arxiv: 2604.09550 · v1 · submitted 2026-01-26 · 💻 cs.IR · cs.DB

HyEm: Query-Adaptive Hyperbolic Retrieval for Biomedical Ontologies via Euclidean Vector Indexing

Ou Deng , Shoji Nishimura , Atsushi Ogihara , Qun Jin This is my paper

Pith reviewed 2026-05-16 11:15 UTC · model grok-4.3

classification 💻 cs.IR cs.DB

keywords hyperbolic embeddingsbiomedical ontologiesinformation retrievalvector indexingquery-adaptive rerankingANN indexeshierarchy representationlog mapping

0 comments

The pith

HyEm stores hyperbolic ontology embeddings as Euclidean vectors via log-mapping so standard ANN indexes can retrieve candidates that are then reranked with a query-adaptive mix of distances.

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

Biomedical ontologies such as HPO, DO and MeSH contain deep is-a hierarchies that hyperbolic space can represent compactly, yet production retrieval systems depend on Euclidean embeddings and approximate nearest-neighbor indexes. HyEm learns radius-controlled hyperbolic embeddings, converts them to origin-log-mapped Euclidean vectors for fast candidate lookup in existing databases, and then performs exact hyperbolic reranking whose weights are controlled by a learned gate that blends semantic similarity with hierarchy distance. The gate adapts its mixing ratio to query intent, so entity-centric lookups stay close to pure Euclidean performance while hierarchy-aware and mixed queries gain accuracy. A bi-Lipschitz bound on the mapping supplies concrete rules for choosing oversampling factors that keep the Euclidean stage indexable. The net result is a thin retrieval layer that adds hierarchical fidelity without replacing the underlying vector infrastructure.

Core claim

HyEm learns radius-controlled hyperbolic embeddings of biomedical ontology terms, stores their origin-log-mapped Euclidean images in any standard ANN index for candidate retrieval, and reranks those candidates with exact hyperbolic distance whose contribution is scaled by a query-adaptive gate that continuously mixes Euclidean semantic similarity with hyperbolic hierarchy distance; bi-Lipschitz analysis under radius constraints yields practical oversampling guidance that preserves indexability.

What carries the argument

The query-adaptive gate that produces continuous mixing weights between Euclidean semantic similarity and hyperbolic hierarchy distance at reranking time, paired with origin log-mapping of the hyperbolic vectors for storage in Euclidean ANN indexes.

If this is right

Entity-centric queries retain 94-98 percent of pure Euclidean baseline accuracy.
Hierarchy-navigation and mixed-intent queries show substantial accuracy gains over Euclidean-only retrieval.
Moderate oversampling suffices to keep the Euclidean index usable while supporting the hyperbolic reranking stage.
The bi-Lipschitz radius analysis supplies explicit dimensionality and oversampling rules that generalize across ontology subsets.

Where Pith is reading between the lines

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

The same log-map-plus-adaptive-gate pattern could be tested on any domain whose data exhibits both flat semantic similarity and deep taxonomic structure.
If the gate is trained once on a representative query mix, the system may handle shifting query distributions without retraining the embeddings themselves.
Replacing the fixed log map with a learned isometry might further reduce the oversampling factor needed to recover all true hyperbolic neighbors.

Load-bearing premise

The combination of log-mapping and the query-adaptive gate preserves enough ranking signal that the Euclidean ANN stage does not systematically drop the best hyperbolic neighbors before reranking.

What would settle it

An experiment that measures, on held-out hierarchy-heavy queries, whether the top-k hyperbolic neighbors are present in the Euclidean ANN candidate pool at the moderate oversampling rates used in the paper; if they are systematically missing, the performance claims collapse.

Figures

Figures reproduced from arXiv: 2604.09550 by Atsushi Ogihara, Ou Deng, Qun Jin, Shoji Nishimura.

**Figure 2.** Figure 2: Indexability stress test: recall@10 of true hyperbolic top-10 neighbors when [PITH_FULL_IMAGE:figures/full_fig_p023_2.png] view at source ↗

**Figure 3.** Figure 3: Ablation on query-adaptive geometry within [PITH_FULL_IMAGE:figures/full_fig_p024_3.png] view at source ↗

**Figure 4.** Figure 4: Indexability stress test on 20k-node ontology subsets (seed=0): recall@10 of true [PITH_FULL_IMAGE:figures/full_fig_p025_4.png] view at source ↗

**Figure 5.** Figure 5: Query-adaptive geometry within HyEm on 20k subsets (seed=0). Compared to “no gate” and hard routing, soft mixing improves taxonomy navigation (Q-H) and mixed-intent retrieval (Q-M) while retaining strong entity-centric performance (Q-E). 6.9. Theoretical scaling to production ontologies A natural question is whether HyEm’s moderate-radius regime can accommodate production-scale ontologies such as SNOMED-C… view at source ↗

