betaFeedback

WorksMathematics

Identifiable alignment of unpaired representations

Abstract

Aligning two models without paired data requires a distance that both share, yet existing methods use Euclidean or spherical geometry. This is either an arbitrary choice or imposed by their loss function. We show that, under a precise form of the manifold hypothesis, the pullback of the data metric gives such a distance for any two models of one data-generating process. This holds whether the models are encoders, decoders or both, and whatever their architecture, loss, prior or training data. The alignment map preserves this distance and the distribution of representations, and is identifiable if and only if the data distribution breaks every symmetry of the data manifold. We show that this symmetry breaking holds generically. Identifiability makes the alignment coupling the unique optimum of Gromov–Wasserstein and of its weighted combinations with a linear program. The linear program alone has the same optimum if almost every data point has a distinct distribution of distances. The coupling, and with it the alignment map, is recovered exactly from permuted samples, even without identifiability, and consistently from independent ones. Our method, Neural Gromov–Monge Alignment (NGMA), uses no pairs or labels and outperforms every unpaired and label-informed baseline. It relies on pullback distances to align image models with different priors and single-cell models trained on disjoint cells, where Euclidean distances fail.

Introduction

Building multimodal foundation models [53], sharing knowledge across data-privacy walls [48], and auditing proprietary vector-database security [34] all rely on learning an alignment map between two representation spaces, which sends the representation of each data point in one space to the representation of the same data point in the other. With paired data, this is a regression problem. Without pairs, individual representations are meaningless in absolute terms. What remains in each space are the relative distances between points and their distribution, which unpaired methods use to recover the alignment.

This idea rests on the Platonic representation hypothesis [31], which posits that independently trained models converge to the same similarity structure on the data, defined as a kernel on pairs of data points computed from their representations. Given two sets of representations, unpaired methods therefore learn a map that matches the kernels, in the hope that it is the alignment.

Recent methods align text encoders whose loss function places the representations on a sphere [81, 34, 13], where equal kernels make the alignment map a rotation. Their cross-modal counterpart [68] extends this to vision encoders by imposing a spherical geometry on their representations before alignment. The advances therefore predominantly cover one family of loss functions and two modalities, and whether they extend beyond these is an open problem. The case to settle first is a single modality under arbitrary loss functions, which any account of representational convergence must pass before it can extend across modalities.

The immediate problem is that both signals, the relative distances and the distribution of the representations, depend on the choice of geometry. Intuitively, leveraging them requires a distance tied to the intrinsic geometry of the data, invariant to arbitrary modelling choices. Unfortunately, the field almost universally assumes the Euclidean or spherical geometry for the representations, which the data do not necessarily determine. Standard contrastive losses build the sphere into training [81]. Balestriero and LeCun [2026] prove that, at their optimum, whitened joint-embedding predictive architectures (JEPAs) trained on the same finite set of data points with the same positive pairs share a metric, but only on those points, not on their generating distribution. In both cases, the loss function sets the geometry, so models trained with different loss functions need not share it. For general loss functions, models can freely reparameterise their representations, which distorts Euclidean distances [74, 29].

More fundamentally, the literature offers no general account of why trained models should share a similarity structure and what that structure is. Without one, we cannot say how the alignment map is defined, whether a map that matches the kernels is unique, or whether its recovery is well-posed.

In this paper, we show that aligning two models of the same data without pairs is well-posed only when both representations induce the same metric on the data. A formalised manifold hypothesis defines this metric, and treating models as charts of the data manifold defines the alignment map. With this metric, the map is uniquely identifiable if and only if the data distribution breaks the symmetries of the underlying manifold, a condition we prove holds generically. We turn this identifiability into recovery guarantees by casting the problem as transporting one distribution of representations onto the other while preserving pairwise distances. Finally, we instantiate our theory as an architecture-agnostic alignment method for representations of the same dimension. Figure 1 visualises the main contributions of the paper, the relationships between them and how they are presented.

Figure 1. Outline of Sections 2–4: the setting, identifiability and recovery from samples.

Section 2 states the metric manifold hypothesis (H1) and treats each model as a chart of the data manifold (model coordinates, H2), so the alignment map is a coordinate change. It preserves latent distances only if both models assign the same distances to data points, as the pullback of the data metric always does. With the distribution of representations, each latent space becomes a metric measure space (mm-space) [27, 72], and the alignment map is an isomorphism.

Section 3 shows that AA is identifiable, the only isomorphism between the latent mm-spaces, if and only if no isometry of the data manifold other than the identity preserves the data distribution. This symmetry breaking (C1) depends on the data-generating process alone and holds generically.

Section 4 recovers AA by fitting a map to a coupling of the samples. Under C1, the coupling uniquely solves a non-convex transport problem [51]. Under distinguishable points (C2), a linear program finds the same optimum. We show that fused GW, which combines the two, keeps the unique optimum under C1 without requiring C2. Recovery is exact when both models encode the same observations and consistent for independent samples.

Section 5 instantiates the theory as Neural Gromov–Monge Alignment (NGMA), which computes the coupling with entropic optimal transport solvers [61, 12] and fits a map that preserves the pullback metric. Section 6 tests each claim on the sphere and on image and single-cell models. NGMA outperforms unpaired and label-informed baselines, and the pullback geometry is what makes it work across priors and training cells.

