-
Stationary Kernels and Gaussian Processes on Lie Groups and their Homogeneous Spaces II: non-compact symmetric spaces
Authors:
Iskander Azangulov,
Andrei Smolensky,
Alexander Terenin,
Viacheslav Borovitskiy
Abstract:
Gaussian processes are arguably the most important class of spatiotemporal models within machine learning. They encode prior information about the modeled function and can be used for exact or approximate Bayesian learning. In many applications, particularly in physical sciences and engineering, but also in areas such as geostatistics and neuroscience, invariance to symmetries is one of the most f…
▽ More
Gaussian processes are arguably the most important class of spatiotemporal models within machine learning. They encode prior information about the modeled function and can be used for exact or approximate Bayesian learning. In many applications, particularly in physical sciences and engineering, but also in areas such as geostatistics and neuroscience, invariance to symmetries is one of the most fundamental forms of prior information one can consider. The invariance of a Gaussian process' covariance to such symmetries gives rise to the most natural generalization of the concept of stationarity to such spaces. In this work, we develop constructive and practical techniques for building stationary Gaussian processes on a very large class of non-Euclidean spaces arising in the context of symmetries. Our techniques make it possible to (i) calculate covariance kernels and (ii) sample from prior and posterior Gaussian processes defined on such spaces, both in a practical manner. This work is split into two parts, each involving different technical considerations: part I studies compact spaces, while part II studies non-compact spaces possessing certain structure. Our contributions make the non-Euclidean Gaussian process models we study compatible with well-understood computational techniques available in standard Gaussian process software packages, thereby making them accessible to practitioners.
△ Less
Submitted 1 July, 2024; v1 submitted 30 January, 2023;
originally announced January 2023.
-
On power sum kernels on symmetric groups
Authors:
Iskander Azangulov,
Viacheslav Borovitskiy,
Andrei Smolensky
Abstract:
In this note, we introduce a family of "power sum" kernels and the corresponding Gaussian processes on symmetric groups $\mathrm{S}_n$. Such processes are bi-invariant: the action of $\mathrm{S}_n$ on itself from both sides does not change their finite-dimensional distributions. We show that the values of power sum kernels can be efficiently calculated, and we also propose a method enabling approx…
▽ More
In this note, we introduce a family of "power sum" kernels and the corresponding Gaussian processes on symmetric groups $\mathrm{S}_n$. Such processes are bi-invariant: the action of $\mathrm{S}_n$ on itself from both sides does not change their finite-dimensional distributions. We show that the values of power sum kernels can be efficiently calculated, and we also propose a method enabling approximate sampling of the corresponding Gaussian processes with polynomial computational complexity. By doing this we provide the tools that are required to use the introduced family of kernels and the respective processes for statistical modeling and machine learning.
△ Less
Submitted 28 November, 2022; v1 submitted 10 November, 2022;
originally announced November 2022.
-
Stationary Kernels and Gaussian Processes on Lie Groups and their Homogeneous Spaces I: the compact case
Authors:
Iskander Azangulov,
Andrei Smolensky,
Alexander Terenin,
Viacheslav Borovitskiy
Abstract:
Gaussian processes are arguably the most important class of spatiotemporal models within machine learning. They encode prior information about the modeled function and can be used for exact or approximate Bayesian learning. In many applications, particularly in physical sciences and engineering, but also in areas such as geostatistics and neuroscience, invariance to symmetries is one of the most f…
▽ More
Gaussian processes are arguably the most important class of spatiotemporal models within machine learning. They encode prior information about the modeled function and can be used for exact or approximate Bayesian learning. In many applications, particularly in physical sciences and engineering, but also in areas such as geostatistics and neuroscience, invariance to symmetries is one of the most fundamental forms of prior information one can consider. The invariance of a Gaussian process' covariance to such symmetries gives rise to the most natural generalization of the concept of stationarity to such spaces. In this work, we develop constructive and practical techniques for building stationary Gaussian processes on a very large class of non-Euclidean spaces arising in the context of symmetries. Our techniques make it possible to (i) calculate covariance kernels and (ii) sample from prior and posterior Gaussian processes defined on such spaces, both in a practical manner. This work is split into two parts, each involving different technical considerations: part I studies compact spaces, while part II studies non-compact spaces possessing certain structure. Our contributions make the non-Euclidean Gaussian process models we study compatible with well-understood computational techniques available in standard Gaussian process software packages, thereby making them accessible to practitioners.