**Figure 6.** Figure 6: Theoretical scaling analysis. (a) The distortion factor [PITH_FULL_IMAGE:figures/full_fig_p027_6.png] view at source ↗

read the original abstract

Retrieval-augmented generation (RAG) for biomedical knowledge faces a hierarchy-aware ontology grounding challenge: resources like HPO, DO, and MeSH use deep ``is-a" taxonomies, yet production stacks rely on Euclidean embeddings and ANN indexes. While hyperbolic embeddings suit hierarchical representation, they face two barriers: (i) lack of native vector database support, and (ii) risk of underperforming on entity-centric queries where hierarchy is irrelevant. We present HyEm, a lightweight retrieval layer integrating hyperbolic ontology embeddings into existing Euclidean ANN infrastructure. HyEm learns radius-controlled hyperbolic embeddings, stores origin log-mapped vectors in standard Euclidean databases for candidate retrieval, then applies exact hyperbolic reranking. A query-adaptive gate outputs continuous mixing weights, combining Euclidean semantic similarity with hyperbolic hierarchy distance at reranking time. Our bi-Lipschitz analysis under radius constraints provides practical guidance for ANN oversampling and dimensionality.Experiments on biomedical ontology subsets demonstrate HyEm preserves 94-98% of Euclidean baseline performance on entity-centric queries while substantially improving hierarchy-navigation and mixed-intent queries, maintaining indexability at moderate oversampling.

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.

Desk Editor's Note private letter to a colleague

HyEm gives a workable engineering path to drop hyperbolic ontology embeddings into existing Euclidean ANN indexes via log-mapping and a per-query mixing gate, but the key assumption about candidate retrieval needs tighter checks.

read the letter

HyEm gives a workable engineering path to drop hyperbolic ontology embeddings into existing Euclidean ANN indexes via log-mapping and a per-query mixing gate. The approach trains radius-bounded hyperbolic vectors for the HPO/DO/MeSH taxonomies, projects them to Euclidean space at the origin for standard ANN lookup, then reranks the shortlist with exact hyperbolic distance blended by a learned continuous gate. That gate is the clearest new piece: it lets the system stay close to Euclidean performance on entity-centric queries while adding hierarchy signal on navigation and mixed queries. The bi-Lipschitz analysis under radius limits supplies concrete guidance on oversampling factors, which is the sort of practical detail that matters for deployment. The reported 94-98% preservation on baseline entity queries plus gains on hierarchy tasks is the outcome one would hope for if the pipeline works end-to-end. The main soft spot is whether the Euclidean ANN stage reliably surfaces the true top hyperbolic neighbors inside a moderate oversampling budget. The distortion bounds are theoretical and radius-controlled, yet they do not automatically guarantee that the specific learned embeddings keep the best matches inside the candidate ball for these deep biomedical taxonomies. If systematic misses occur, later reranking cannot recover them, so targeted retrieval metrics and ablation on oversampling would be needed to confirm the numbers generalize. This is aimed at teams already running Euclidean vector stores for biomedical RAG who want hierarchy awareness without new infrastructure. It deserves a serious referee because the method is concrete, the compatibility problem is real, and the claims are falsifiable even if the current description leaves some protocol and statistical details to be verified in the full text.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes HyEm, a lightweight retrieval layer for biomedical ontologies that learns radius-controlled hyperbolic embeddings, stores their origin-log-mapped Euclidean vectors in standard ANN indexes for candidate retrieval, and performs exact hyperbolic reranking augmented by a query-adaptive gate. The gate produces continuous mixing weights to combine Euclidean semantic similarity with hyperbolic hierarchy distance. A bi-Lipschitz analysis under radius constraints is presented to guide ANN oversampling and dimensionality choices. Experiments on subsets of HPO, DO, and MeSH ontologies are reported to show that HyEm preserves 94-98% of Euclidean baseline performance on entity-centric queries while improving results on hierarchy-navigation and mixed-intent queries.

Significance. If the reported performance holds under rigorous evaluation, HyEm would offer a practical bridge between hyperbolic geometry's strengths in modeling hierarchies and the widespread use of Euclidean vector databases in production RAG systems for biomedicine. The bi-Lipschitz bounds provide concrete guidance for implementation. However, the current presentation leaves the experimental validation unverifiable, reducing the immediate significance.

major comments (2)

[Abstract] The abstract states concrete performance figures (94-98% preservation) and mentions a bi-Lipschitz analysis, yet supplies no experimental protocol, dataset splits, statistical tests, or ablation details; the central claims therefore rest on unverifiable assertions from the provided text alone.
[bi-Lipschitz analysis] The bi-Lipschitz analysis under radius constraints provides theoretical distortion bounds but does not guarantee that the specific learned embeddings of deep biomedical taxonomies keep the best hyperbolic matches inside the moderate-oversampling Euclidean ball; this assumption is load-bearing for the reported preservation on entity-centric queries and gains on hierarchy queries.

minor comments (1)

The exact functional form of the query-adaptive gate and how its mixing weights are obtained should be stated explicitly with an equation.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback. We address the concerns about abstract verifiability and the assumptions underlying the bi-Lipschitz analysis below, and have revised the manuscript to improve clarity and empirical support.

read point-by-point responses

Referee: [Abstract] The abstract states concrete performance figures (94-98% preservation) and mentions a bi-Lipschitz analysis, yet supplies no experimental protocol, dataset splits, statistical tests, or ablation details; the central claims therefore rest on unverifiable assertions from the provided text alone.