The metric manifold hypothesis defines the alignment and latent geometry

Representation learning is often motivated by the manifold hypothesis [7], which posits that the high-dimensional data concentrates near a lower-dimensional submanifold of the observation space. This idea has proven effective in, e.g., learning causal representations [69], biological mechanisms [47], text embeddings [54], and image generation [65].

Each of these successes uses the learned structure as a space. Clusters are found through distances, latent codes are interpolated along paths, and the representations of two models are compared by how far apart their points lie. All of this implicitly requires the data manifold to be connected, so that any two points are joined, and compact, so that distances stay bounded. It also requires the data to cover the manifold, since a model learns only the part of it that the data reach. Our metric manifold hypothesis (H1) states these assumptions explicitly.

Metric manifold hypothesis (H1). Observations x∼PXx \sim P_X lie in X=RD\mathcal{X} = \mathbb{R}^{D}, equipped with a Riemannian metric gXg^{\mathcal{X}}, Euclidean unless stated otherwise. We further assume:

(i) Data concentrate near a submanifold M⊂X\mathcal{M} \subset\mathcal{X} of dimension dim⁡M<D\dim\mathcal{M} < D that is smooth, compact and connected, possibly with a boundary [20, 58].

(ii) The noise-free data follow a distribution PMP_{\mathcal{M}} with a smooth positive density on M\mathcal{M}.

(iii) Each observation xx has a unique nearest point π(x)\pi(x) on M\mathcal{M}, and it is the noise-free point xx came from. This holds when noise moves points perpendicular to M\mathcal{M} by less than its reach, the largest distance within which every point has a unique nearest point on M\mathcal{M} [19, 1].

Distances on M\mathcal{M} are measured with the restriction gMg^{\mathcal{M}} of gXg^{\mathcal{X}}. The resulting shortest-path distance dMd_{\mathcal{M}} is finite because M\mathcal{M} is connected and bounded because it is compact. Compactness also makes the reach positive, so (iii) allows noise of positive size. Projecting the observations onto M\mathcal{M} recovers the noise-free data (Remark B.1), so PMP_{\mathcal{M}} is the distribution of π(x)\pi(x) for x∼PXx \sim P_X, the pushforward π#PX\pi_{\#}P_X. A metric space with a probability measure is a metric measure space, or mm-space [27, 72]. We call (M,dM,PM)(\mathcal{M}, d_{\mathcal{M}}, P_{\mathcal{M}}) the data mm-space.

What it means for a model to represent this structure is also usually left implicit. Using latent distances as distances between data points presupposes that distinct data points have distinct representations and that nearby representations belong to nearby data points. Identifiability theory assumes this explicitly [33, 35, 82, 74]. Our model coordinates hypothesis states it precisely:

Model coordinates (H2). A model provides coordinates on M\mathcal{M} through a parameterisation, a diffeomorphism f:Z→Mf : \mathcal{Z} \to\mathcal{M} from a latent space Z\mathcal{Z} of the same dimension onto M\mathcal{M}, that is, a smooth bijection with a smooth inverse. Then Z\mathcal{Z} inherits the compactness, connectedness and boundary of M\mathcal{M}. A parameterisation can be built from either an encoder or a decoder (Lemma B.2):

(i) A decoder dec⁡:Rq→X\operatorname{dec}: \mathbb{R}^{q} \to\mathcal{X}, injective with a Jacobian of rank qq and image containing M\mathcal{M}. Then f=dec⁡∣Zf = \operatorname{dec}|_{\mathcal{Z}} is a parameterisation on Z:=dec⁡−1(M)\mathcal{Z} := \operatorname{dec}^{-1}(\mathcal{M}) if Z\mathcal{Z} is compact.

(ii) An encoder enc⁡:X→Rq\operatorname{enc}: \mathcal{X} \to\mathbb{R}^{q} with enc⁡=enc⁡∣M∘π\operatorname{enc} = \operatorname{enc}|_{\mathcal{M}} \circ\pi, injective on M\mathcal{M} with a Jacobian of rank dim⁡M\dim\mathcal{M} there. Then f=(enc⁡∣M)−1f = (\operatorname{enc}|_{\mathcal{M}})^{-1} is a parameterisation on Z:=enc⁡(M)\mathcal{Z} := \operatorname{enc}(\mathcal{M}).

Since dim⁡M\dim\mathcal{M} is unknown, decoders usually have q>dim⁡Mq > \dim\mathcal{M}, and compactness of Z\mathcal{Z} keeps neighbouring data points from having distant representations. We write fa,fbf_a, f_b for the parameterisations of two models and Za,Zb\mathcal{Z}_a, \mathcal{Z}_b for their latent spaces. Two models of the same data are two coordinate systems on one manifold, and aligning them is changing coordinates.

Definition 2.1 (Alignment map). Under H1–H2, the alignment map A:Za→ZbA : \mathcal{Z}_a \to\mathcal{Z}_b is:

A=fb−1∘fa(1)A = f_b^{-1} \circ f_a \tag*{(1)}

AA sends each representation in model aa to the representation of the same data point in model bb.

Although AA has a closed form, it cannot be evaluated from unpaired samples and must be inferred from distances and distributions. Both depend on the geometry, and since AA only changes coordinates, it preserves a geometry only if both models measure the same distances on the data manifold M\mathcal{M}.

