pith. machine review for the scientific record. sign in

arxiv: 2604.06752 · v1 · submitted 2026-04-08 · 💻 cs.LG

Recognition: 2 theorem links

· Lean Theorem

Busemann energy-based attention for emotion analysis in Poincar\'e discs

Zinaid Kapi\'c , Vladimir Ja\'cimovi\'c
Authors on Pith no claims yet

Pith reviewed 2026-05-10 17:52 UTC · model grok-4.3

classification 💻 cs.LG
keywords hyperbolic geometryPoincaré discemotion analysisBusemann energyattention mechanismaffective computingtext classificationdeep learning
0
0 comments X

The pith

EmBolic places emotion analysis in the Poincaré disc and scores alignments with Busemann energy to capture semantic hierarchies.

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

The paper introduces EmBolic, a fully hyperbolic neural architecture for classifying emotions in text messages. It treats emotions as points in a continuous space whose metric structure reflects hierarchical relationships created by word ambiguities rather than as isolated categories. Text inputs generate query points inside the disc while emotion classes appear as keys on the boundary; the Busemann energy between each query and the keys determines the prediction. The model therefore learns both the placement of messages and the directions that best represent each emotion. Experiments indicate that this setup maintains accuracy and generalizes reliably even when the embedding dimension is kept small.

Core claim

EmBolic is a novel fully hyperbolic deep learning architecture for fine-grained emotion analysis from textual messages. The underlying idea is that hyperbolic geometry efficiently captures hierarchies between both words and emotions that arise from semantic ambiguities. EmBolic aims to infer the curvature on the continuous space of emotions rather than treating them as a categorical set without any metric structure. The model generates queries from text and lets keys emerge automatically at the boundary of the disc; predictions rest on the Busemann energy between those queries and keys, measuring alignment with class directions.

What carries the argument

The Busemann energy attention mechanism inside the Poincaré disc, in which text-derived query points are scored against automatically generated boundary keys that represent emotion classes.

If this is right

  • Affective computing tasks benefit from hyperbolic representations that preserve hierarchy induced by ambiguity.
  • The architecture can infer a metric curvature on the space of emotions instead of using flat categorical labels.
  • Prediction accuracy stays reasonable even when the representation dimension is reduced.
  • Strong generalization is observed across different textual emotion datasets.

Where Pith is reading between the lines

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

  • The same query-to-boundary alignment could be tested on other NLP problems that involve graded or overlapping categories, such as multi-label intent detection.
  • Because performance holds in low dimensions, the approach may reduce memory and compute costs for real-time emotion monitoring on edge devices.
  • The automatic emergence of class keys suggests the model discovers emotion prototypes directly from data rather than requiring hand-specified directions.

Load-bearing premise

Hyperbolic geometry is especially good at representing the hierarchical relations among words and emotions that come from semantic ambiguities, and the Busemann energy supplies a sufficient score for deciding which emotion a message expresses.

What would settle it

An otherwise identical Euclidean attention model trained on the same emotion datasets achieves equal or higher accuracy and generalization across the same range of embedding dimensions.

Figures

Figures reproduced from arXiv: 2604.06752 by Vladimir Ja\'cimovi\'c, Zinaid Kapi\'c.