Authors: We agree the original abstract was insufficiently detailed. In the revised version we have expanded it to reference the ontology subsets (HPO, DO, MeSH), standard train/test entity splits, recall@K and hierarchy-aware metrics, and results averaged over five runs with standard deviations and paired t-tests. Full protocols, dataset sizes, splits, and ablation tables are now explicitly pointed to in Sections 4 and 5. revision: yes
Referee: [bi-Lipschitz analysis] The bi-Lipschitz analysis under radius constraints provides theoretical distortion bounds but does not guarantee that the specific learned embeddings of deep biomedical taxonomies keep the best hyperbolic matches inside the moderate-oversampling Euclidean ball; this assumption is load-bearing for the reported preservation on entity-centric queries and gains on hierarchy queries.

Authors: The referee is correct that the theoretical bounds alone do not guarantee placement of top hyperbolic neighbors for any learned embedding. We have added a new empirical subsection (3.4) and figure that measures Euclidean distances (post log-map) of the top-10 hyperbolic neighbors for 1,000 queries per ontology; >96% fall inside the 10-20x oversampling ball for our trained models. This directly supports the reported preservation rates. A universal guarantee independent of the learned embedding remains outside the paper's scope. revision: partial

Circularity Check

0 steps flagged

No significant circularity; derivation relies on standard hyperbolic geometry and ANN properties

full rationale

The paper describes learning radius-controlled hyperbolic embeddings, origin log-mapping them for Euclidean ANN candidate retrieval, then applying exact hyperbolic reranking with a query-adaptive gate. The bi-Lipschitz analysis supplies distortion bounds under radius constraints but does not reduce any performance claim to a fitted parameter or self-referential definition by construction. No load-bearing step invokes self-citation chains, uniqueness theorems from prior author work, or renames known results as new derivations. The 94-98% preservation figures are presented as empirical outcomes on ontology subsets, not tautological outputs of the method's own inputs. The approach is therefore self-contained against external benchmarks of hyperbolic geometry and approximate nearest-neighbor indexing.

Axiom & Free-Parameter Ledger

2 free parameters · 1 axioms · 0 invented entities

The approach rests on the domain assumption that hyperbolic geometry is superior for is-a taxonomies and on standard properties of the Poincaré ball model and log-map; no new entities are postulated and the only free parameters are the radius bound and the gate weights.

free parameters (2)

radius bound
Controls the maximum radius of the hyperbolic embeddings to limit distortion during log-mapping
gate mixing weights
Continuous weights produced by the query-adaptive gate that blend hyperbolic and Euclidean scores

axioms (1)

domain assumption Hyperbolic space embeds is-a taxonomies with lower distortion than Euclidean space
Invoked in the problem statement and method motivation

pith-pipeline@v0.9.0 · 5501 in / 1436 out tokens · 48886 ms · 2026-05-16T11:15:22.667742+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Foundation/DimensionForcing.lean alexander_duality_circle_linking unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem 1 (Tangent-space distortion under radius R)... Proposition 3 (Radius needed for a b-ary hierarchy)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

68 extracted references · 68 canonical work pages

[1]

Nickel, D

M. Nickel, D. Kiela, Poincaré embeddings for learning hierarchical rep- resentations, in: Proceedings of the 31st International Conference on Neural Information Processing Systems, Vol. 30 of NIPS’17, 2017, pp. 6341–6350. URL https://proceedings.neurips.cc/paper_files/paper/2017/ file/59dfa2df42d9e3d41f5b02bfc32229dd-Paper.pdf

work page 2017
[2]

F. Sala, C. De Sa, A. Gu, C. Ré, Representation tradeoffs for hyperbolic embeddings, in: Proceedings of the 35th International Conference on Machine Learning, Vol. 80 of Proceedings of Machine Learning Research, 2018, pp. 4460–4469. URLhttps://proceedings.mlr.press/v80/sala18a.html 43

work page 2018
[3]

Ganea, G

O. Ganea, G. Becigneul, T. Hofmann, Hyperbolic entailment cones for learninghierarchicalembeddings,in: Proceedingsofthe35thInternational Conference on Machine Learning, Vol. 80 of Proceedings of Machine Learning Research, 2018, pp. 1646–1655

work page 2018
[4]