Proposition 2.2 (A shared data metric). Under H1–H2, let dad_a and dbd_b be metrics on Za\mathcal{Z}_a and Zb\mathcal{Z}_b. The alignment preserves them, db(A(z),A(z′))=da(z,z′)d_b(A(z), A(z')) = d_a(z, z') for all z,z′∈Zaz, z' \in\mathcal{Z}_a, if and only if both models

assign the same distance to every pair of data points,

da(fa−1(x),fa−1(y))=db(fb−1(x),fb−1(y))for all x,y∈M.d_a\left(f_a^{-1}(x), f_a^{-1}(y)\right)=d_b\left(f_b^{-1}(x), f_b^{-1}(y)\right)\qquad\text{for all }x,y\in\mathcal{M}.

Euclidean or spherical latent distances satisfy this condition only when AA extends to a rigid motion, which is what the methods of § 1 assume and what the data alone do not guarantee. Balestriero and LeCun [2026] prove that at the optimum, whitened JEPAs trained alike on the same data points share a Euclidean metric on them. This metric satisfies Proposition 2.2, but only models trained with the same loss on the same data points share it.

Latent metric measure spaces. By Proposition 2.2, the metric has to be one that both models induce on M\mathcal{M} regardless of their training, and M\mathcal{M} already carries a canonical one, dMd_{\mathcal{M}}, from H1. Each model reaches it through its own Jacobian, via the pullback metric ga(u,v):=gM(dfau,dfav)g_a(u,v):=g^{\mathcal{M}}(df_a u,df_a v), written fa∗gMf_a^*g^{\mathcal{M}}. The length dad_a of the shortest curve in Za\mathcal{Z}_a under gag_a then satisfies da(z,z′)=dM(fa(z),fa(z′))d_a(z,z')=d_{\mathcal{M}}(f_a(z),f_a(z')). We pair it with the distribution of the representations, Pa:=(fa−1)#PMP_a:=(f_a^{-1})_\#P_{\mathcal{M}}, and likewise for bb. Two mm-spaces are isomorphic if a bijection between them preserves distances and carries one measure onto the other. By construction, faf_a is such an isomorphism from the latent mm-space (Za,da,Pa)(\mathcal{Z}_a,d_a,P_a) onto the data mm-space, and so is fbf_b. Composing the two gives a key result:

Proposition 2.3. Under H1–H2, the alignment map is an isomorphism of the latent mm-spaces.

Unlike the Euclidean metric of whitened JEPAs, the pullback metric is shared by any two models satisfying H2, whatever their architecture, loss function or training points.

Identifiability of the alignment map

Proposition 2.3 tells us to search for an isomorphism of the latent mm-spaces, but not whether the search will find AA. Without pairs, the samples reveal only distances and distributions, which every such isomorphism, including AA, matches. If several exist, the search may return any of them. We call AA identifiable if it is the only isomorphism. Since both latent mm-spaces inherit their geometry from the data mm-space, identifiability is a property of the data-generating process rather than of the models, and we show that it holds generically.1

To illustrate when multiple isomorphisms exist, we build a running example of three data mm-spaces, each the sphere S2⊂R3\mathbb{S}^2\subset\mathbb{R}^3 with the round metric and a different distribution (Figure 2). The hypersphere is the geometry that contrastive losses [81] and alignment methods [68] impose on representations, even when the data manifold itself is not a sphere. The example therefore mirrors such models, but all results hold for any data mm-space satisfying the metric manifold hypothesis (H1).

Figure 2. Three data mm-spaces on the round sphere, with the symmetries SS that preserve each distribution (olive), the vMF centres (teal, area proportional to weight) and iso-density contours. O(3)O(3) consists of all isometries of the sphere, and O(2)O(2) of those that fix the vMF centre.

On each sphere, we take two parameterisations, fa=id⁡f_a=\operatorname{id} and fb=R−1f_b=R^{-1} for a fixed rotation RR, so model bb represents every data point rotated by RR and the alignment map is A=fb−1∘fa=RA=f_b^{-1}\circ f_a=R. Rotations preserve the round metric, so the latent mm-spaces (S2,dS2,P)(S^2,d_{S^2},P) and (S2,dS2,R#P)(S^2,d_{S^2},R_\#P) are isomorphic. For the uniform distribution, P=PunifP=P_{\mathrm{unif}} (Figure 2, left panel), every rotation RR is allowed, and the identity moves points but leaves the distances between them and their distribution unchanged. The map fb−1∘S∘fa=RSf_b^{-1}\circ S\circ f_a=RS is then an isomorphism of the latent mm-spaces just like AA, and the

search may return either. Concentrating the data around one point with a von Mises–Fisher (vMF) distribution PvMFP_{\mathrm{vMF}} [21] rules out some of these symmetries (middle), and the mixture PmixP_{\mathrm{mix}} of three vMF distributions rules out the rest (right), so AA is the only isomorphism (Appendix B.4).

A symmetry of the data mm-space is a bijection S:M→MS:\mathcal{M}\to\mathcal{M} that preserves distances and PMP_{\mathcal{M}}, and the symmetries form the automorphism group Aut⁡(M,dM,PM)\operatorname{Aut}(\mathcal{M},d_{\mathcal{M}},P_{\mathcal{M}}). Each symmetry SS yields an isomorphism fb−1∘S∘faf_b^{-1}\circ S\circ f_a that is inseparable from the alignment map without pairs, but differs from AA unless SS is the identity. The following condition ensures that AA is identifiable:

Symmetry breaking (C1). The identity is the only symmetry of the data mm-space.