△ Less
Submitted 7 November, 2023; v1 submitted 31 August, 2022;
originally announced August 2022.
-
Geometry-aware Bayesian Optimization in Robotics using Riemannian Matérn Kernels
Authors:
Noémie Jaquier,
Viacheslav Borovitskiy,
Andrei Smolensky,
Alexander Terenin,
Tamim Asfour,
Leonel Rozo
Abstract:
Bayesian optimization is a data-efficient technique which can be used for control parameter tuning, parametric policy adaptation, and structure design in robotics. Many of these problems require optimization of functions defined on non-Euclidean domains like spheres, rotation groups, or spaces of positive-definite matrices. To do so, one must place a Gaussian process prior, or equivalently define…
▽ More
Bayesian optimization is a data-efficient technique which can be used for control parameter tuning, parametric policy adaptation, and structure design in robotics. Many of these problems require optimization of functions defined on non-Euclidean domains like spheres, rotation groups, or spaces of positive-definite matrices. To do so, one must place a Gaussian process prior, or equivalently define a kernel, on the space of interest. Effective kernels typically reflect the geometry of the spaces they are defined on, but designing them is generally non-trivial. Recent work on the Riemannian Matérn kernels, based on stochastic partial differential equations and spectral theory of the Laplace-Beltrami operator, offers promising avenues towards constructing such geometry-aware kernels. In this paper, we study techniques for implementing these kernels on manifolds of interest in robotics, demonstrate their performance on a set of artificial benchmark functions, and illustrate geometry-aware Bayesian optimization for a variety of robotic applications, covering orientation control, manipulability optimization, and motion planning, while showing its improved performance.
△ Less
Submitted 17 March, 2023; v1 submitted 2 November, 2021;
originally announced November 2021.
-
The centralizers of root subgroups in Kac-Moody Steinberg groups
Authors:
Andrei Smolensky
Abstract:
For the affine and hyperbolic root system the symmetric part of the centralizers of root subgroups in the corresponding Steinberg groups are calculated. In the affine case the corresponding root subsystems can be computed in term of the centralizers in the spherical root systems, while in the hyperbolic case there emerges a "zoo" of examples, many of them non-hyperbolic. This also delivers many ex…
▽ More
For the affine and hyperbolic root system the symmetric part of the centralizers of root subgroups in the corresponding Steinberg groups are calculated. In the affine case the corresponding root subsystems can be computed in term of the centralizers in the spherical root systems, while in the hyperbolic case there emerges a "zoo" of examples, many of them non-hyperbolic. This also delivers many examples of naturally occuring root subsystems of infinite rank.
△ Less
Submitted 15 August, 2021;
originally announced August 2021.
-
Real roots in the root system $\mathsf{T}_{2,p,q}$
Authors:
Karin Baur,
Jian-Rong Li,
Andrei Smolensky
Abstract:
Motivated by the recent advances in the categorification of the cluster structure on the coordinate rings of Grassmannians of $k$-subspaces in $n$-space, we investigate a particular construction of root systems of type $\mathsf{T}_{2,p,q}$, including the type $\mathsf{E}_n$. This construction generalizes Manin's ``hyperbolic construction'' of $\mathsf{E}_8$ and reveals a lot of otherwise hidden re…
▽ More
Motivated by the recent advances in the categorification of the cluster structure on the coordinate rings of Grassmannians of $k$-subspaces in $n$-space, we investigate a particular construction of root systems of type $\mathsf{T}_{2,p,q}$, including the type $\mathsf{E}_n$. This construction generalizes Manin's ``hyperbolic construction'' of $\mathsf{E}_8$ and reveals a lot of otherwise hidden regularities in this family of root systems.
△ Less
Submitted 17 August, 2023; v1 submitted 8 January, 2021;
originally announced January 2021.