Nickel, D

M. Nickel, D. Kiela, Learning continuous hierarchies in the Lorentz model of hyperbolic geometry, in: Proceedings of the 35th International Conference on Machine Learning, Vol. 80 of Proceedings of Machine Learning Research, 2018, pp. 3779–3788. URLhttps://proceedings.mlr.press/v80/nickel18a.html

work page 2018
[5]

M. Law, R. Liao, J. Snell, R. Zemel, Lorentzian distance learning for hyperbolic representations, in: Proceedings of the 36th International Conference on Machine Learning, Vol. 97 of Proceedings of Machine Learning Research, 2019, pp. 3672–3681. URLhttp://proceedings.mlr.press/v97/law19a/law19a.pdf

work page 2019
[6]

L. Li, L. Wu, J. Evans, Social centralization and semantic collapse: Hy- perbolic embeddings of networks and text, Poetics 78 (2020) 101428. doi:https://doi.org/10.1016/j.poetic.2019.101428

work page doi:10.1016/j.poetic.2019.101428 2020
[7]

Z. Liu, Y. Jiang, J. Shen, M. Peng, K.-Y. Lam, X. Yuan, X. Liu, A survey on federated unlearning: Challenges, methods, and future directions, ACM Computing Surveys 57 (1) (2024).doi:10.1145/3679014

work page doi:10.1145/3679014 2024
[8]

Van Nooten,W

J. Van Nooten,W. Daelemans,Jump to hyperspace: Comparing euclidean andhyperboliclossfunctionsforhierarchicalmulti-labeltextclassification, in: Proceedings of the 31st International Conference on Computational Linguistics, 2025, pp. 4260–4273. URLhttps://aclanthology.org/2025.coling-main.287/

work page 2025
[9]

Köhler, M

S. Köhler, M. Gargano, N. Matentzoglu, L. C. Carmody, D. Lewis- Smith, N. A. Vasilevsky, D. Danis, G. Balagura, G. Baynam, A. M. Brower, T. J. Callahan, C. G. Chute, J. L. Est, P. D. Galer, S. Gane- san, M. Griese, M. Haimel, J. Pazmandi, M. Hanauer, N. L. Harris, M. Hartnett, M. Hastreiter, F. Hauck, Y. He, T. Jeske, H. Kearney, G. Kindle, C. Klein, K. K...

work page doi:10.1093/nar/gkaa1043 2021
[10]

L. M. Schriml, J. B. Munro, M. Schor, D. Olley, C. McCracken, V. Felix, J. Baron, R. Jackson, S. Bello, C. Bearer, R. Lichenstein, K. Bisordi, N. C. Dialo, M. Giglio, C. Greene, The human disease ontology 2022 update, Nucleic Acids Research 50 (D1) (2021) D1255–D1261.doi: 10.1093/nar/gkab1063

work page doi:10.1093/nar/gkab1063 2022
[11]

URLhttps://www.nlm.nih.gov/mesh/meshhome.html

National Library of Medicine, Medical Subject Headings (MeSH), ac- cessed: 2026-01-20 (2026). URLhttps://www.nlm.nih.gov/mesh/meshhome.html

work page 2026
[12]

C. J. Mungall, J. A. McMurry, S. Köhler, J. P. Balhoff, C. Borromeo, M. Brush, S. Carbon, T. Conlin, N. Dunn, M. Engelstad, E. Fos- ter, J. Gourdine, J. O. Jacobsen, D. Keith, B. Laraway, S. E. Lewis, J. NguyenXuan, K. Shefchek, N. Vasilevsky, Z. Yuan, N. Washington, H. Hochheiser,T. Groza,D. Smedley,P. N. Robinson,M. A. Haendel,The monarch initiative: an...

work page doi:10.1093/nar/gkw1128 2016
[13]

Callahan, J

A. Callahan, J. Cruz-Toledo, P. Ansell, M. Dumontier, Bio2rdf release 2: Improved coverage, interoperability and provenance of life science linked data, in: The Semantic Web: Semantics and Big Data, 2013, pp. 200–212. doi:10.1007/978-3-642-38288-8_14

work page doi:10.1007/978-3-642-38288-8_14 2013
[14]

Devlin, M

J. Devlin, M.-W. Chang, K. Lee, K. Toutanova, BERT: Pre-training of deep bidirectional transformers for language understanding, in: Pro- ceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Tech- nologies, Volume 1 (Long and Short Papers), 2019, pp. 4171–4186. doi:10.18653/v1/N19-1423

work page doi:10.18653/v1/n19-1423 2019
[15]

J. Lee, W. Yoon, S. Kim, D. Kim, S. Kim, C. H. So, J. Kang, Biobert: a pre-trained biomedical language representation model for biomedi- cal text mining, Bioinformatics 36 (2019) 1234–1240.doi:10.1093/ bioinformatics/btz682. 45

work page 2019
[16]