It is also necessary, since every isomorphism of the latent mm-spaces comes from a symmetry:

Proposition 3.1. Under H1–H2, the isomorphisms of the latent mm-spaces are exactly the maps fb−1∘S∘faf_b^{-1}\circ S\circ f_a with S∈Aut⁡(M,dM,PM)S\in\operatorname{Aut}(\mathcal{M},d_{\mathcal{M}},P_{\mathcal{M}}), so AA is identifiable if and only if C1 holds.

The condition may seem demanding, since the uniform sphere already violates it. Yet symmetric densities are the exception on any data manifold:

Proposition 3.2. Fix M\mathcal{M} and gMg^{\mathcal{M}}. The densities in C>0∞(M)C^\infty_{>0}(\mathcal{M}) for which the data mm-space satisfies C1 form an open, dense subset in the C∞C^\infty topology.

A symmetry SS is fragile, because the density ρ\rho of PMP_{\mathcal{M}} must satisfy ρ∘S=ρ\rho\circ S=\rho at every point. An arbitrarily small bump at a point that SS moves, but not at its image, breaks SS. On the sphere, a first vMF bump on PunifP_{\mathrm{unif}} breaks every symmetry that moves its centre. The rotations about its axis and the reflections in planes containing it remain, as for PvMFP_{\mathrm{vMF}} (Figure 2, middle). A second, smaller bump off that axis breaks these rotations, and only the reflection in the plane through both centres remains. A third, even smaller bump off that plane breaks this reflection, which maps the third centre to a point without a bump (right). On any M\mathcal{M}, well-placed bumps lower the dimension of the continuous symmetries until none remain. Since the isometry group of M\mathcal{M} is compact, finitely many symmetries are left, and one more bump breaks each. Densities that satisfy C1 are therefore dense (Appendix B).

Conversely, densities that satisfy C1 form an open set, because a small change cannot create a symmetry. Every isometry of the sphere far from the identity changes the density of PmixP_{\mathrm{mix}} somewhere by at least a fixed amount. A small rotation changes it at a rate set by the slope of the density along the rotation, which is nonzero since no rotation preserves PmixP_{\mathrm{mix}}. A perturbation whose values and slope stay well below this amount and this rate, such as a slight shift of a centre, creates no symmetry. On any M\mathcal{M}, the argument uses only that the isometries form a compact Lie group, like O(3)O(3).

The pullback metric is what makes identifiability a property of the data-generating process alone. With Euclidean or spherical distances, AA preserves them only if AA is a rigid motion, which limits alignment to models trained with the same dot-product loss (§ 2). Even then, AA is unique only if the models spread their representations without symmetry. Contrastive losses and the LeJEPA regulariser do the opposite. They push the representations towards the uniform distribution on the hypersphere towards N(0,I)\mathcal{N}(0,I), which every orthogonal map preserves [81, 5]. Where these losses give two models a shared metric, they do so through this symmetry, which leaves AA inseparable from QAQA for every orthogonal QQ.

Recovery of the alignment from finite samples

Under symmetry breaking (C1), which holds generically, the latent mm-spaces determine AA, so estimating it without pairs is well-posed. We see only independent samples of the two latent mm-spaces, with no one-to-one correspondence between them, so regression is out of reach. Instead, we first estimate a coupling and then fit a map to it, akin to neural optimal transport methods [38]. We now discuss recovery guarantees for this two-stage procedure.

