pith. machine review for the scientific record. sign in

arxiv: 2604.18913 · v1 · submitted 2026-04-20 · 💻 cs.CL

Recognition: unknown

LogosKG: Hardware-Optimized Scalable and Interpretable Knowledge Graph Retrieval

He Cheng , Yifu Wu , Saksham Khatwani , Maya Kruse , Dmitriy Dligach , Timothy A. Miller , Majid Afshar , Yanjun Gao
Authors on Pith no claims yet

Pith reviewed 2026-05-10 04:04 UTC · model grok-4.3

classification 💻 cs.CL
keywords knowledge graphsmulti-hop retrievalscalable retrievalhardware optimizationLLM integrationgraph partitioningbiomedical reasoning
0
0 comments X

The pith

LogosKG performs k-hop retrieval on billion-edge knowledge graphs by executing traversals as hardware-efficient operations over decomposed subject, object, and relation representations.

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

The paper presents LogosKG as a framework that integrates knowledge graphs with large language models for structured reasoning. It targets the core challenge of multi-hop retrieval, where existing methods often fail to scale efficiently while preserving interpretability. By relying on symbolic graph structures and breaking representations into subject, object, and relation components, the approach runs traversals through hardware-aligned operations. Degree-aware partitioning, cross-graph routing, and on-demand caching allow it to handle graphs with billions of edges. Experiments report faster performance than standard CPU and GPU methods with no drop in retrieval accuracy, and a follow-on two-round setup with LLMs shows how graph topology influences diagnostic reasoning in biomedical data.

Core claim

LogosKG enables scalable and interpretable k-hop retrieval on large KGs by building on symbolic KG formulations and executing traversal as hardware-efficient operations over decomposed subject, object, and relation representations, with degree-aware partitioning, cross-graph routing, and on-demand caching to reach billion-edge scale while preserving retrieval fidelity.

What carries the argument

Decomposed subject-object-relation representations combined with degree-aware partitioning, cross-graph routing, and on-demand caching to turn symbolic traversals into hardware-efficient operations.

If this is right

  • Substantial runtime reductions on CPU and GPU hardware for k-hop queries without any loss in retrieved facts.
  • The two-round KG-LLM loop isolates effects of hop distribution and connectivity on alignment between structured knowledge and model outputs.
  • Enables evidence-grounded analysis of how KG topology shapes LLM diagnostic reasoning at scales previously inaccessible.
  • Provides a path to next-generation KG-LLM systems that keep retrieval both fast and fully traceable.

Where Pith is reading between the lines

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

  • The same decomposition might reduce memory pressure in other graph algorithms that currently run on general-purpose processors.
  • If the hardware mapping generalizes, similar techniques could apply to dynamic graphs where edges arrive continuously.
  • The topology-analysis capability could be tested on non-biomedical domains to check whether hop and connectivity patterns affect reasoning in the same way.

Load-bearing premise

The partitioning, routing, and caching steps preserve exact retrieval correctness on arbitrary large graphs while delivering measured efficiency gains without hidden costs or accuracy trade-offs.

What would settle it

Measure end-to-end retrieval time and exact match accuracy on a public billion-edge KG against CPU and GPU baselines; if accuracy falls below baseline or runtime gains disappear after accounting for partitioning overhead, the central claim does not hold.

Figures

Figures reproduced from arXiv: 2604.18913 by Dmitriy Dligach, He Cheng, Majid Afshar, Maya Kruse, Saksham Khatwani, Timothy A. Miller, Yanjun Gao, Yifu Wu.