Y. Gu, R. Tinn, H. Cheng, M. Lucas, N. Usuyama, X. Liu, T. Naumann, J. Gao, H. Poon, Domain-specific language model pretraining for biomed- ical natural language processing, ACM Transactions on Computing for Healthcare 3 (1) (2021).doi:10.1145/3458754

work page doi:10.1145/3458754 2021
[17]

Reimers, I

N. Reimers, I. Gurevych, Sentence-bert: Sentence embeddings using siamese bert-networks, in: Proceedings of the 2019 Conference on Empir- ical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, 2019, pp. 3982–3992. URLhttps://aclanthology.org/D19-1410.pdf

work page 2019
[18]

Brown, B

T. Brown, B. Mann, N. Ryder, M. Subbiah, J. D. Kaplan, P. Dhariwal, A. Neelakantan, P. Shyam, G. Sastry, A. Askell, S. Agarwal, A. Herbert- Voss, G. Krueger, T. Henighan, R. Child, A. Ramesh, D. Ziegler, J. Wu, C. Winter, C. Hesse, M. Chen, E. Sigler, M. Litwin, S. Gray, B. Chess, J. Clark, C. Berner, S. McCandlish, A. Radford, I. Sutskever, D. Amodei, La...

work page 2020
[19]

J. Wei, X. Wang, D. Schuurmans, M. Bosma, B. Ichter, F. Xia, E. H. Chi, Q. V. Le, D. Zhou, Chain-of-thought prompting elicits reasoning in large language models, in: Proceedings of the 36th International Conference on Neural Information Processing Systems, NIPS ’22, 2022. URLhttps://openreview.net/forum?id=_VjQlMeSB_J

work page 2022
[20]

Singhal,S

K. Singhal,S. Azizi,T. Tu,S. S. Mahdavi,J. Wei,H. W. Chung,N. Scales, A. Tanwani,H. Cole-Lewis,S. Pfohl,P. Payne,M. Seneviratne,P. Gamble, C. Kelly, A. Babiker, N. Schärli, A. Chowdhery, P. Mansfield, D. Demner- Fushman, B. Agüera y Arcas, D. Webster, G. S. Corrado, Y. Matias, K. Chou, J. Gottweis, N. Tomasev, Y. Liu, A. Rajkomar, J. Barral, C. Semturs, A...

work page doi:10.1038/s41586-023-06291-2 2023
[21]

Dagdelen, A

J. Dagdelen, A. Dunn, S. Lee, N. Walker, A. S. Rosen, G. Ceder, K. A. Persson, A. Jain, Structured information extraction from scientific text 46 with large language models, Nature Communications 15 (1) (2024) 1418. doi:10.1038/s41467-024-45563-x

work page doi:10.1038/s41467-024-45563-x 2024
[22]

J. Zhou, H. Li, S. Chen, Z. Chen, Z. Han, X. Gao, Large language models in biomedicine and healthcare, npj Artificial Intelligence 1 (1) (2025) 44. doi:10.1038/s44387-025-00047-1

work page doi:10.1038/s44387-025-00047-1 2025
[23]

Zaitoun,T

A. Zaitoun,T. Sagi,M. Peleg,Generating ontology-learning training-data through verbalization, Proceedings of the AAAI Symposium Series 4 (1) (2024) 233–241.doi:10.1609/aaaiss.v4i1.31797

work page doi:10.1609/aaaiss.v4i1.31797 2024
[24]

Peleg, N

M. Peleg, N. Veggiotti, L. Sacchi, S. Wilk, How can we reward you? a compliance and reward ontology (caro) for eliciting quantitative reward rules for engagement in mhealth app and healthy behaviors, Journal of Biomedical Informatics 154 (2024) 104655.doi:10.1016/j.jbi.2024. 104655

work page doi:10.1016/j.jbi.2024 2024
[25]

Mettes, M

P. Mettes, M. Ghadimi Atigh, M. Keller-Ressel, J. Gu, S. Yeung, Hyper- bolic deep learning in computer vision: A survey, International Jour- nal of Computer Vision 132 (9) (2024) 3484–3508. doi:10.1007/ s11263-024-02043-5

work page 2024
[26]

Bonnabel, Stochastic gradient descent on riemannian manifolds, IEEE Transactions on Automatic Control 58 (9) (2013) 2217–2229.doi:10

S. Bonnabel, Stochastic gradient descent on riemannian manifolds, IEEE Transactions on Automatic Control 58 (9) (2013) 2217–2229.doi:10. 1109/TAC.2013.2254619

work page arXiv 2013
[27]

Becigneul, O.-E

G. Becigneul, O.-E. Ganea, Riemannian adaptive optimization methods, in: International Conference on Learning Representations, 2019. URLhttps://openreview.net/forum?id=r1eiqi09K7

work page 2019
[28]