References

  1. [1]Eddie Aamari, Jisu Kim, Frédéric Chazal, Bertrand Michel, Alessandro Rinaldo, and Larry Wasserman. Estimating the reach of a manifold. Electronic Journal of Statistics, 13(1):1359 – 1399, 2019. doi: 10.1214/19-EJS1551. URL https://doi.org/10.1214/19-EJS1551.
  2. [2]David Alvarez-Melis and Tommi Jaakkola. Gromov-Wasserstein alignment of word embedding spaces. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, pages 1881–1890, 2018. doi: 10.18653/v1/d18-1214.
  3. [3]Mikel Artetxe, Gorka Labaka, and Eneko Agirre. A robust self-learning method for fully unsupervised cross-lingual mappings of word embeddings. In Proceedings of the 56th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 789–798, 2018. doi: 10.18653/v1/P18-1073.DOI
  4. [4]Georgios Arvanitidis, Lars Kai Hansen, and Søren Hauberg. Latent space oddity: on the curvature of deep generative models. In International Conference on Learning Representations, 2018. doi: 10.48550/arxiv.1710.11379.DOI
  5. [5]Randall Balestriero and Yann LeCun. LeJEPA: Provable and scalable self-supervised learning without the heuristics. arXiv preprint arXiv:2511.08544, 2025.arxiv.org/abs/2511.08544
  6. [7]Yoshua Bengio, Aaron Courville, and Pascal Vincent. Representation learning: A review and new perspectives. IEEE Transactions on Pattern Analysis and Machine Intelligence, 35:1798–1828, 2013. doi: 10.1109/tpami.2013.50.DOI
  7. [8]Patrick Billingsley. Convergence of Probability Measures. Wiley Series in Probability and Statistics. John Wiley & Sons, 2 edition, 1999. ISBN 9780471197454. doi: 10.1002/9780470316962.DOI
  8. [9]Isaac Chavel. Eigenvalues in Riemannian Geometry, volume 115 of Pure and Applied Mathematics. Academic Press, 1984. ISBN 9780121706401.
  9. [10]Alexis Conneau, Guillaume Lample, Marc’Aurelio Ranzato, Ludovic Denoyer, and Hervé Jégou. Word translation without parallel data. In International Conference on Learning Representations, 2018. doi: 10.48550/arxiv.1710.04087.DOI
  10. [11]Adrián Csiszárik, Péter Kőrösi-Szabó, Ákos K. Matszangosz, Gergely Papp, and Dániel Varga. Similarity and matching of neural network representations. In Advances in Neural Information Processing Systems, volume 34, pages 5656–5668, 2021. doi: 10.48550/arxiv.2110.14633.DOI
  11. [12]Marco Cuturi. Sinkhorn distances: Lightspeed computation of optimal transport. In Advances in Neural Information Processing Systems, volume 26, pages 2292–2300, 2013. doi: 10.48550/arxiv.1306.0895.DOI
  12. [13]Guy Dar. mini-vec2vec: Scaling universal geometry alignment with linear transformations. arXiv preprint arXiv:2510.02348, 2025. URL https://arxiv.org/abs/2510.02348.
  13. [14]Pınar Demetci, Rebecca Santorella, Björn Sandstede, William Stafford Noble, and Ritambhara Singh. SCOT: Single-cell multi-omics alignment with optimal transport. J. Comput. Biol., 29:3–18, 2022. doi: 10.1089/cmb.2021.0446.DOI
  14. [16]A. Dvoretzky, J. Kiefer, and J. Wolfowitz. Asymptotic Minimax Character of the Sample Distribution Function and of the Classical Multinomial Estimator. The Annals of Mathematical Statistics, 27:642–669, 1956. doi: 10.1214/aoms/1177728174. URL https://doi.org/10.1214/aoms/1177728174.
  15. [17]David G. Ebin. The manifold of Riemannian metrics. In Shiiing-Shen Chern and Stephen Smale, editors, Global Analysis, volume 15 of Proceedings of Symposia in Pure Mathematics, pages 11–40. American Mathematical Society, 1970. ISBN 9780821814154. doi: 10.1090/pspum/015/0267604.DOI
  16. [18]Lawrence C. Evans and Ronald F. Gariepy. Measure Theory and Fine Properties of Functions. Textbooks in Mathematics. CRC Press, revised edition, 2015. ISBN 9781482242386.
  17. [19]Herbert Federer. Curvature measures. Transactions of the American Mathematical Society, 93: 418–491, 1959. doi: 10.1090/s0002-9947-1959-0110078-1.DOI
  18. [20]Charles Fefferman, Sanjoy Mitter, and Hariharan Narayanan. Testing the manifold hypothesis. Journal of the American Mathematical Society, 29(4):983–1049, 2016. doi: 10.1090/jams/852. URL https://doi.org/10.1090/jams/852.
  19. [21]R. A. Fisher. Dispersion on a sphere. Proceedings of the Royal Society of London. Series A, Mathematical and Physical Sciences, 217:295–305, 1953. ISSN 1364-5021. doi: 10.1098/rspa.1953.0064. URL https://doi.org/10.1098/rspa.1953.0064.
  20. [22]Rémi Flamary, Nicolas Courty, Alexandre Gramfort, Mokhtar Z. Alaya, Aurélie Boisbuon, Stanislas Chambon, Laetitia Chapel, Adrien CornelJos, Kilian FATRAS, Nemo Fournier, Léo Gautherton, Nathalie T. H. Gayraud, Hicham Janati, Alain Rakotomamonjy, Ievgen Redko, Antoine Rolet, Antony Schutz, Vivien Seguy, Danica J. Sutherland, Romain Tavenard, Alexander Tong, and Titouan Vayer. POT: Python optimal transport. J. Mach. Learn. Res., 22(78):1–8, 2021. URL http://jmlr.org/papers/v22/20-451.html.
  21. [24]Marco Fumero, Marco Pegoraro, Valentino Maiorca, Francesco Locatello, and Emanuele Rodolà. Latent functional maps: a spectral framework for representation alignment. In Advances in Neural Information Processing Systems (NeurIPS), volume 37, pages 66178–66203, 2024. doi: 10.52202/079017-2115.DOI