Figure 1
Figure 1. Figure 1: Overview of the LogosKG retrieval framework. LogosKG adopts a linear algebra–based retrieval method that supports both small and large KGs. For large graphs, LogosKG incorporates graph partitioning, cross-graph routing, and on-demand caching to enable efficient multi-hop retrieval across distributed subgraphs. The retrieved evidence can provide knowledge-grounded support for medical diagnosis generation. c… view at source ↗
Figure 2
Figure 2. Figure 2: Hop-distance distribution of pair entities for the ProbSum and DDXPlus dataset on UMLS and PrimeKG. and scalable performance on billion-edge graphs, establishing it as a reliable and hardware-optimized solution for large-scale KG retrieval. 6 Interaction Regimes between High-Hop KGs and LLM for Diagnosis Existing approaches that integrate KGs with LLMs often restrict reasoning to shallow neighborhoods due … view at source ↗
Figure 3
Figure 3. Figure 3: Performance comparison of KG filtering (Round 1) and enhancement (Round 2) across varying hop distances (k = 1 to k = 5) on ProbSum and DDXPlus datasets using UMLS and PrimeKG. F1-score metrics are shown for GPT-4.1, Qwen-2.5-7B-Instruct, and Llama-3.1-8B-Instruct in the zero-shot setting, with baseline performance included for reference [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: PDSQI-9 comparison for three models on DDX￾Plus dataset with UMLS (k=5). Each subplot displays seven evaluation dimensions comparing Baseline, Round 1 (KG￾filtered), and Round 2 (KG-enhanced) approaches. its prediction with the medical ontology. In Round 2, LOGOSKG provides the LLM with a long-tail candidate space that was previously invisible, mak￾ing the optimal “depth” not a fixed parameter. As shown in… view at source ↗
Figure 5
Figure 5. Figure 5: Illustrative example of the LOGOSKG two-round prediction pipeline on a clinical case from ProbSum. (A) Clinical input consists of a patient progress note with the gold standard diagnosis. (B) Round 1: The baseline LLM generates differential diagnoses, which are then filtered by validating against KG neighbors retrieved within k = 1 hop. (C) Round 2: Additional candidate diseases from the KG are presented t… view at source ↗
Figure 6
Figure 6. Figure 6: Performance comparison of KG filtering (Round 1) and enhancement (Round 2) across varying hop [PITH_FULL_IMAGE:figures/full_fig_p013_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: Performance comparison of KG filtering (Round 1) and enhancement (Round 2) across varying hop [PITH_FULL_IMAGE:figures/full_fig_p014_7.png] view at source ↗
read the original abstract

Knowledge graphs (KGs) are increasingly integrated with large language models (LLMs) to provide structured, verifiable reasoning. A core operation in this integration is multi-hop retrieval, yet existing systems struggle to balance efficiency, scalability, and interpretability. We introduce LogosKG, a novel, hardware-aligned framework that enables scalable and interpretable k-hop retrieval on large KGs by building on symbolic KG formulations and executing traversal as hardware-efficient operations over decomposed subject, object, and relation representations. To scale to billion-edge graphs, LogosKG integrates degree-aware partitioning, cross-graph routing, and on-demand caching. Experiments show substantial efficiency gains over CPU and GPU baselines without loss of retrieval fidelity. With proven performance in KG retrieval, a downstream two-round KG-LLM interaction demonstrates how LogosKG enables large-scale, evidence-grounded analysis of how KG topology, such as hop distribution and connectivity, shapes the alignment between structured biomedical knowledge and LLM diagnostic reasoning, thereby opening the door for next-generation KG-LLM integration. The source code is publicly available at https://github.com/LARK-NLP-Lab/LogosKG, and an online demo is available at https://lark-nlp-lab-logoskg.hf.space/.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The paper introduces LogosKG, a hardware-aligned framework for scalable and interpretable k-hop retrieval on large knowledge graphs. It builds on symbolic KG formulations by decomposing into subject, object, and relation representations for efficient traversal operations, and scales to billion-edge graphs via degree-aware partitioning, cross-graph routing, and on-demand caching. Experiments claim substantial efficiency gains over CPU and GPU baselines with no loss in retrieval fidelity. A downstream two-round KG-LLM interaction is used to analyze how KG topology (hop distribution, connectivity) affects alignment between structured biomedical knowledge and LLM diagnostic reasoning. Source code and an online demo are provided.

Significance. If the efficiency and fidelity claims hold with rigorous validation, LogosKG would offer a meaningful advance in KG-LLM integration by delivering hardware-efficient, interpretable multi-hop retrieval at scale. The open-source code and demo are positive contributions that could facilitate adoption. The biomedical topology analysis provides a concrete use case for evidence-grounded reasoning, though its isolation of topology effects would require careful controls.