Figure 1
Figure 1. Figure 1: Circumplex model of affect by Russell [11]. Permissions obtained. Furthermore, Scherer with collaborators [15] developed an instrument named Geneva Emotion Wheel for measuring emotional reactions to objects, events and situations. Circumplex models have also been deployed for the emotion analysis from facial expressions. The paper [16] proposed a method for computing emotional parameters from facial expres… view at source ↗
Figure 2
Figure 2. Figure 2: shows geodesics and distances in the Poincaré disc. A  [PITH_FULL_IMAGE:figures/full_fig_p005_2.png] view at source ↗
Figure 3
Figure 3. Figure 3 [PITH_FULL_IMAGE:figures/full_fig_p008_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Hyperbolic GloVe word embeddings in the Poincaré disc [PITH_FULL_IMAGE:figures/full_fig_p009_4.png] view at source ↗
Figure 5
Figure 5. Figure 5: Randomly selected words with their representations in the Poincaré disc. This illustrates emotion-related semantic ambiguity of certain words. representations as points in the disc. Embeddings of instances annotated by three emotions in the disc are depicted in [PITH_FULL_IMAGE:figures/full_fig_p009_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Learned representations of textual messages for Desire, Love, and Nervousness (top row) and the corre￾sponding representations after isometric (Möbius) correction (bottom row). between emotions and angles. These angles correspond to dominant directions, calculated as the weighted average, where the weight of each point is proportional to the square of its modulus ψj = ∑m k=1 r 2 kφk. Here, m is the number … view at source ↗
Figure 7
Figure 7. Figure 7: Visualization of the validation results. Representations of test messages for Desire, Love and Nervousness and the radii encoding these emotions. Correct predictions correspond to the points which are placed on the radii. The confidence levels are encoded by the moduli (distances from the center) [PITH_FULL_IMAGE:figures/full_fig_p012_7.png] view at source ↗
Figure 8
Figure 8. Figure 8: Emotion correlation matrix learned by EmBolic. Each cell (i, j) show how frequently emotion i appears together with j in top 5. Darker cells indicate stronger pairwise correlations [PITH_FULL_IMAGE:figures/full_fig_p014_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Map of emotions in the Poincaré disc learned from the co-occurrence matrix of emotions depicted in [PITH_FULL_IMAGE:figures/full_fig_p014_9.png] view at source ↗
read the original abstract

We present EmBolic - a novel fully hyperbolic deep learning architecture for fine-grained emotion analysis from textual messages. The underlying idea is that hyperbolic geometry efficiently captures hierarchies between both words and emotions. In our context, these hierarchical relationships arise from semantic ambiguities. EmBolic aims to infer the curvature on the continuous space of emotions, rather than treating them as a categorical set without any metric structure. In the heart of our architecture is the attention mechanism in the hyperbolic disc. The model is trained to generate queries (points in the hyperbolic disc) from textual messages, while keys (points at the boundary) emerge automatically from the generated queries. Predictions are based on the Busemann energy between queries and keys, evaluating how well a certain textual message aligns with the class directions representing emotions. Our experiments demonstrate strong generalization properties and reasonably good prediction accuracy even for small dimensions of the representation space. Overall, this study supports our claim that affective computing is one of the application domains where hyperbolic representations are particularly advantageous.

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 EmBolic, a novel fully hyperbolic deep learning architecture for fine-grained emotion analysis from textual messages using Poincaré discs. The model generates queries from text and derives boundary keys automatically, basing predictions on the Busemann energy between queries and keys to align with emotion class directions. It claims that hyperbolic geometry captures hierarchies arising from semantic ambiguities and demonstrates strong generalization and good prediction accuracy even in small representation dimensions.

Significance. If the experimental results hold, this work would provide evidence that hyperbolic representations are particularly advantageous for affective computing tasks, enabling effective modeling of hierarchical semantic structures with low-dimensional embeddings and potentially improving generalization in emotion classification.

major comments (2)
  1. [Abstract] Abstract: The abstract asserts experimental results including 'strong generalization properties' and 'reasonably good prediction accuracy' but provides no details on the datasets used, baseline models, training procedures, quantitative metrics, error bars, or statistical tests. This makes it impossible to determine whether the performance follows from the Busemann energy-based attention or from other design choices, which is central to validating the main claim.
  2. [Architecture description] Architecture description: The statement that keys 'emerge automatically from the generated queries' risks circularity, as the alignment score via Busemann energy may depend on the same learned mapping that produces the queries. The paper should clarify how this setup ensures an independent and meaningful alignment measure, perhaps by providing a parameter-free derivation or external benchmark.
minor comments (1)
  1. The abstract contains a LaTeX escape sequence 'Poincaré' which should be rendered properly in the final version.

Simulated Author's Rebuttal

2 responses · 0 unresolved

Thank you for the opportunity to respond to the referee's comments on our manuscript arXiv:2604.06752. We address each major comment below and propose targeted revisions to improve clarity and support for our claims regarding EmBolic.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The abstract asserts experimental results including 'strong generalization properties' and 'reasonably good prediction accuracy' but provides no details on the datasets used, baseline models, training procedures, quantitative metrics, error bars, or statistical tests. This makes it impossible to determine whether the performance follows from the Busemann energy-based attention or from other design choices, which is central to validating the main claim.

