact values not reported
Defines the start of length expert selection in Algorithm 1 and affects the gating context c_q.
L = not reported; described as inherited from AdaProp defaults
Upper bound of path exploration; central to computation and to the length expert set.
tau = tuned in (0.5, 2.5)
Sharpens or flattens expert gating in Eq. (4).
lambda1 = range 1e-2 to 1e-4
Weights the length and pruning importance losses in Eq. (10); hand-tuned.
lambda2 = range 1e-3 to 1e-5
Weights the length load balance loss in Eq. (10); hand-tuned.
T = predefined but value not reported
Controls the deterministic truncation and early stopping rule in Eq. (18).
Adaptive sampling parameters (K_s, K_h, K_l, l_i, a) = values not reported
Determine the per-layer retained entity count K_l in Eq. (24); not specified in the experiments.
PPR exploration scope = not specified
Selects how many PPR-ranked entities form the subgraph on YAGO3-10; central to the large-scale result in Appendix D.1.

assumptions (4)

standard math Robbins-Monro SGD convergence conditions (bounded below loss, Lipschitz gradients, decreasing learning rates) hold for the MoKGR objective.
Invoked in Theorem F.1; smoothness and boundedness are asserted, not verified for gating with Gumbel noise and top-K selection.
domain assumption For every evaluated query, the true answer is reachable within L hops and the optimal reasoning evidence is concentrated in lengths [Lmin, L].
Section 3.2 and Appendix A.2 motivate this with a test-set length distribution table, but it is not guaranteed for unseen queries.
ad hoc to paper At least one of the three pruning experts ranks every optimal-path entity in its top-K set, and expert failures are independent.
Theorem F.3 requires this to bound the exclusion probability; no argument connects scoring, attention, and semantic experts to top-K retention probabilities.
ad hoc to paper The learned compatibility score [Q(c_q)]_l correlates with negative expected reasoning error and with mutual information between answer and path length.
Theorems F.2 and F.4 conclude optimality from this correlation, but the correlation is exactly what training is supposed to produce, making the proof circular.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Mixture of Length and Pruning Experts for Knowledge Graphs Reasoning." pith.science (2026). https://pith.science/paper/C7XVLYNA

@misc{pith2026250720498,
  author       = {Pith},
  title        = {Pith review of: Mixture of Length and Pruning Experts for Knowledge Graphs Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/C7XVLYNA}},
  note         = {Machine review of arXiv:2507.20498}
}

read the original abstract

Knowledge Graph (KG) reasoning, which aims to infer new facts from structured knowledge repositories, plays a vital role in Natural Language Processing (NLP) systems. Its effectiveness critically depends on constructing informative and contextually relevant reasoning paths. However, existing graph neural networks (GNNs) often adopt rigid, query-agnostic path-exploration strategies, limiting their ability to adapt to diverse linguistic contexts and semantic nuances. To address these limitations, we propose \textbf{MoKGR}, a mixture-of-experts framework that personalizes path exploration through two complementary components: (1) a mixture of length experts that adaptively selects and weights candidate path lengths according to query complexity, providing query-specific reasoning depth; and (2) a mixture of pruning experts that evaluates candidate paths from a complementary perspective, retaining the most informative paths for each query. Through comprehensive experiments on diverse benchmark, MoKGR demonstrates superior performance in both transductive and inductive settings, validating the effectiveness of personalized path exploration in KGs reasoning.

Figures

Figures reproduced from arXiv: 2507.20498 by the authors.

**Figure 2.** Comparison of MoKGR with supervised state [PITH_FULL_IMAGE:figures/full_fig_p006_2.png] view at source ↗

**Figure 3.** Comparison between MoKGR and current state-of-the-art methods in YAGO3-10 dataset. ing the out-of-memory problem of full-exploration methods, and greatly improving the accuracy compared with other pruning methods. Learning Process Comparison. To comprehensively evaluate the effectiveness of MoKGR, we analyze a few SOTA methods on the YAGO3-10 dataset. We tracked both the MRR performance and computational time (tra… view at source ↗

Figures from the paper (3 more)

**Figure 5.** Figure 5: Comparison of Pruning Selection Strategy. [PITH_FULL_IMAGE:figures/full_fig_p008_5.png]

**Figure 6.** Figure 6: Heatmaps of relation type ratios in the rea [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]

**Figure 7.** Figure 7: Visualization of the transmission path on the [PITH_FULL_IMAGE:figures/full_fig_p017_7.png]

Review history (2 revisions) →

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

48 extracted references · 39 canonical work pages

[1]