major comments (2)
  1. [Abstract] Abstract: The central claim of 'substantial efficiency gains over CPU and GPU baselines without loss of retrieval fidelity' is asserted without any reported baselines, metrics for fidelity, datasets, quantitative results, or error analysis. This absence prevents assessment of whether the degree-aware partitioning and caching preserve correctness or deliver net gains on arbitrary large graphs.
  2. [Methods (inferred from abstract claims)] The description of the core mechanisms (decomposed subject/object/relation representations, degree-aware partitioning, cross-graph routing, on-demand caching) remains at a high level with no pseudocode, formal definitions, complexity analysis, or proof of correctness. Without these, it is impossible to verify that the approach maintains retrieval fidelity or avoids hidden overheads for billion-edge graphs.
minor comments (1)
  1. [Abstract] The abstract mentions 'proven performance in KG retrieval' but does not specify the exact fidelity metric or how the two-round KG-LLM setup isolates topology effects; clarifying this in the main text would improve readability.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their thoughtful and constructive feedback. We address each major comment below and indicate the changes we will make in the revised manuscript.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim of 'substantial efficiency gains over CPU and GPU baselines without loss of retrieval fidelity' is asserted without any reported baselines, metrics for fidelity, datasets, quantitative results, or error analysis. This absence prevents assessment of whether the degree-aware partitioning and caching preserve correctness or deliver net gains on arbitrary large graphs.

    Authors: We agree that the abstract would be strengthened by including specific quantitative support for the claims. The full manuscript reports these details in the Experiments section (baselines, fidelity metrics such as exact retrieval match, datasets including large biomedical KGs, and error analysis). We will revise the abstract to incorporate key results, baselines, metrics, and dataset references so that the central claims can be assessed directly from the abstract. revision: yes

  2. Referee: [Methods (inferred from abstract claims)] The description of the core mechanisms (decomposed subject/object/relation representations, degree-aware partitioning, cross-graph routing, and on-demand caching) remains at a high level with no pseudocode, formal definitions, complexity analysis, or proof of correctness. Without these, it is impossible to verify that the approach maintains retrieval fidelity or avoids hidden overheads for billion-edge graphs.

    Authors: The manuscript provides algorithmic descriptions and implementation details in Sections 2–3, but we acknowledge that additional formalization would improve clarity and verifiability. We will add pseudocode for the core traversal, partitioning, routing, and caching procedures to an appendix; include formal definitions of the decomposed representations in Section 2; and provide a complexity analysis (time and space) in Section 3. On proof of correctness, the optimizations preserve the semantics of standard symbolic k-hop retrieval (degree-aware partitioning affects only load balance and does not alter reachable nodes), which we will state explicitly and support via the empirical fidelity results already reported. A full formal proof is not provided because the system combines deterministic traversal with heuristic scheduling; we believe the combination of semantic equivalence argument and empirical validation is appropriate for this work. revision: partial

Circularity Check

0 steps flagged

No significant circularity detected

full rationale

The abstract and available context describe a hardware-aligned KG retrieval framework using symbolic formulations, decomposed representations, degree-aware partitioning, cross-graph routing, and on-demand caching, with claims supported by efficiency experiments and a downstream KG-LLM setup. No equations, self-definitional constructs, fitted inputs renamed as predictions, or load-bearing self-citations appear in the provided text. All central claims rest on independently described methods and external experimental validation rather than reducing to inputs by construction, making the derivation chain self-contained.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides no explicit free parameters, axioms, or invented entities; framework described at conceptual level without mathematical details or assumptions listed.

pith-pipeline@v0.9.0 · 5539 in / 1018 out tokens · 27848 ms · 2026-05-10T04:04:54.157545+00:00 · methodology

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

60 extracted references · 24 canonical work pages · 2 internal anchors

  1. [1]

    Knowledge Graph Retrieval-Augmented Generation for LLM -based Recommendation

    Wang, Shijie and Fan, Wenqi and Feng, Yue and Shanru, Lin and Ma, Xinyu and Wang, Shuaiqiang and Yin, Dawei. Knowledge Graph Retrieval-Augmented Generation for LLM -based Recommendation. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2025. doi:10.18653/v1/2025.acl-long.1317

    work page doi:10.18653/v1/2025.acl-long.1317 2025

  2. [2]

    Biomedical knowledge graph: A survey of domains, tasks, and real-world applications

    Biomedical knowledge graph: A survey of domains, tasks, and real-world applications , author=. arXiv preprint arXiv:2501.11632 , year=

    work page arXiv

  3. [3]

    IEEE Transactions on Computational Social Systems , volume=

    A knowledge graph-based many-objective model for explainable social recommendation , author=. IEEE Transactions on Computational Social Systems , volume=. 2023 , publisher=

    2023

  4. [4]