    Authors: We agree that the abstract is high-level and would benefit from additional specifics to better contextualize the results. Abstracts are length-constrained, but we will revise it to name the evaluation datasets, report key quantitative metrics (accuracy and generalization measures), and reference the baselines. The full experimental details—including training procedures, metrics with error bars, and statistical tests—are already provided in the Experiments section. Our ablation studies and direct comparisons to Euclidean and other hyperbolic baselines demonstrate that the reported performance and generalization arise specifically from the Busemann energy-based attention rather than ancillary design choices; we will add a concise statement to the abstract highlighting this attribution. revision: yes

  2. Referee: [Architecture description] Architecture description: The statement that keys 'emerge automatically from the generated queries' risks circularity, as the alignment score via Busemann energy may depend on the same learned mapping that produces the queries. The paper should clarify how this setup ensures an independent and meaningful alignment measure, perhaps by providing a parameter-free derivation or external benchmark.

    Authors: We thank the referee for identifying this potential ambiguity. The queries are produced by a dedicated hyperbolic text encoder that maps input embeddings into the Poincaré disc. The keys are class-specific ideal points on the boundary, learned as directional prototypes for each emotion; they are not outputs of the query-generation mapping. The Busemann energy itself is the standard, parameter-free Busemann function of hyperbolic geometry, which measures alignment between an interior point and a boundary direction using only the fixed hyperbolic metric. We will revise the architecture section to explicitly separate these components with equations, remove the potentially misleading phrasing, and include an ablation isolating the contribution of the Busemann scoring. This will confirm the alignment measure is geometrically independent and meaningful. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained

full rationale

The paper applies standard hyperbolic geometry tools (Poincaré disc model and Busemann functions) to a supervised text classification task. Queries are generated from input text via a learned mapping, keys are derived from those queries as class-direction representatives, and alignment is scored via Busemann energy; this is an ordinary attention-style architecture whose parameters are optimized against labeled data. No equation or claim reduces a prediction to its own fitted inputs by construction, no uniqueness theorem is imported from the authors' prior work, and no ansatz is smuggled via self-citation. The central claim rests on experimental generalization results rather than tautological redefinition of inputs as outputs.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

The central claim rests on the premise that semantic hierarchies in language are faithfully represented by hyperbolic geometry and that Busemann energy provides a meaningful alignment score. No explicit free parameters, axioms, or invented entities are enumerated in the abstract, but the learned curvature and the automatic key emergence are implicit modeling choices.

pith-pipeline@v0.9.0 · 5475 in / 1221 out tokens · 35315 ms · 2026-05-10T17:52:59.619733+00:00 · methodology

discussion (0)

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

Lean theorems connected to this paper

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

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

46 extracted references · 31 canonical work pages

  1. [1]

    Low distortion delaunay embedding of trees in hyperbolic plane

    Rik Sarkar. Low distortion delaunay embedding of trees in hyperbolic plane. In International symposium on graph drawing, pages 355–366. Springer, 2011. doi: 10.1007/978-3-642-25878-7_34

    work page doi:10.1007/978-3-642-25878-7_34 2011

  2. [2]

    Hyperbolic geometry of complex networks

    Dmitri Krioukov, Fragkiskos Papadopoulos, Maksim Kitsak, Amin V ahdat, and Marián Boguná. Hyperbolic geometry of complex networks. Physical Review EStatistical, Nonlinear , and Soft Matter Physics, 82(3):036106,

  3. [3]

    doi: 10.1103/PhysRevE.82.036106

    work page doi:10.1103/physreve.82.036106

  4. [4]

    Poincaré embeddings for learning hierarchical representations

    Maximillian Nickel and Douwe Kiela. Poincaré embeddings for learning hierarchical representations. Advances in neural information processing systems , 30, 2017

    2017

  5. [5]

    Poincaré glove: Hyperbolic word embeddings

    Alexandru Tifrea, Gary Bécigneul, and Octavian-Eugen Ganea. Poincaré glove: Hyperbolic word embeddings. arXiv preprint arXiv:1810.06546, 2018. doi: 10.48550/arXiv.1810.06546

    work page doi:10.48550/arxiv.1810.06546 2018

  6. [6]

    Skip-gram word embeddings in hyperbolic space

    M Leimeister and BJ Wilson. Skip-gram word embeddings in hyperbolic space. arXiv preprint arXiv:1809.01498, 2018. doi: 10.48550/arXiv.1809.01498

    work page doi:10.48550/arxiv.1809.01498 2018

  7. [7]