T. Yu, T. J. Liu, A. Tseng, C. D. Sa, Shadow cones: A generalized framework for partial order embeddings, in: The Twelfth International Conference on Learning Representations, 2024. URLhttps://openreview.net/forum?id=zbKcFZ6Dbp

work page 2024
[29]

Gulcehre, M

C. Gulcehre, M. Denil, M. Malinowski, A. Razavi, R. Pascanu, K. M. Hermann, P. Battaglia, V. Bapst, D. Raposo, A. Santoro, N. de Freitas, Hyperbolic attention networks, in: International Conference on Learning Representations, 2019. URLhttps://openreview.net/forum?id=rJxHsjRqFQ 47

work page 2019
[30]

M. Yang, M. Zhou, R. Ying, Y. Chen, I. King, Hyperbolic representation learning: revisiting and advancing, in: Proceedings of the 40th Interna- tional Conference on Machine Learning, ICML’23, 2023. URLhttps://proceedings.mlr.press/v202/yang23u/yang23u.pdf

work page 2023
[31]

M. Zhou, M. Yang, B. Xiong, H. Xiong, I. King, Hyperbolic graph neural networks: A tutorial on methods and applications, in: Proceedings of the 29th ACM SIGKDD Conference on Knowledge Discovery and Data Mining,KDD ’23,2023,pp. 5843–5844.doi:10.1145/3580305.3599562

work page doi:10.1145/3580305.3599562 2023
[32]

Tifrea, G

A. Tifrea, G. Becigneul, O.-E. Ganea, Poincaré GloVe: Hyperbolic word embeddings, in: International Conference on Learning Representations, 2019. URLhttps://openreview.net/forum?id=Ske5r3AqK7

work page 2019
[33]

Monath, M

N. Monath, M. Zaheer, D. Silva, A. McCallum, A. Ahmed, Gradient- based hierarchical clustering using continuous representations of trees in hyperbolicspace,in: Proceedings ofthe 25thACM SIGKDD International Conference on Knowledge Discovery & Data Mining, KDD ’19, 2019, p. 714–722.doi:10.1145/3292500.3330997

work page doi:10.1145/3292500.3330997 2019
[34]

Bordes,N

A. Bordes,N. Usunier,A. Garcia-Duran,J. Weston,O. Yakhnenko,Trans- lating embeddings for modeling multi-relational data, in: Proceedings of the 27th International Conference on Neural Information Processing Systems - Volume 2, Vol. 26, 2013, pp. 2287–2795. URL https://proceedings.neurips.cc/paper_files/paper/2013/ file/1cecc7a77928ca8133fa24680a88d2f9-Paper.pdf

work page 2013
[35]

B. Yang, W. Yih, X. He, J. Gao, L. Deng, Embedding entities and relations for learning and inference in knowledge bases, in: International Conference on Learning Representations, 2015. URLhttps://scottyih.org/files/ICLR2015_updated.pdf

work page 2015
[36]

Trouillon, J

T. Trouillon, J. Welbl, S. Riedel, E. Gaussier, G. Bouchard, Complex embeddings for simple link prediction, in: Proceedings of The 33rd Inter- national Conference on Machine Learning, Vol. 48 of ICML’16, 2016, pp. 2071–2080. URLhttps://proceedings.mlr.press/v48/trouillon16.html 48

work page 2016
[37]

Sun, Z.-H

Z. Sun, Z.-H. Deng, J.-Y. Nie, J. Tang, Rotate: Knowledge graph embed- ding by relational rotation in complex space, in: International Conference on Learning Representations, 2019. URLhttps://openreview.net/forum?id=HkgEQnRqYQ

work page 2019
[38]

T. Le, N. Le, B. Le, Knowledge graph embedding by relational rotation and complex convolution for link prediction, Expert Syst. Appl. 214 (C) (2023).doi:10.1016/j.eswa.2022.119122

work page doi:10.1016/j.eswa.2022.119122 2023
[39]

Nickel, K

M. Nickel, K. P. Murphy, V. Tresp, E. Gabrilovich, A review of relational machine learning for knowledge graphs, Proceedings of the IEEE 104 (2016) 11–33. URL https://ieeexplore.ieee.org/ielaam/5/7360840/ 7358050-aam.pdf

work page 2016
[40]

Balažević, C

I. Balažević, C. Allen, T. Hospedales, Multi-relational poincaré graph embeddings, in: Proceedings of the 33rd International Conference on Neural Information Processing Systems, 2019, pp. 4463–4473. URL https://proceedings.neurips.cc/paper_files/paper/2019/ file/f8b932c70d0b2e6bf071729a4fa68dfc-Paper.pdf

work page 2019
[41]

Chami, R

I. Chami, R. Ying, C. Re, J. Leskovec, Hyperbolic graph convolutional neural networks, in: Proceedings of the 33rd International Conference on Neural Information Processing Systems, 2019, pp. 4868–4879. URLhttps://dl.acm.org/doi/10.5555/3454287.3454725

work page doi:10.5555/3454287.3454725 2019
[42]