    H op RAG : Multi-Hop Reasoning for Logic-Aware Retrieval-Augmented Generation

    Liu, Hao and Wang, Zhengren and Chen, Xi and Li, Zhiyu and Xiong, Feiyu and Yu, Qinhan and Zhang, Wentao. H op RAG : Multi-Hop Reasoning for Logic-Aware Retrieval-Augmented Generation. Findings of the Association for Computational Linguistics: ACL 2025. 2025. doi:10.18653/v1/2025.findings-acl.97

    work page doi:10.18653/v1/2025.findings-acl.97 2025

  5. [5]

    The Thirteenth International Conference on Learning Representations , year=

    Reasoning of Large Language Models over Knowledge Graphs with Super-Relations , author=. The Thirteenth International Conference on Learning Representations , year=

  6. [6]

    Verify-in-the-Graph: Entity Disambiguation Enhancement for Complex Claim Verification with Interactive Graph Representation

    Pham, Hoang and Nguyen, Thanh-Do and Bui, Khac-Hoai Nam. Verify-in-the-Graph: Entity Disambiguation Enhancement for Complex Claim Verification with Interactive Graph Representation. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers)...

    work page doi:10.18653/v1/2025.naacl-long.268 2025

  7. [7]

    arXiv preprint arXiv:2403.09724 , year=

    Claimver: Explainable claim-level verification and evidence attribution of text through knowledge graphs , author=. arXiv preprint arXiv:2403.09724 , year=

    work page arXiv

  8. [8]

    arXiv preprint arXiv:2506.00783 , year=

    KG-TRACES: Enhancing Large Language Models with Knowledge Graph-constrained Trajectory Reasoning and Attribution Supervision , author=. arXiv preprint arXiv:2506.00783 , year=

    work page arXiv

  9. [9]

    Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

    Path-based explanation for knowledge graph completion , author=. Proceedings of the 30th ACM SIGKDD Conference on Knowledge Discovery and Data Mining , pages=

  10. [10]

    arXiv preprint arXiv:2305.06590 , year=

    FactKG: Fact verification via reasoning on knowledge graphs , author=. arXiv preprint arXiv:2305.06590 , year=

    work page arXiv

  11. [11]

    Nature Scientific Data , doi=

    Building a knowledge graph to enable precision medicine , author=. Nature Scientific Data , doi=

  12. [12]

    International Conference on Learning Representations , year=

    Scalable Neural Methods for Reasoning With a Symbolic Knowledge Base , author=. International Conference on Learning Representations , year=

  13. [13]

    arXiv preprint arXiv:2505.22993 , year=

    Verify-in-the-Graph: Entity Disambiguation Enhancement for Complex Claim Verification with Interactive Graph Representation , author=. arXiv preprint arXiv:2505.22993 , year=

    work page arXiv

  14. [14]

    arXiv preprint arXiv:2502.16514 , year=

    GraphCheck: Breaking Long-Term Text Barriers with Extracted Knowledge Graph-Powered Fact-Checking , author=. arXiv preprint arXiv:2502.16514 , year=

    work page arXiv

  15. [15]

    arXiv preprint arXiv:2503.07282 , year=

    A Graph-based Verification Framework for Fact-Checking , author=. arXiv preprint arXiv:2503.07282 , year=

    work page arXiv

  16. [16]

    Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

    Fact verification in knowledge graphs using LLMs , author=. Proceedings of the 48th International ACM SIGIR Conference on Research and Development in Information Retrieval , pages=

  17. [17]

    Thirty-seventh Conference on Neural Information Processing Systems , year=

    A*Net: A Scalable Path-based Reasoning Approach for Knowledge Graphs , author=. Thirty-seventh Conference on Neural Information Processing Systems , year=

  18. [18]

    Proceedings of the ACM on Web Conference 2025 , pages=

    Paths-over-graph: Knowledge graph empowered large language model reasoning , author=. Proceedings of the ACM on Web Conference 2025 , pages=

    2025

  19. [19]

    arXiv preprint arXiv:2412.15235 , year=

    Og-rag: Ontology-grounded retrieval-augmented generation for large language models , author=. arXiv preprint arXiv:2412.15235 , year=

    work page arXiv

  20. [20]

    arXiv preprint arXiv:1901.08248 , year=

    Tigergraph: A native MPP graph database , author=. arXiv preprint arXiv:1901.08248 , year=

    work page arXiv 1901

  21. [21]

    Complex syst , volume=

    The igraph software , author=. Complex syst , volume=

  22. [22]