    Learning continuous hierarchies in the lorentz model of hyperbolic geom- etry

    Maximillian Nickel and Douwe Kiela. Learning continuous hierarchies in the lorentz model of hyperbolic geom- etry. In International conference on machine learning , pages 3779–3788. PMLR, 2018

    2018

  8. [8]

    Representation tradeoffs for hyperbolic embeddings

    Frederic Sala, Chris De Sa, Albert Gu, and Christopher Ré. Representation tradeoffs for hyperbolic embeddings. In International conference on machine learning , pages 4460–4469. PMLR, 2018

    2018

  9. [9]

    A graph-based algo- rithm for inducing lexical taxonomies from scratch

    Sarang Patil, Zeyong Zhang, Yiran Huang, Tengfei Ma, and Mengjia Xu. Hyperbolic large language models. arXiv preprint arXiv:2509.05757, 2025. doi: 10.48550/arXiv.2509.05757

    work page doi:10.48550/arxiv.2509.05757 2025

  10. [10]

    Affective computing

    Rosalind W Picard. Affective computing. MIT press, 2000

    2000

  11. [11]

    Fuzzy concepts in a fuzzy hierarchy: varieties of anger

    James A Russell and Beverly Fehr. Fuzzy concepts in a fuzzy hierarchy: varieties of anger. Journal of personality and social psychology, 67(2):186, 1994. doi: 10.1037/0022-3514.67.2.186

    work page doi:10.1037/0022-3514.67.2.186 1994

  12. [12]

    A Circumplex Model of Affect

    James A Russell. A circumplex model of affect. Journal of personality and social psychology , 39(6):1161, 1980. doi:10.1037/h0077714

    work page doi:10.1037/h0077714 1980

  13. [13]

    A cross-cultural study of a circumplex model of affect

    James A Russell, Maria Lewicka, and Toomas Niit. A cross-cultural study of a circumplex model of affect. Journal of personality and social psychology , 57(5):848, 1989. doi: 10.1037/0022-3514.57.5.848

    work page doi:10.1037/0022-3514.57.5.848 1989

  14. [14]

    What are emotions? and how can they be measured? Social science information , 44(4): 695–729, 2005

    Klaus R Scherer. What are emotions? and how can they be measured? Social science information , 44(4): 695–729, 2005. doi: 10.1177/0539018405058216

    work page doi:10.1177/0539018405058216 2005

  15. [15]

    Seeing stars of valence and arousal in blog posts

    Georgios Paltoglou and Michael Thelwall. Seeing stars of valence and arousal in blog posts. IEEE Transactions on Affective Computing, 4(1):116–123, 2012. doi: 10.1109/T-AFFC.2012.36

    work page doi:10.1109/t-affc.2012.36 2012

  16. [16]

    2013 , PAGES =

    Klaus R Scherer, V era Shuman, Johnny RJ Fontaine, and Cristina Soriano. The grid meets the wheel: Assess- ing emotional feeling via self-report. Components of emotional meaning: A sourcebook , 53:1689–1699, 2013. doi:10.1093/acprof:oso/9780199592746.003.0019

    work page doi:10.1093/acprof:oso/9780199592746.003.0019 2013

  17. [17]

    Asymmetrical facial expressions based on an advanced interpretation of two-dimensional russells emotional model

    Junghyun Ahn, Stephane Gobron, Quentin Silvestre, and Daniel Thalmann. Asymmetrical facial expressions based on an advanced interpretation of two-dimensional russells emotional model. Proceedings of ENGAGE, 2, 2010

    2010

  18. [18]

    C. Tomasi. The Circumplex Model of Affects: Understanding the Science Behind It and How Emotion AI Leverages Its Power. https://www.morphcast.com/blog/circumplex-model-of-affects/ , 2024

    2024

  19. [19]

    Semeval-2007 task 14: Affective text

    Carlo Strapparava and Rada Mihalcea. Semeval-2007 task 14: Affective text. In Proceedings of the fourth international workshop on semantic evaluations (SemEval-2007) , pages 70–74, 2007

    2007

  20. [20]

    Semeval-2018 task 1: Affect in tweets

    Saif Mohammad, Felipe Bravo-Marquez, Mohammad Salameh, and Svetlana Kiritchenko. Semeval-2018 task 1: Affect in tweets. In Proceedings of the 12th international workshop on semantic evaluation , pages 1–17, 2018. doi:10.18653/v1/S18-1001

    work page doi:10.18653/v1/s18-1001 2018

  21. [21]