  22. [25]Edouard Grave, Armand Joulin, and Quentin Berthet. Unsupervised alignment of embeddings with Wasserstein Procrustes. In Proceedings of the Twenty-Second International Conference on Artificial Intelligence and Statistics, volume 89 of Proceedings of Machine Learning Research, pages 1880–1890. PMLR, 2019.arxiv.org/abs/1805.11222
  23. [27]Misha Gromov. Metric Structures for Riemannian and Non-Riemannian Spaces, volume 152 of Progress in Mathematics. Birkhäuser, 1999. ISBN 9780817638986.
  24. [29]Søren Hauberg. Only Bayes should learn a manifold (on the estimation of differential geometric structure from data). arXiv preprint arXiv:1806.04994, 2018.arxiv.org/abs/1806.04994
  25. [30]Dan Hendrycks and Kevin Gimpel. Gaussian error linear units (GELUs). arXiv preprint arXiv:1606.08415, 2016. doi: 10.48550/arxiv.1606.08415.
  26. [31]Minyoung Huh, Brian Cheung, Tongzhou Wang, and Phillip Isola. Position: The Platonic representation hypothesis. In Proceedings of the 41st International Conference on Machine Learning, volume 235 of Proceedings of Machine Learning Research, pages 20617–20642. PMLR, 2024. URL https://proceedings.mlr.press/v235/huh24a.html.arxiv.org/abs/2405.07987
  27. [32]YoonHaeng Hur and Yuehaw Khoo. Robust point matching with distance profiles. Journal of Machine Learning Research, 26(205):1–38, 2025. URL http://jmlr.org/papers/v26/24-2224.html.
  28. [33]Aapo Hyvärinen and Petteri Pajunen. Nonlinear independent component analysis: Existence and uniqueness results. Neural Networks, 12(3):429–439, 1999. doi: 10.1016/S0893-6080(98)00140-3.DOI
  29. [34]Rishi Jha, Collin Zhang, Vitaly Shmatikov, and John X. Morris. Harnessing the universal geometry of embeddings. In Advances in Neural Information Processing Systems, volume 38, pages 51335–51359, 2025. doi: 10.52202/085713-1530. URL https://openreview.net/forum?id=jiCLUPq5xv.DOI
  30. [35]Ilyes Khemakhem, Diederik Kingma, Ricardo Monti, and Aapo Hyvärinen. Variational autoencoders and nonlinear ICA: A unifying framework. In Proceedings of the Twenty Third International Conference on Artificial Intelligence and Statistics, volume 108 of Proceedings of Machine Learning Research, pages 2207–2220. PMLR, 2020.arxiv.org/abs/1907.04809
  31. [36]Diederik P. Kingma and Jimmy Ba. Adam: A method for stochastic optimization. In International Conference on Learning Representations, 2015.arxiv.org/abs/1412.6980
  32. [37]Diederik P. Kingma and Max Welling. Auto-encoding variational Bayes. In International Conference on Learning Representations, 2014. doi: 10.48550/arxiv.1312.6114.DOI
  33. [38]Dominik Klein, Théo Uscidda, Fabian Theis, and Marco Cuturi. GENOT: Entropic (Gromov) Wasserstein flow matching with applications to single-cell genomics. In Advances in Neural Information Processing Systems 37, pages 103897–103944, 2024. doi: 10.52202/079017-3301.DOI
  34. [39]Shoshichi Kobayashi and Katsumi Nomizu. Foundations of Differential Geometry, Volume I. Wiley Classics Library. John Wiley & Sons, 1996. ISBN 9780471157335.
  35. [40]Steven G. Krantz and Harold R. Parks. A Primer of Real Analytic Functions. Birkhäuser Advanced Texts Basler Lehrbücher. Birkhäuser, 2 edition, 2002. ISBN 9780817642648. doi: 10.1007/978-0-8176-8134-0. URL https://doi.org/10.1007/978-0-8176-8134-0.
  36. [41]Natalia Kravtsova. Note on computational complexity of the Gromov-Wasserstein distance. arXiv preprint arXiv:2408.06525, 2026. doi: 10.48550/arXiv.2408.06525. v4. Earlier versions were titled “The NP-hardness of the Gromov-Wasserstein distance”; v4 proves non-convexity and gives no NP-hardness proof.DOI
  37. [42]Zorah Lähner and Michael Moeller. On the direct alignment of latent spaces. In Proceedings of UniReps: the First Workshop on Unifying Representations in Neural Models, volume 243 of Proceedings of Machine Learning Research, pages 158–169. PMLR, 2024.
  38. [43]Yann LeCun, Léon Bottou, Yoshua Bengio, and Patrick Haffner. Gradient-based learning applied to document recognition. Proceedings of the IEEE, 86(11):2278–2324, 1998. doi: 10.1109/5.726791.
  39. [45]John M. Lee. Introduction to Riemannian Manifolds, volume 176 of Graduate Texts in Mathematics. Springer, 2 edition, 2018. ISBN 9783319917542. doi: 10.1007/978-3-319-91755-9.DOI
  40. [46]Ziwei Liu, Ping Luo, Xiaogang Wang, and Xiaoou Tang. Deep learning face attributes in the wild. In Proceedings of the IEEE International Conference on Computer Vision (ICCV), pages 3730–3738, 2015. doi: 10.1109/ICCV.2015.425.DOI
  41. [47]Romain Lopez, Jeffrey Regier, Michael B. Cole, Michael I. Jordan, and Nir Yosef. Deep generative modeling for single-cell transcriptomics. Nature Methods, 15:1053–1058, 2018. doi: 10.1038/s41592-018-0229-2. URL https://doi.org/10.1038/s41592-018-0229-2.