    IEEE High Performance Extreme Computing Conference (HPEC) , year =

    Mathematical Foundations of the GraphBLAS , author =. IEEE High Performance Extreme Computing Conference (HPEC) , year =

  23. [23]

    ACM Transactions on Intelligent Systems and Technology (TIST) , volume=

    Snap: A general-purpose network analysis and graph-mining library , author=. ACM Transactions on Intelligent Systems and Technology (TIST) , volume=. 2016 , publisher=

    2016

  24. [24]

    Deep graph library: A graph-centric, highly-performant package for graph neural networks,

    Deep graph library: A graph-centric, highly-performant package for graph neural networks , author=. arXiv preprint arXiv:1909.01315 , year=

    work page arXiv 1909

  25. [25]

    figshare , author =

    The graph-tool python library , url =. figshare , author =. 2014 , keywords =. doi:10.6084/m9.figshare.1164194 , urldate =

    work page doi:10.6084/m9.figshare.1164194 2014

  26. [26]

    Fast Graph Representation Learning with PyTorch Geometric

    Fast graph representation learning with PyTorch Geometric , author=. arXiv preprint arXiv:1903.02428 , year=

    work page internal anchor Pith review arXiv 1903

  27. [27]

    2008 , institution=

    Exploring network structure, dynamics, and function using NetworkX , author=. 2008 , institution=

    2008

  28. [28]

    doi:10.5281/zenodo.10631255 , url =

    Erik Welch and Jim Kitchen and Sultan Orazbayev and ParticularMiner and Stan Seibert and William Zijie Zhang and Adam Lugowski and Paul Nguyen , title =. doi:10.5281/zenodo.10631255 , url =

    work page doi:10.5281/zenodo.10631255

  29. [29]

    Nucleic acids research , volume=

    The unified medical language system (UMLS): integrating biomedical terminology , author=. Nucleic acids research , volume=. 2004 , publisher=

    2004

  30. [30]

    Scientific data , volume=

    Building a PubMed knowledge graph , author=. Scientific data , volume=. 2020 , publisher=

    2020

  31. [31]

    MedIR workshop, sigir , pages=

    Quickumls: a fast, unsupervised approach for medical concept extraction , author=. MedIR workshop, sigir , pages=

  32. [32]

    Overview of the Problem List Summarization ( P rob S um) 2023 Shared Task on Summarizing Patients' Active Diagnoses and Problems from Electronic Health Record Progress Notes

    Gao, Yanjun and Dligach, Dmitriy and Miller, Timothy and Afshar, Majid. Overview of the Problem List Summarization ( P rob S um) 2023 Shared Task on Summarizing Patients' Active Diagnoses and Problems from Electronic Health Record Progress Notes. Proceedings of the 22nd Workshop on Biomedical Natural Language Processing and BioNLP Shared Tasks. 2023. doi:...

    work page doi:10.18653/v1/2023.bionlp-1.43 2023

  33. [33]

    Advances in neural information processing systems , volume=

    Ddxplus: A new dataset for automatic medical diagnosis , author=. Advances in neural information processing systems , volume=

  34. [34]

    Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining , pages=

    Smore: Knowledge graph completion and multi-hop reasoning in massive knowledge graphs , author=. Proceedings of the 28th ACM SIGKDD conference on knowledge discovery and data mining , pages=

  35. [35]

    arXiv preprint arXiv:2412.05547 , year=

    Kg-retriever: Efficient knowledge indexing for retrieval-augmented large language models , author=. arXiv preprint arXiv:2412.05547 , year=

    work page arXiv

  36. [36]

    arXiv preprint arXiv:2504.20114 , year=

    TreeHop: Generate and Filter Next Query Embeddings Efficiently for Multi-hop Question Answering , author=. arXiv preprint arXiv:2504.20114 , year=

    work page arXiv

  37. [37]

    Divide by question, conquer by agent: Split-rag with question-driven graph partitioning.arXiv preprint arXiv:2505.13994, 2025

    Divide by Question, Conquer by Agent: SPLIT-RAG with Question-Driven Graph Partitioning , author=. arXiv preprint arXiv:2505.13994 , year=

    work page arXiv

  38. [38]

    Proceedings of the first workshop on Parallel programming for analytics applications , pages=

    A performance evaluation of open source graph databases , author=. Proceedings of the first workshop on Parallel programming for analytics applications , pages=

  39. [39]

    Social Network Analysis and Mining , volume=

    A comparative evaluation of social network analysis tools: performance and community engagement perspectives , author=. Social Network Analysis and Mining , volume=. 2025 , publisher=