    Dens: A dataset for multi-class emotion analysis

    Chen Liu, Muhammad Osama, and Anderson De Andrade. Dens: A dataset for multi-class emotion analysis. arXiv preprint arXiv:1910.11769, 2019. doi: 10.18653/v1/D19-1656

    work page doi:10.18653/v1/d19-1656 1910

  22. [22]

    An argument for basic emotions

    Paul Ekman. An argument for basic emotions. Cognition & emotion , 6(3-4):169–200, 1992. doi:10.1080/02699939208411068

    work page doi:10.1080/02699939208411068 1992

  23. [23]

    A general psychoevolutionary theory of emotion

    Robert Plutchik. A general psychoevolutionary theory of emotion. In Theories of emotion, pages 3–33. Elsevier,

  24. [24]

    doi: 10.1016/B978-0-12-558701-3.50007-7

    work page doi:10.1016/b978-0-12-558701-3.50007-7

  25. [25]

    G o E motions: A Dataset of Fine-Grained Emotions

    Dorottya Demszky, Dana Movshovitz-Attias, Jeongwoo Ko, Alan Cowen, Gaurav Nemade, and Sujith Ravi. GoEmotions: A dataset of fine-grained emotions. In Proceedings of the 58th Annual Meeting of the As- sociation for Computational Linguistics , pages 4040–4054. Association for Computational Linguistics, 2020. doi:10.18653/v1/2020.acl-main.372. 20 Busemann ene...

    work page doi:10.18653/v1/2020.acl-main.372 2020

  26. [26]

    Fine-grained emotion detection in health-related online posts

    Hamed Khanpour and Cornelia Caragea. Fine-grained emotion detection in health-related online posts. In Proceedings of the 2018 conference on empirical methods in natural language processing , pages 1160–1166,

    2018

  27. [27]

    doi: 10.18653/v1/D18-1147

    work page doi:10.18653/v1/d18-1147

  28. [28]

    Bert: Pre-training of deep bidirectional transformers for language understanding

    Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. Bert: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 conference of the North American chapter of the association for computational linguistics: human language technologies, volume 1 (long and short papers), pages 4171–4186, 2019

    2019

  29. [29]

    arXiv preprint arXiv:2403.06108 , year=

    Kaipeng Wang, Zhi Jing, Y ongye Su, and Yikun Han. Large language models on fine-grained emotion detection dataset with data augmentation and transfer learning. arXiv preprint arXiv:2403.06108 , 2024. doi:10.48550/arXiv.2403.06108

    work page doi:10.48550/arxiv.2403.06108 2024

  30. [30]

    Ju- rafsky, J

    Da Yin, Tao Meng, and Kai-Wei Chang. Sentibert: A transferable transformer-based architecture for composi- tional sentiment semantics. arXiv preprint arXiv:2005.04114, 2020. doi: 10.18653/v1/2020.acl-main.341

    work page doi:10.18653/v1/2020.acl-main.341 2005

  31. [31]

    emlm: A new pre-training objective for emotion related tasks

    Tiberiu Sosea and Cornelia Caragea. emlm: A new pre-training objective for emotion related tasks. In Pro- ceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th Interna- tional Joint Conference on Natural Language Processing (V olume 2: Short Papers) , pages 286–293, 2021. doi:10.18653/v1/2021.acl-short.38

    work page doi:10.18653/v1/2021.acl-short.38 2021

  32. [32]

    Not all negatives are equal: Label-aware contrastive loss for fine-grained text classification

    V arsha Suresh and Desmond Ong. Not all negatives are equal: Label-aware contrastive loss for fine-grained text classification. In Proceedings of the 2021 conference on empirical methods in natural language processing , pages 4381–4394, 2021. doi: 10.18653/v1/2021.emnlp-main.359

    work page doi:10.18653/v1/2021.emnlp-main.359 2021

  33. [33]

    Cluster-level contrastive learning for emotion recognition in conversations

    Kailai Y ang, Tianlin Zhang, Hassan Alhuzali, and Sophia Ananiadou. Cluster-level contrastive learning for emotion recognition in conversations. IEEE Transactions on Affective Computing , 14(4):3269–3280, 2023. doi:10.1109/TAFFC.2023.3243463

    work page doi:10.1109/taffc.2023.3243463 2023

  34. [34]