Muhammad Ali, Abubakar Abid, and Parisa Kordjamshidi. 2022. Knowledge graphs: A comprehensive survey. IEEE Transactions on Knowledge and Data Engineering, 34(1):123--145

work page 2022
[2]

Ke Cheng, Jie Liu, Wei Wang, and Yizhou Sun. 2022. Rlogic: Recursive logical rule learning from knowledge graphs. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 179--189

work page 2022
[3]

R. Das, S. Dhuliawala, M. Zaheer, L. Vilnis, I. Durugkar, A. Krishnamurthy, A. Smola, and A. McCallum. 2017. Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning. In International Conference on Learning Representations (ICLR)

work page 2017
[4]

Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. 2017 a . Convolutional 2d knowledge graph embeddings. In Proceedings of the AAAI Conference on Artificial Intelligence

work page 2017
[5]

Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. 2017 b . Convolutional 2d knowledge graph embeddings. In AAAI

work page 2017
[6]

Charles Dugas, Yoshua Bengio, François B \'e lisle, Claude Nadeau, and René Garcia. 2001. Incorporating second-order functional knowledge for better option pricing. Advances in Neural Information Processing Systems (NeurIPS), 13:472--478

work page 2001
[7]

Mikhail Galkin, Xinyu Yuan, Hesham Mostafa, Jian Tang, and Zhaocheng Zhu. 2024. https://arxiv.org/abs/2310.04562 Towards foundation models for knowledge graph reasoning . In Proceedings of the International Conference on Learning Representations (ICLR). Published as a conference paper at ICLR 2024

arXiv 2024
[8]

Jinyang Huang, Xiachong Feng, Qiguang Chen, Hanjie Zhao, Zihui Cheng, Jiesong Bai, Jingxuan Zhou, Min Li, and Libo Qin. 2025. Mldebugging: Towards benchmarking code debugging across multi-library scenarios. ACL Findings

work page 2025

Show all 48 references

[9]

Eric Jang, Shixiang Gu, and Ben Poole. 2017. Categorical reparameterization with gumbel-softmax. In International Conference on Learning Representations (ICLR)

2017
[10]

Shaoxiong Ji, Shirui Pan, Erik Cambria, Pekka Marttinen, and Philip S Yu. 2021. A survey on knowledge graphs: Representation, acquisition, and applications. IEEE transactions on neural networks and learning systems, 33(2):494--514

2021
[11]

Shaoxiong Ji, Shirui Pan, Erik Cambria, Pekka Marttinen, and Philip S Yu. 2022. A survey on knowledge graphs: Representation, acquisition, and applications. IEEE Transactions on Neural Networks and Learning Systems, 33(2):494--514

2022
[12]

Michael I Jordan and Robert A Jacobs. 1994. Hierarchical mixtures of experts and the em algorithm. Neural Computation, 6(2):181--214

1994
[13]

Stanley Kok and Pedro Domingos. 2007. Statistical predicate invention. In ICML, pages 433--440

2007
[14]

Dmitry Lepikhin, HyoukJoong Lee, Yuanzhong Xu, Dehao Chen, Orhan Firat, Yanping Huang, Maxim Krikun, Noam Shazeer, and Zhifeng Chen. 2020. Gshard: Scaling giant models with conditional computation and automatic sharding. arXiv preprint arXiv:2006.16668

2020 arXiv
[15]

Jiang Li, Xiangdong Su, Fujun Zhang, and Guanglai Gao. 2024. https://github.com/dellixx/DuASE Learning low-dimensional multi-domain knowledge graph embedding via dual archimedean spirals . In Findings of the Association for Computational Linguistics: ACL 2024, pages 1982--1994...

2024
[16]

Ke Liang, Lingyuan Meng, Meng Liu, Yue Liu, Wenxuan Tu, Siwei Wang, Sihang Zhou, Xinwang Liu, Fuchun Sun, and Kunlun He. 2024. A survey of knowledge graph reasoning on graph types: Static, dynamic, and multi-modal. IEEE Transactions on Pattern Analysis and Machine Intelligence

2024
[17]

Shengchao Mai, Shen Zheng, Yunhao Yang, and Hongxia Hu. 2021. Communicative message passing for inductive relation reasoning. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 35, pages 4294--4302

2021
[18]

Zhenzhen Mai, Wenjun Wang, Xueli Liu, Xiaoyang Feng, Jun Wang, and Wenzhi Fu. 2025. A reinforcement learning approach for graph rule learning. Big Data Mining and Analytics, 8(1):31--44

2025
[19]

Christian Meilicke, Melisachew Wudage Chekol, Manuel Fink, and Heiner Stuckenschmidt. 2020. Reinforced anytime bottom up rule learning for knowledge graph completion. arXiv preprint arXiv:2004.04412