-
Suzuki-Ree groups and Tits mixed groups over rings
Authors:
Andrei Smolensky
Abstract:
It is shown that Suzuki-Ree groups can be easily defined by means of comparing two fundamental representations of the ambient Chevalley group in characteristic 2 or 3. This eliminates the distinction between the Suzuki-Ree groups over perfect and imperfect fields and gives a natural definition for the analogues of such groups over commutative rings. As an application of the same idea, we explicitl…
▽ More
It is shown that Suzuki-Ree groups can be easily defined by means of comparing two fundamental representations of the ambient Chevalley group in characteristic 2 or 3. This eliminates the distinction between the Suzuki-Ree groups over perfect and imperfect fields and gives a natural definition for the analogues of such groups over commutative rings. As an application of the same idea, we explicitly construct a pair of polynomial maps between the groups of types B_n and C_n in characteristic 2 that compose to the Frobenius endomorphism. This, in turn, provides a simple definition for the Tits mixed groups over rings.
△ Less
Submitted 17 August, 2018;
originally announced August 2018.
-
On the combinatorics of circular codes
Authors:
Aleksandr Serdiukov,
Andrei Smolensky
Abstract:
The present paper is devoted to the study of the combinatorics of 216 maximal $C^3$ circular codes --- a particular type of structure arising in the analysis of genomic sequences. Their circularity property is believed to be intimately connected to the protection against the reading frame shift in the process of RNA translation. We present some new observations concerning the internal structure of…
▽ More
The present paper is devoted to the study of the combinatorics of 216 maximal $C^3$ circular codes --- a particular type of structure arising in the analysis of genomic sequences. Their circularity property is believed to be intimately connected to the protection against the reading frame shift in the process of RNA translation. We present some new observations concerning the internal structure of circular codes, which give a way to construct all of them in a relatively simple manner.
△ Less
Submitted 13 May, 2018;
originally announced May 2018.
-
On the definition of Suzuki groups over rings
Authors:
Andrei Smolensky
Abstract:
The definition of Suzuki groups over rings is given by means of an explicit description as a difference-algebraic group. For a (not necessarily perfect) field with more than two elements this construction produces a simple group.
The definition of Suzuki groups over rings is given by means of an explicit description as a difference-algebraic group. For a (not necessarily perfect) field with more than two elements this construction produces a simple group.
△ Less
Submitted 8 January, 2018;
originally announced January 2018.
-
Decompositions of congruence subgroups of Chevalley groups
Authors:
Sergey Sinchuk,
Andrei Smolensky
Abstract:
We formulate and prove relative versions of several classical decompositions known in the theory of Chevalley groups over commutative rings. As an application we obtain upper estimates for the width of principal congruence subgroups in terms of several families of generators. Some of our results are new even in the absolute case and were previously studied only for groups over finite fields.
We formulate and prove relative versions of several classical decompositions known in the theory of Chevalley groups over commutative rings. As an application we obtain upper estimates for the width of principal congruence subgroups in terms of several families of generators. Some of our results are new even in the absolute case and were previously studied only for groups over finite fields.
△ Less
Submitted 29 September, 2018; v1 submitted 9 November, 2015;
originally announced November 2015.
-
Products of Sylow subgroups in Suzuki and Ree groups
Authors:
Andrei Smolensky
Abstract:
An explicit and elementary proof is given to the fact that Suzuki and Ree groups can be decomposed into the product of 4 of their Sylow p-subgroups, where p is the defining characterictic.
An explicit and elementary proof is given to the fact that Suzuki and Ree groups can be decomposed into the product of 4 of their Sylow p-subgroups, where p is the defining characterictic.
△ Less
Submitted 18 February, 2015; v1 submitted 21 January, 2015;
originally announced January 2015.
-
Commutator width of Chevalley groups over rings of stable rank 1
Authors:
Andrei Smolensky
Abstract:
An estimate on the commutator width is given for Chevalley groups over rings of stable rank 1, and the general method suitable for other rings of small dimension.
An estimate on the commutator width is given for Chevalley groups over rings of stable rank 1, and the general method suitable for other rings of small dimension.
△ Less
Submitted 13 October, 2014;
originally announced October 2014.
-
Gauss decomposition for Chevalley groups, revisited
Authors:
A. Smolensky,
B. Sury,
N. Vavilov
Abstract:
In the 1960's Noboru Iwahori and Hideya Matsumoto, Eiichi Abe and Kazuo Suzuki, and Michael Stein discovered that Chevalley groups $G=G(Φ,R)$ over a semilocal ring admit remarkable Gauss decomposition $G=TUU^-U$, where $T=T(Φ,R)$ is a split maximal torus, whereas $U=U(Φ,R)$ and $U^-=U^-(Φ,R)$ are unipotent radicals of two opposite Borel subgroups $B=B(Φ,R)$ and $B^-=B^-(Φ,R)$ containing $T$. It fo…
▽ More
In the 1960's Noboru Iwahori and Hideya Matsumoto, Eiichi Abe and Kazuo Suzuki, and Michael Stein discovered that Chevalley groups $G=G(Φ,R)$ over a semilocal ring admit remarkable Gauss decomposition $G=TUU^-U$, where $T=T(Φ,R)$ is a split maximal torus, whereas $U=U(Φ,R)$ and $U^-=U^-(Φ,R)$ are unipotent radicals of two opposite Borel subgroups $B=B(Φ,R)$ and $B^-=B^-(Φ,R)$ containing $T$. It follows from the classical work of Hyman Bass and Michael Stein that for classical groups Gauss decomposition holds under weaker assumptions such as $\sr(R)=1$ or $\asr(R)=1$. Later the second author noticed that condition $\sr(R)=1$ is necessary for Gauss decomposition. Here, we show that a slight variation of Tavgen's rank reduction theorem implies that for the elementary group $E(Φ,R)$ condition $\sr(R)=1$ is also sufficient for Gauss decomposition. In other words, $E=HUU^-U$, where $H=H(Φ,R)=T\cap E$. This surprising result shows that stronger conditions on the ground ring, such as being semi-local, $\asr(R)=1$, $\sr(R,Λ)=1$, etc., were only needed to guarantee that for simply connected groups $G=E$, rather than to verify the Gauss decomposition itself.
△ Less
Submitted 10 October, 2011; v1 submitted 24 September, 2011;
originally announced September 2011.
-
Unitriangular factorisations of Chevalley groups
Authors:
N. A. Vavilov,
A. V. Smolensky,
B. Sury
Abstract:
Lately, the following problem has attracted a lot of attention in various contexts: find the shortest factorisation $G=UU^-UU^-...U^{\pm}$ of a Chevalley group $G=G(Φ,R)$ in terms of the unipotent radical $U=U(Φ,R)$ of the standard Borel subgroup $B=B(Φ,R)$ and the unipotent radical $U^-=U^-(Φ,R)$ of the opposite Borel subgroup $B^-=B^-(Φ,R)$. So far, the record over a finite field was established…
▽ More
Lately, the following problem has attracted a lot of attention in various contexts: find the shortest factorisation $G=UU^-UU^-...U^{\pm}$ of a Chevalley group $G=G(Φ,R)$ in terms of the unipotent radical $U=U(Φ,R)$ of the standard Borel subgroup $B=B(Φ,R)$ and the unipotent radical $U^-=U^-(Φ,R)$ of the opposite Borel subgroup $B^-=B^-(Φ,R)$. So far, the record over a finite field was established in a 2010 paper by Babai, Nikolov, and Pyber, where they prove that a group of Lie type admits unitriangular factorisation $G=UU^-UU^-U$ of length 5. Their proof invokes deep analytic and combinatorial tools. In the present paper we notice that from the work of Bass and Tavgen one immediately gets a much more general result, asserting that over any ring of stable rank 1 one has unitriangular factorisation $G=UU^-UU^-$ of length 4. Moreover, we give a detailed survey of triangular factorisations, prove some related results, discuss prospects of generalisation to other classes of rings, and state several unsolved problems. Another main result of the present paper asserts that, in the assumption of the Generalised Riemann's Hypothesis, Chevalley groups over the ring $\Int\Big[\displaystyle{1\over p}\Big]$ admit unitriangular factorisation $G=UU^-UU^-UU^-$ of length 6. Otherwise, the best length estimate for Hasse domains with infinite multiplicative groups that follows from the work of Cooke and Weinberger, gives 9 factors.
△ Less
Submitted 27 July, 2011;
originally announced July 2011.