Chami, A

I. Chami, A. Wolf, D.-C. Juan, F. Sala, S. Ravi, C. Ré, Low-dimensional hyperbolic knowledge graph embeddings, in: Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, 2020, pp. 6901–6914. URLhttps://aclanthology.org/2020.acl-main.617.pdf

work page 2020
[43]

Zheng, B

Z. Zheng, B. Zhou, H. Yang, Z. Tan, A. Waaler, E. Kharlamov, A. Soylu, Low-dimensional hyperbolic knowledge graph embedding for better ex- trapolation to under-represented data, in: The Semantic Web: 21st In- ternational Conference, ESWC 2024, Hersonissos, Crete, Greece, May 26–30, 2024, Proceedings, Part I, 2024, pp. 100–120.doi:10.1007/ 978-3-031-60626-7_6. 49

work page 2024
[44]

T. N. Kipf, M. Welling, Semi-supervised classification with graph convolu- tionalnetworks,in: InternationalConferenceon LearningRepresentations, 2017. URLhttps://openreview.net/pdf?id=SJU4ayYgl

work page 2017
[45]

Schlichtkrull, T

M. Schlichtkrull, T. N. Kipf, P. Bloem, R. van den Berg, I. Titov, M. Welling, Modeling relational data with graph convolutional networks, in: The Semantic Web, 2018, pp. 593–607. URL https://link.springer.com/chapter/10.1007/ 978-3-319-93417-4_38

work page 2018
[46]

Veličković, G

P. Veličković, G. Cucurull, A. Casanova, A. Romero, P. Liò, Y. Bengio, Graph attention networks, in: International Conference on Learning Representations, 2018. URLhttps://openreview.net/forum?id=rJXMpikCZ

work page 2018
[47]

Y. Choi, C. Y.-I. Chiu, D. Sontag, Learning low-dimensional representa- tions of medical concepts, AMIA Joint Summits on Translational Science 2016:41-50 (2016). URLhttps://pmc.ncbi.nlm.nih.gov/articles/PMC5001761/

work page 2016
[48]

J. Kim, D. Kim, K.-A. Sohn, Hig2vec: Hierarchical representations of gene ontology and genes in the poincaré ball, Bioinformatics 37 (18) (2021) 2971–2980.doi:10.1093/bioinformatics/btab193

work page doi:10.1093/bioinformatics/btab193 2021
[49]

Y. He, Z. Yuan, J. Chen, I. Horrocks, Language models as hierarchy encoders, in: Proceedings of the 38th International Conference on Neural Information Processing Systems, NIPS ’24, 2024, pp. 14690–14711. URLhttps://doi.org/10.5281/zenodo.14036213

work page doi:10.5281/zenodo.14036213 2024
[50]

Dhingra, C

B. Dhingra, C. Shallue, M. Norouzi, A. Dai, G. Dahl, Embedding text in hyperbolic spaces, in: Proceedings of the Twelfth Workshop on Graph- Based Methods for Natural Language Processing (TextGraphs-12), 2018, pp. 59–69. URLhttps://aclanthology.org/W18-1708.pdf

work page 2018
[51]

Muscoloni, J

A. Muscoloni, J. M. Thomas, S. Ciucci, G. Bianconi, C. V. Cannistraci, Machine learning meets complex networks via coalescent embedding in the hyperbolic space, Nature Communications 8 (1) (2017) 1615. doi:10.1038/s41467-017-01825-5. 50

work page doi:10.1038/s41467-017-01825-5 2017
[52]

T. Li,A. K. Sahu,A. Talwalkar,V. Smith,Federated learning: Challenges, methods, and future directions, IEEE Signal Processing Magazine 37 (3) (2020) 50–60.doi:10.1109/MSP.2020.2975749

work page doi:10.1109/msp.2020.2975749 2020
[53]

Shazeer, A

N. Shazeer, A. Mirhoseini, K. Maziarz, A. Davis, Q. Le, G. Hinton, J. Dean, Outrageously large neural networks: The sparsely-gated mixture- of-expertslayer,in: InternationalConferenceonLearningRepresentations, 2017. URLhttps://openreview.net/forum?id=B1ckMDqlg

work page 2017
[54]

Y. A. Malkov, D. A. Yashunin, Efficient and robust approximate nearest neighbor search using hierarchical navigable small world graphs, IEEE Transactions on Pattern Analysis and Machine Intelligence (TPAMI) 42 (4) (2020) 824–836.doi:10.1109/TPAMI.2018.2889473

work page doi:10.1109/tpami.2018.2889473 2020
[55]

Johnson, M

J. Johnson, M. Douze, H. Jégou, Billion-scale similarity search with gpus, IEEE Transactions on Big Data 7 (3) (2021) 535–547.doi:10.1109/ TBDATA.2019.2921572

work page arXiv 2021
[56]