2020 arXiv
[20]

Christian Meilicke, Mathias Fink, Yanjie Wang, Daniel Ruffinelli, Rainer Gemulla, and Heiner Stuckenschmidt. 2018. Fine-grained evaluation of rule- and embedding-based systems for knowledge graph completion. In International Semantic Web Conference, pages 3--20. Springer

2018
[21]

Basil Mustafa, Carlos Riquelme, Joan Puigcerver, Rodolphe Jenatton, and Neil Houlsby. 2022. Multimodal contrastive learning with limoe: The language-image mixture of experts. arXiv preprint arXiv:2206.02770

2022 arXiv
[22]

Maximilian Nickel, Kevin Murphy, Volker Tresp, and Evgeniy Gabrilovich. 2015. A review of relational machine learning for knowledge graphs. Proceedings of the IEEE, 104(1):11--33

2015
[23]

M. Qu, J. Chen, L. Xhonneux, Y. Bengio, and J. Tang. 2021. Rnnlogic: Learning logic rules for reasoning on knowledge graphs. In International Conference on Learning Representations (ICLR)

2021
[24]

Carlos Riquelme, Joan Puigcerver, Basil Mustafa, Maxim Neumann, Rodolphe Jenatton, André Susano Pinto, Daniel Keysers, and Neil Houlsby. 2021. Scaling vision with sparse mixture of experts. In Advances in Neural Information Processing Systems (NeurIPS), volume 34, pages 8583--8595

2021
[25]

Amirmohammad Sadeghian, Mohammadreza Armandpour, Pasquale Ding, and D Wang. 2019. Drum: End-to-end differentiable rule mining on knowledge graphs. In Advances in Neural Information Processing Systems (NeurIPS), pages 15347--15357

2019
[26]

Noam Shazeer, Azalia Mirhoseini, Krzysztof Maziarz, Andy Davis, Quoc Le, Geoffrey Hinton, and Jeff Dean. 2017. Outrageously large neural networks: The sparsely-gated mixture-of-experts layer. In International Conference on Learning Representations (ICLR). Under review

2017
[27]

Fabian Suchanek, Gjergji Kasneci, and Gerhard Weikum. 2007. Yago: A core of semantic knowledge. In The WebConf, pages 697--706

2007
[28]

Zhiqing Sun, Zhi-Hong Deng, Jian-Yun Nie, and Jian Tang. 2019. Rotate: Knowledge graph embedding by relational rotation in complex space. In International Conference on Learning Representations (ICLR)

2019
[29]

Zhiqing Sun, Chao Huang, Jianyu Chen, Qianhan Wang, Xiang Wang, Yiyang Li, and Liqiang Nie. 2021. Rotate: Knowledge graph embedding by relational rotations. In Advances in Neural Information Processing Systems, pages 10057--10068

2021
[30]

K Teru, E

K. K Teru, E. Denis, and W. L Hamilton. 2020. Inductive relation prediction by subgraph reasoning. arXiv preprint arXiv:1911.06962

2020 arXiv
[31]

Kristina Toutanova and Danqi Chen. 2015. Observed versus latent features for knowledge base and text inference. In PWCVSMC, pages 57--66

2015
[32]

Nitin, and Partha Talukdar

Shikhar Vashishth, Soumya Sanyal, V. Nitin, and Partha Talukdar. 2019. Composition-based multi-relational graph convolutional networks

2019
[33]

Haotao Wang, Ziyu Jiang, Yuning You, Yan Han, Gaowen Liu, Jayanth Srinivasa, Ramana Rao Kompella, and Zhangyang Wang. 2023 a . https://github.com/VITA-Group/Graph-Mixture-of-Experts Graph mixture of experts: Learning on large-scale graphs with explicit diversity modeling . In ...

2023
[34]

Li Wang, Xiaohui Yan, and Yansong Feng. 2023 b . Enhancing knowledge graph embeddings with graph neural networks. Journal of Artificial Intelligence Research, 68:789--805

2023
[35]

Wenhan Xiong, Thien Hoang, and William Yang Wang. 2017. Deeppath: A reinforcement learning method for knowledge graph reasoning. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 564--573

2017
[36]

Fan Yang, Zhilin Yang, and William W Cohen. 2017. Differentiable learning of logical rules for knowledge base reasoning. In Advances in Neural Information Processing Systems (NeurIPS), pages 2319--2328

2017
[37]

Wenhao Yu, Chenguang Zhu, Lianhui Qin, Zhihan Zhang, Tong Zhao, and Meng Jiang. 2022. Diversifying content generation for commonsense reasoning with mixture of knowledge graph experts. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (ACL)