  42. [48]Mohammad Lotfollahi, Mohsen Naghipourfar, Malte D. Luecken, Matin Khajavi, Maren Büttner, Marco Wagenstetter, Žiga Avsec, Adam Gayoso, Nir Yosef, Marta Interlandi, Sergei Rybakov, Alexander V. Misharin, and Fabian J. Theis. Mapping single-cell data to reference atlases by transfer learning. Nature Biotechnology, 40(1):121–130, 2022. doi: 10.1038/s41587-021-01001-7.DOI
  43. [49]Valentino Maiorca, Luca Moschella, Antonio Norelli, Marco Fumero, Francesco Locatello, and Emanuele Rodolà. Latent space translation via semantic alignment. In Advances in Neural Information Processing Systems, volume 36, pages 55394–55414, 2023. doi: 10.52202/075280-2418.DOI
  44. [50]P. Massart. The Tight Constant in the Dvoretzky-Kiefer-Wolfowitz Inequality. Ann. Probab., 18: 1269–1283, 1990. doi: 10.1214/aop/1176990746. URL https://doi.org/10.1214/aop/1176990746.
  45. [51]Facundo Mémoli. Gromov–Wasserstein distances and the metric approach to object matching. Found. Comput. Math., 11:417–487, 2011. doi: 10.1007/s10208-011-9093-5. URL https://doi.org/10.1007/s10208-011-9093-5.
  46. [52]Facundo Mémoli and Tom Needham. Distance distributions and inverse problems for metric measure spaces. Stud. Appl. Math., 149:943–1001, 2022. doi: 10.1111/sapm.12526.DOI
  47. [53]Jack Merullo, Louis Castricato, Carsten Eickhoff, and Ellie Pavlick. Linearly mapping from image to text space. In International Conference on Learning Representations, 2023.arxiv.org/abs/2209.15162
  48. [54]Tomas Mikolov, Kai Chen, Greg Corrado, and Jeffrey Dean. Efficient estimation of word representations in vector space. In International Conference on Learning Representations (ICLR), Workshop Track, 2013. URL https://arxiv.org/abs/1301.3781.
  49. [55]B. S. Mityagin. The zero set of a real analytic function. Mathematical Notes, 107:529–530, 2020. doi: 10.1134/S0001434620030189. URL https://doi.org/10.1134/S0001434620030189.
  50. [56]Luca Moschella, Valentino Maiorca, Marco Fumero, Antonio Norelli, Francesco Locatello, and Emanuele Rodolà. Relative representations enable zero-shot latent space communication. International Conference on Learning Representations, 2023. doi: 10.48550/arxiv.2209.15430. URL https://openreview.net/forum?id=SrC-nwieGJ.DOI
  51. [57]S. B. Myers and N. E. Steenrod. The group of isometries of a Riemannian manifold. Annals of Mathematics, 40:400–416, 1939. doi: 10.2307/1968928.DOI
  52. [58]Partha Niyogi, Stephen Smale, and Shmuel Weinberger. Finding the homology of submanifolds with high confidence from random samples. Discrete & Computational Geometry, 39(1-3):419–441, 2008. ISSN 1432-0444. doi: 10.1007/s00454-008-9053-2. URL https://doi.org/10.1007/s00454-008-9053-2.
  53. [59]Richard S. Palais. On the differentiability of isometries. Proc. Amer. Math. Soc., 8:805–807, 1957. doi: 10.1090/s0002-9939-1957-0088000-x.DOI
  54. [60]Peter Petersen. Riemannian Geometry, volume 171 of Graduate Texts in Mathematics. Springer International Publishing, 3rd edition, 2016. ISBN 9783319266527. doi: 10.1007/978-3-319-26654-1.
  55. [61]Gabriel Peyré, Marco Cuturi, and Justin Solomon. Gromov–Wasserstein averaging of kernel and distance matrices. In International Conference on Machine Learning, volume 48 of Proceedings of Machine Learning Research, pages 2664–2672. PMLR, 2016.
  56. [62]Moritz Piening and Robert Beinert. A novel sliced fused Gromov-Wasserstein distance. Proceedings of the AAAI Conference on Artificial Intelligence, 40(29):24828–24836, 2026. doi: 10.1609/aaai.v40i29.39669.DOI
  57. [63]Friedrich Pukelsheim. Optimal Design of Experiments, volume 50 of Classics in Applied Mathematics. Society for Industrial and Applied Mathematics, 2006. ISBN 9780898716047. doi: 10.1137/1.9780898719109. URL https://epubs.siam.org/doi/abs/10.1137/1.9780898719109.DOI
  58. [64]Dan Raviv, Anastasia Dubrovina, and Ron Kimmel. Hierarchical matching of non-rigid shapes. In Scale Space and Variational Methods in Computer Vision (SSVM 2011), volume 6667 of Lecture Notes in Computer Science, pages 604–615. Springer, 2012. doi: 10.1007/978-3-642-24785-9_51.
  59. [65]Robin Rombach, Andreas Blattmann, Dominik Lorenz, Patrick Esser, and Björn Ommer. High-resolution image synthesis with latent diffusion models. In 2022 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 10674–10685. IEEE, 2022. doi: 10.1109/cvpr52688.2022.01042.DOI
  60. [66]Ryoma Sato, Marco Cuturi, Makoto Yamada, and Hisashi Kashima. Fast and robust comparison of probability measures in heterogeneous spaces. arXiv preprint arXiv:2002.01615, 2020.arxiv.org/abs/2002.01615
  61. [67]Dominik Schnaus, Nikita Araslanov, and Daniel Cremers. It’s a (blind) match! Towards vision-language correspondence without parallel data. In 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR), pages 24983–24992, 2025. doi: 10.1109/cvpr52734.2025.02326.DOI
  62. [68]Dominik Schnaus, Thomas Dagès, Daniel Cremers, Xi Wang, and Phillip Isola. Shared geometry as a Rosetta Stone: Cross-modal alignment without paired data. arXiv preprint arXiv:2610.09411, 2026. URL https://arxiv.org/abs/2610.09411.