    2025

  40. [40]

    arXiv preprint arXiv:2406.14326 , year=

    medIKAL: Integrating knowledge graphs as assistants of LLMs for enhanced clinical diagnosis on EMRs , author=. arXiv preprint arXiv:2406.14326 , year=

    work page arXiv

  41. [41]

    AAAI Bridge Program on AI for Medicine and Healthcare , pages=

    Kg4diagnosis: A hierarchical multi-agent llm framework with knowledge graph enhancement for medical diagnosis , author=. AAAI Bridge Program on AI for Medicine and Healthcare , pages=. 2025 , organization=

    2025

  42. [42]

    arXiv e-prints , pages=

    Adaptive Knowledge Graphs Enhance Medical Question Answering: Bridging the Gap Between LLMs and Evolving Medical Knowledge , author=. arXiv e-prints , pages=

  43. [43]

    Proceedings of the conference

    Benchmark and best practices for biomedical knowledge graph embeddings , author=. Proceedings of the conference. Association for Computational Linguistics. Meeting , volume=

  44. [44]

    Jmir Ai , volume=

    Leveraging medical knowledge graphs into large language models for diagnosis prediction: Design and application study , author=. Jmir Ai , volume=. 2025 , publisher=

    2025

  45. [45]

    Scientific Data , volume=

    PubMed knowledge graph 2.0: Connecting papers, patents, and clinical trials in biomedical science , author=. Scientific Data , volume=. 2025 , publisher=

    2025

  46. [46]

    Journal of the American Medical Informatics Association , volume=

    Development and validation of the provider documentation summarization quality instrument for large language models , author=. Journal of the American Medical Informatics Association , volume=. 2025 , publisher=

    2025

  47. [47]

    Scientific data , volume=

    MIMIC-III, a freely accessible critical care database , author=. Scientific data , volume=. 2016 , publisher=

    2016

  48. [48]

    2025 , eprint=

    Qwen2.5 Technical Report , author=. 2025 , eprint=

    2025

  49. [49]

    arXiv e-prints , pages=

    The llama 3 herd of models , author=. arXiv e-prints , pages=

  50. [50]

    Gemma: Open Models Based on Gemini Research and Technology

    Gemma: Open models based on gemini research and technology , author=. arXiv preprint arXiv:2403.08295 , year=

    work page internal anchor Pith review arXiv

  51. [51]

    2023 , eprint=

    Mistral 7B , author=. 2023 , eprint=

    2023

  52. [52]

    2025 , howpublished =

    OpenAI o3-mini System Card , author =. 2025 , howpublished =

    2025

  53. [53]

    2025 , month =

    Introducing GPT-4.1 , author =. 2025 , month =

    2025

  54. [54]

    2025 , month =

    GPT-5 Mini Model Card , author =. 2025 , month =

    2025

  55. [55]

    arXiv 2024 , author=

    Retrieval-Augmented Generation with Graphs (GraphRAG). arXiv 2024 , author=

    2024

  56. [56]

    npj Digital Medicine , volume=

    Evaluating clinical AI summaries with large language models as judges , author=. npj Digital Medicine , volume=. 2025 , publisher=

    2025

  57. [57]

    12th International Conference on Learning Representations, ICLR 2024 , year=

    GRAPHCARE: ENHANCING HEALTHCARE PREDICTIONS WITH PERSONALIZED KNOWLEDGE GRAPHS , author=. 12th International Conference on Learning Representations, ICLR 2024 , year=

    2024

  58. [58]

    arXiv preprint arXiv:2509.18316 , year=

    Brittleness and Promise: Knowledge Graph Based Reward Modeling for Diagnostic Reasoning , author=. arXiv preprint arXiv:2509.18316 , year=

    work page arXiv

  59. [59]

    Proceedings of the 34th ACM International Conference on Information and Knowledge Management , pages=

    Agentigraph: A multi-agent knowledge graph framework for interactive, domain-specific llm chatbots , author=. Proceedings of the 34th ACM International Conference on Information and Knowledge Management , pages=

  60. [60]

    arXiv preprint arXiv:2507.02773 , year=

    KERAP: A Knowledge-Enhanced Reasoning Approach for Accurate Zero-shot Diagnosis Prediction Using Multi-agent LLMs , author=. arXiv preprint arXiv:2507.02773 , year=

    work page arXiv