2022
[38]

Guibin Zhang, Xiangguo Sun, Yanwei Yue, Chonghe Jiang, Kun Wang, Tianlong Chen, and Shirui Pan. 2023 a . Graph sparsification via mixture of graphs. arXiv preprint

2023
[39]

Shuai Zhang, Yi Tay, Lina Yao, and Qi Liu. 2019. Quaternion knowledge graph embeddings. In Advances in Neural Information Processing Systems (NeurIPS)

2019
[40]

Yuning Zhang and Quanming Yao. 2022. Knowledge graph reasoning with relational directed graph. In Proceedings of TheWebConf

2022
[41]

Yuning Zhang, Quanming Yao, Weinan Dai, and Lei Chen. 2020. Autosf: Searching scoring functions for knowledge graph embedding. In Proceedings of the IEEE International Conference on Data Engineering (ICDE), pages 433--444. IEEE

2020
[42]

Yuning Zhang, Zhen Zhou, Quanming Yao, Xia Chu, and Bo Han. 2023 b . Adaprop: Learning adaptive propagation for knowledge graph reasoning. In Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD)

2023
[43]

Zhanke Zhou, Yongqi Zhang, Jiangchao Yao, Quanming Yao, and Bo Han. 2024. Less is more: One-shot-subgraph link prediction on large-scale knowledge graphs. In International Conference on Learning Representations (ICLR)

2024
[44]

Jinguo Zhu, Xizhou Zhu, Wenhai Wang, Xiaohua Wang, Hongsheng Li, Xiaogang Wang, and Jifeng Dai. 2022. Uni-perceiver-moe: Learning sparse generalist models with conditional moes. arXiv preprint arXiv:2206.04674

2022 arXiv
[45]

Zhaocheng Zhu, Xinyu Yuan, Mikhail Galkin, Sophie Xhonneux, Ming Zhang, Maxime Gazeau, and Jian Tang. 2023. A*net: A scalable path-based reasoning approach for knowledge graphs. In Advances in Neural Information Processing Systems (NeurIPS)

2023
[46]

Ziniu Zhu, Zhaocheng Zhang, Louis Xhonneux, and Jian Tang. 2021. Neural bellman-ford networks: A general graph neural network framework for link prediction. In Advances in Neural Information Processing Systems (NeurIPS)

2021
[47]

online" 'onlinestring :=

ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...
[48]

write newline

" write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

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

[1] [1]

Muhammad Ali, Abubakar Abid, and Parisa Kordjamshidi. 2022. Knowledge graphs: A comprehensive survey. IEEE Transactions on Knowledge and Data Engineering, 34(1):123--145

work page 2022

[2] [2]

Ke Cheng, Jie Liu, Wei Wang, and Yizhou Sun. 2022. Rlogic: Recursive logical rule learning from knowledge graphs. In Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining (KDD), pages 179--189

work page 2022

[3] [3]

R. Das, S. Dhuliawala, M. Zaheer, L. Vilnis, I. Durugkar, A. Krishnamurthy, A. Smola, and A. McCallum. 2017. Go for a walk and arrive at the answer: Reasoning over paths in knowledge bases using reinforcement learning. In International Conference on Learning Representations (ICLR)

work page 2017

[4] [4]

Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. 2017 a . Convolutional 2d knowledge graph embeddings. In Proceedings of the AAAI Conference on Artificial Intelligence

work page 2017

[5] [5]

Tim Dettmers, Pasquale Minervini, Pontus Stenetorp, and Sebastian Riedel. 2017 b . Convolutional 2d knowledge graph embeddings. In AAAI

work page 2017

[6] [6]

Charles Dugas, Yoshua Bengio, François B \'e lisle, Claude Nadeau, and René Garcia. 2001. Incorporating second-order functional knowledge for better option pricing. Advances in Neural Information Processing Systems (NeurIPS), 13:472--478

work page 2001

[7] [7]

Mikhail Galkin, Xinyu Yuan, Hesham Mostafa, Jian Tang, and Zhaocheng Zhu. 2024. https://arxiv.org/abs/2310.04562 Towards foundation models for knowledge graph reasoning . In Proceedings of the International Conference on Learning Representations (ICLR). Published as a conference paper at ICLR 2024

arXiv 2024

[8] [8]

Jinyang Huang, Xiachong Feng, Qiguang Chen, Hanjie Zhao, Zihui Cheng, Jiesong Bai, Jingxuan Zhou, Min Li, and Libo Qin. 2025. Mldebugging: Towards benchmarking code debugging across multi-library scenarios. ACL Findings

work page 2025