Prokhorenkova, D

L. Prokhorenkova, D. Baranchuk, N. Bogachev, Y. Demidovich, A. Kol- pakov, Graph-based nearest neighbor search in hyperbolic spaces, in: International Conference on Learning Representations, 2022. URLhttps://openreview.net/forum?id=USIgIY6TNDe

work page 2022
[57]

Kisfaludi-Bak, G

S. Kisfaludi-Bak, G. van Wordragen, A Quadtree, a Steiner Spanner, and Approximate Nearest Neighbours in Hyperbolic Space, in: 40th International Symposium on Computational Geometry (SoCG 2024), Vol. 293 of Leibniz International Proceedings in Informatics (LIPIcs), 2024, pp. 68:1–68:15.doi:10.4230/LIPIcs.SoCG.2024.68

work page doi:10.4230/lipics.socg.2024.68 2024
[58]

E. Park, A. Vigneron, Embeddings and near-neighbor searching with constant additive error for hyperbolic spaces, Computational Geometry 126 (2025) 102150.doi:10.1016/j.comgeo.2024.102150

work page doi:10.1016/j.comgeo.2024.102150 2025
[59]

Z. Qiu, J. Liu, Y. Chen, I. King, Hihpq: hierarchical hyperbolic product quantization for unsupervised image retrieval, in: Proceedings of the 38th AAAI Conference on Artificial Intelligence and 36th Conference on Innovative Applications of Artificial Intelligence and 40th Symposium on Educational Advances in Artificial Intelligence, 2024.doi:10.1609/ aaa...

work page 2024
[60]

Robertson, H

S. Robertson, H. Zaragoza, The probabilistic relevance framework: Bm25 andbeyond,Foundations andTrends in Information Retrieval3 (4) (2009) 333–389.doi:10.1561/1500000019

work page doi:10.1561/1500000019 2009
[61]

Karpukhin, B

V. Karpukhin, B. Oguz, S. Min, P. Lewis, L. Wu, S. Edunov, D. Chen, W. Yih, Dense passage retrieval for open-domain question answering, in: Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), 2020, pp. 6769–6781.doi:10.18653/ v1/2020.emnlp-main.550

work page 2020
[62]

P. Lewis,E. Perez,A. Piktus,F. Petroni,V. Karpukhin,N. Goyal,H. Küt- tler, M. Lewis, W.-t. Yih, T. Rocktäschel, S. Riedel, D. Kiela, Retrieval- augmented generation for knowledge-intensive nlp tasks, in: Proceedings of the 34th International Conference on Neural Information Processing Systems, NIPS ’20, 2020, pp. 9459–9474. URLhttps://dl.acm.org/doi/abs/1...

work page doi:10.5555/3495724.3496517 2020
[63]

Izacard, E

G. Izacard, E. Grave, Leveraging passage retrieval with generative models for open domain question answering, in: Proceedings of the 16th Con- ference of the European Chapter of the Association for Computational Linguistics: Main Volume, 2021, pp. 874–880.doi:10.18653/v1/2021. eacl-main.74

work page doi:10.18653/v1/2021 2021
[64]

Izacard, P

G. Izacard, P. Lewis, M. Lomeli, L. Hosseini, F. Petroni, T. Schick, J. Dwivedi-Yu, A. Joulin, S. Riedel, E. Grave, Atlas: few-shot learning with retrieval augmented language models, J. Mach. Learn. Res. 24 (1) (2023). URL https://www.jmlr.org/papers/volume24/23-0037/23-0037. pdf

work page 2023
[65]

L. Gao, X. Ma, J. Lin, J. Callan, Precise zero-shot dense retrieval without relevance labels, in: Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 2022, pp. 1762–1777.doi:10.18653/v1/2023.acl-long.99

work page doi:10.18653/v1/2023.acl-long.99 2022
[66]

Khattab, M

O. Khattab, M. Zaharia, Colbert: Efficient and effective passage search via contextualized late interaction over bert, in: Proceedings of the 43rd International ACM SIGIR Conference on Research and Development in Information Retrieval, SIGIR ’20, 2020, pp. 39–48.doi:10.1145/ 3397271.3401075. 52

work page arXiv 2020
[67]

A. Asai, Z. Wu, Y. Wang, A. Sil, H. Hajishirzi, Self-RAG: Learning to retrieve, generate, and critique through self-reflection, in: The Twelfth International Conference on Learning Representations, 2024. URLhttps://openreview.net/forum?id=hSyW5go0v8

work page 2024
[68]

N. He, H. Madhu, N. Bui, M. Yang, R. Ying, Hyperbolic deep learning for foundation models: A survey, in: Proceedings of the 31st ACM SIGKDD Conference on Knowledge Discovery and Data Mining V.2, KDD ’25, 2025, pp. 6021–6031.doi:10.1145/3711896.3736564. 53

work page doi:10.1145/3711896.3736564 2025