    Rethinking tabular data understanding with large language models

    Fangxu Y u, Junjie Guo, Zhen Wu, and Xinyu Dai. Emotion-anchored contrastive learning framework for emotion recognition in conversation. In Findings of the Association for Computational Linguistics: NAACL 2024 , pages 4521–4534, 2024. doi: 10.18653/v1/2024.findings-naacl.282

    work page doi:10.18653/v1/2024 2024

  35. [35]

    Message pass- ing on semantic-anchor-graphs for fine-grained emotion representation learning and classification

    Pinyi Zhang, Jingyang Chen, Junchen Shen, Zijie Zhai, Ping Li, Jie Zhang, and Kai Zhang. Message pass- ing on semantic-anchor-graphs for fine-grained emotion representation learning and classification. In Proceed- ings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages 2771–2783, 2024. doi:10.18653/v1/2024.emnlp-main.162

    work page doi:10.18653/v1/2024.emnlp-main.162 2024

  36. [36]

    Label-aware hyperbolic embeddings for fine- grained emotion classification

    Chih-Y ao Chen, Tun Min Hung, Yi-Li Hsu, and Lun-Wei Ku. Label-aware hyperbolic embeddings for fine- grained emotion classification. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (V olume 1: Long Papers), pages 10947–10958, 2023. doi: 10.18653/v1/2023.acl-long.613

    work page doi:10.18653/v1/2023.acl-long.613 2023

  37. [37]

    Semantic alignment in hyperbolic space for fine-grained emotion classifi- cation

    Ashish Kumar and Durga Toshniwal. Semantic alignment in hyperbolic space for fine-grained emotion classifi- cation. In Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (V olume 4: Student Research Workshop), pages 806–813, 2025. doi: 10.18653/v1/2025.acl-srw.55

    work page doi:10.18653/v1/2025.acl-srw.55 2025

  38. [38]

    Visual complex analysis

    Tristan Needham. Visual complex analysis. Oxford University Press, 2023

    2023

  39. [39]

    Conformal and holomorphic barycenters in hyperbolic balls

    Vladimir Jaimovi and David Kalaj. Conformal and holomorphic barycenters in hyperbolic balls. Annales Fennici Mathematici, 50(2):407421, 2025. doi: 10.54330/afm.163349

    work page doi:10.54330/afm.163349 2025

  40. [40]

    A group-theoretic framework for machine learning in hyperbolic spaces

    Vladimir Ja ´cimovi´c. A group-theoretic framework for machine learning in hyperbolic spaces. arXiv preprint arXiv:2501.06934, 2025. doi: 10.48550/arXiv.2501.06934

    work page doi:10.48550/arxiv.2501.06934 2025

  41. [41]

    Conformally natural families of probability distributions on hyperbolic disc with a view on geometric deep learning

    Vladimir Jacimovic and Marijan Markovic. Conformally natural families of probability distributions on hyperbolic disc with a view on geometric deep learning. arXiv preprint arXiv:2407.16733 , 2024. doi:10.48550/arXiv.2407.16733

    work page doi:10.48550/arxiv.2407.16733 2024

  42. [42]

    Probabilistic matrix factorization

    Andriy Mnih and Ruslan Salakhutdinov. Probabilistic matrix factorization. in advances in neural information processing systems. Advances in neural information processing systems , 2007

    2007

  43. [43]

    1996 , journal =

    Peter McCullagh. Möbius transformation and cauchy parameter estimation. The Annals of Statistics , 24(2): 787–808, 1996. doi: 10.1214/aos/1032894465

    work page doi:10.1214/aos/1032894465 1996

  44. [44]

    Poincaré and his disk

    Étienne Ghys. Poincaré and his disk. The scientific legacy of Poincaré, 36:17, 2006

    2006

  45. [45]

    Hyperbolic busemann neural networks

    Ziheng Chen, Bernhard Schölkopf, and Nicu Sebe. Hyperbolic busemann neural networks. arXiv preprint arXiv:2602.18858, 2026. doi: 10.48550/arXiv.2602.18858

    work page doi:10.48550/arxiv.2602.18858 2026

  46. [46]

    Hyperbolic busemann learning with ideal proto- types

    Mina Ghadimi Atigh, Martin Keller-Ressel, and Pascal Mettes. Hyperbolic busemann learning with ideal proto- types. Advances in neural information processing systems , 34:103–115, 2021. 21

    2021