  63. [69]Bernhard Schölkopf, Francesco Locatello, Stefan Bauer, Nan Rosemary Ke, Nal Kalchbrenner, Anirudh Goyal, and Yoshua Bengio. Toward causal representation learning. Proceedings of the IEEE, 109(5):612–634, 2021. doi: 10.1109/JPROC.2021.3058954. URL https://doi.org/10.1109/JPROC.2021.3058954.
  64. [70]Peter H. Schönemann. A generalized solution of the orthogonal Procrustes problem. Psychometrika, 31:1–10, 1966. doi: 10.1007/bf02289451.DOI
  65. [71]Athina Sotiropoulou and David Alvarez-Melis. Strongly isomorphic neural optimal transport across incomparable spaces. In Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM), volume 251 of Proceedings of Machine Learning Research, pages 381–393. PMLR, 2024. doi: 10.48550/arxiv.2407.14957.DOI
  66. [72]Karl-Theodor Sturm. On the geometry of metric measure spaces. Acta Mathematica, 196:65–131, 2006. doi: 10.1007/s11511-006-0002-8.DOI
  67. [73]Stas Syrota, Pablo Moreno-Muñoz, and Søren Hauberg. Decoder ensembling for learned latent geometries. In Proceedings of the Geometry-grounded Representation Learning and Generative Modeling Workshop (GRaM), volume 251 of Proceedings of Machine Learning Research, pages 277–285. PMLR, 2024. doi: 10.48550/arxiv.2408.07507.DOI
  68. [74]Stas Syrota, Yevgen Zainchkovskyy, Johnny Xi, Benjamin Bloem-Reddy, and Søren Hauberg. Identifying metric structures of deep latent variable models. In International Conference on Machine Learning, volume 267 of Proceedings of Machine Learning Research, pages 58065–58087. PMLR, 2025. doi: 10.48550/arxiv.2502.13757.DOI
  69. [75]The Tabula Sapiens Consortium. The Tabula Sapiens: A multiple-organ, single-cell transcriptomic atlas of humans. Science, 376(6594):eabl4896, 2022. doi: 10.1126/science.abl4896.
  70. [76]Jakub M. Tomczak and Max Welling. VAE with a VampPrior. In Amos Storkey and Fernando Perez-Cruz, editors, International Conference on Artificial Intelligence and Statistics, volume 84 of Proceedings of Machine Learning Research, pages 1214–1223. PMLR, 2018. doi: 10.48550/arxiv.1705.07120. URL https://proceedings.mlr.press/v84/tomczak18a.html.DOI
  71. [77]Théo Uscidda, Luca Eyring, Karsten Roth, Fabian Theis, Zeynep Akata, and Marco Cuturi. Disentangled representation learning with the Gromov-Monge gap. In International Conference on Learning Representations, 2025. doi: 10.48550/arxiv.2407.07829.DOI
  72. [78]Titouan Vayer, Laetitia Chapel, Rémi Flamary, Romain Tavenard, and Nicolas Courty. Optimal transport for structured data with application on graphs. In Proceedings of the 36th International Conference on Machine Learning, volume 97 of PMLR, pages 6275–6284, 2019.arxiv.org/abs/1805.09114
  73. [79]Cédric Villani. Optimal Transport: Old and New, volume 338 of Grundlehren der mathematischen Wissenschaften. Springer, 2009. ISBN 9783540710509. doi: 10.1007/978-3-540-71050-9.
  74. [80]Tao Wang and Ziv Goldfeld. Neural entropic optimal transport and Gromov-Wasserstein alignment. IEEE Transactions on Information Theory, 72:2424–2443, 2026. doi: 10.1109/tit.2026.3661439.
  75. [81]Tongzhou Wang and Phillip Isola. Understanding contrastive representation learning through alignment and uniformity on the hypersphere. In Proceedings of the 37th International Conference on Machine Learning, volume 119 of Proceedings of Machine Learning Research, pages 9929–9939. PMLR, 2020.arxiv.org/abs/2005.10242
  76. [82]Quanhан Xi and Benjamin Bloem-Reddy. Indeterminacy in generative models: Characterization and strong identifiability. In International Conference on Artificial Intelligence and Statistics, volume 206 of Proceedings of Machine Learning Research, pages 6912–6939. PMLR, 2023. doi: 10.48550/arXiv.2206.00801.DOI
  77. [83]Han Xiao, Kashif Rasul, and Roland Vollgraf. Fashion-MNIST: a novel image dataset for benchmarking machine learning algorithms. arXiv preprint arXiv:1708.07747, 2017.arxiv.org/abs/1708.07747
  78. [84]Hanlin Yu, Berfin Inal, Georgios Arvanitidis, Søren Hauberg, Francesco Locatello, and Marco Fumero. Connecting neural models latent geometries with relative geodesic representations. In Advances in Neural Information Processing Systems, volume 38, pages 125316–125360, 2025. doi: 10.52202/085713-3769.DOI
  79. [85]Hongyi Zhang, Yann N. Dauphin, and Tengyu Ma. Fixup initialization: Residual learning without normalization. In International Conference on Learning Representations, 2019. doi: 10.48550/arxiv.1901.09321.DOI
  80. [86]Jun-Yan Zhu, Taesung Park, Phillip Isola, and Alexei A. Efros. Unpaired image-to-image translation using cycle-consistent adversarial networks. In 2017 IEEE International Conference on Computer Vision (ICCV), pages 2242–2251, 2017. ISBN 9781538610329. doi: 10.1109/iccv.2017.244.DOI

Paper details

vlnc:W_hjnqy75k
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 0

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents