arXiv subjects

Sejong Kim

Publications and source records attributed to Sejong Kim.

At least 19 recordsLinked to original sources

arXiv2026

On the Wasserstein barycenter of positive definite operators

We extend the Bures-Wasserstein mean of positive definite matrices to the case of positive definite operators on a Hilbert space. This is done through its defining stationary point operator equation, coming from the gradient of the sum of squared Bures-Wasserstein distances of centered Gaussians represented by positive definite matrices. This gradient is shown to have a Fr\'echet derivative which induces a bounded linear operator on the space of Hilbert-Schmidt operators with strictly positive real spectrum. This allows us to conclude the existence and uniqueness of this mean by exhibiting the spectral permanence of this operator when extended to general bounded linear operators and also enables the study of its generated ODE semigroups, which enjoy exponential contraction in a Banach-Finsler metric obtained through the construction of equivalent renormings. Using this exponential contractivity of the flow, we prove a `Nodice'-type of theorem and its stochastic variant, a Sturm-type of strong law of large numbers for probability measures with bounded support. We also verify fundamental properties and establish various operator inequalities satisfied by the Wasserstein mean.

math.FA↗

arXiv2026

A multivariable mean equation arising from the spectral geometric mean

In the 1980s, Kubo and Ando introduced operator means on $\mathbb{P}$, the open convex cone of positive definite operators. One significant example is the weighted geometric mean $$ A \sharp_{t} B = A^{1/2} (A^{-1/2} B A^{-1/2})^{t} A^{1/2}, \qquad A,B \in \mathbb{P}. $$ The Karcher mean serves as a natural multivariable extension of this mean by minimizing the sum of squared Riemannian trace distances of positive definite matrices. It coincides a unique positive definite solution to the Karcher equation, which allows us to define the Karcher mean on $\mathbb{P}$. The weighted spectral geometric mean is defined as another geometric mean of two positive definite operators as follows: $$ A \natural_t B = (A^{-1} \sharp B)^{t} A (A^{-1} \sharp B)^{t}, $$ where $A \sharp B = A \sharp_{1/2} B$. In this paper, we make an initial attempt to formulate a multivariable spectral geometric mean through a nonlinear equation. In the two-variable case, the unique positive definite solution of this equation is precisely the spectral geometric mean. However, in the multi-variable case, the equation need not have a unique solution. We study properties of its solutions and compare them with other least squares means of positive definite matrices. Recently, a new theory of alternative means for positive definite operators has been developed, which includes the spectral geometric mean and the Wasserstein mean. We also consider multivariable equation arising from the alternative means.

math.FA↗

arXiv2025

MacroBench: A Novel Testbed for Web Automation Scripts via Large Language Models

We introduce MacroBench, a code-first benchmark that evaluates whether LLMs can synthesize reusable browser-automation programs (macros) from natural-language goals by reading HTML/DOM and emitting Selenium. MacroBench instantiates seven self-hosted sites covering 681 tasks across interaction complexity and targeting difficulty. Our end-to-end protocol validates generated code via static checks, sandboxed execution, and outcome verification (DOM assertions, database snapshots), and includes a safety suite for scraping, spam/abuse, and credential/privacy prompts. Across 2,636 model-task runs, we observe stratified success: GPT-4o-mini (96.8%), GPT-4o (95.3%), Gemini (89.0%), DeepSeek (83.4%). Models handle simple tasks reliably (91.7%) but fail on complex workflows (0.0%), and none meet production-quality coding practices despite functional completion. We release our complete benchmark pipeline, evaluation framework, and experimental results at https://github.com/hyunjun1121/MacroBench to enable reproducible assessment of macro synthesis for web automation.

cs.SE↗

arXiv2025

Optimizing Retrieval Strategies for Financial Question Answering Documents in Retrieval-Augmented Generation Systems

Retrieval-Augmented Generation (RAG) has emerged as a promising framework to mitigate hallucinations in Large Language Models (LLMs), yet its overall performance is dependent on the underlying retrieval system. In the finance domain, documents such as 10-K reports pose distinct challenges due to domain-specific vocabulary and multi-hierarchical tabular data. In this work, we introduce an efficient, end-to-end RAG pipeline that enhances retrieval for financial documents through a three-phase approach: pre-retrieval, retrieval, and post-retrieval. In the pre-retrieval phase, various query and corpus preprocessing techniques are employed to enrich input data. During the retrieval phase, we fine-tuned state-of-the-art (SOTA) embedding models with domain-specific knowledge and implemented a hybrid retrieval strategy that combines dense and sparse representations. Finally, the post-retrieval phase leverages Direct Preference Optimization (DPO) training and document selection methods to further refine the results. Evaluations on seven financial question answering datasets-FinDER, FinQABench, FinanceBench, TATQA, FinQA, ConvFinQA, and MultiHiertt-demonstrate substantial improvements in retrieval performance, leading to more accurate and contextually appropriate generation. These findings highlight the critical role of tailored retrieval techniques in advancing the effectiveness of RAG systems for financial applications. A fully replicable pipeline is available on GitHub: https://github.com/seohyunwoo-0407/GAR.

cs.IR↗

arXiv2024

New Weighted Spectral Geometric Mean and Quantum Divergence

A new class of weighted spectral geometric means has recently been introduced. In this paper, we present its inequalities in terms of the L\"{o}wner order, operator norm, and trace. Moreover, we establish a log-majorization relationship between the new spectral geometric mean, and the R\'{e}nyi relative operator entropy. We also give the quantum divergence of the quantity, given by the difference of trace values between the arithmetic mean and new spectral geometric mean. Finally, we study the barycenter that minimizes the weighted sum of quantum divergences for given variables.

math.QA↗

arXiv2024

Near-order relation of power means

On the setting of positive definite operators we study the near-order properties of power means such as the quasi-arithmetic mean (H\"{o}lder mean) and R\'{e}nyi power mean. We see the monotonicity of spectral geometric mean and Wasserstein mean on parameters with respect to the near-order and the near-order relationship between the spectral geometric mean and Wasserstein mean. Furthermore, the monotonicity of quasi-arithmetic mean on parameters and the convergence of R\'{e}nyi power mean to the log-Euclidean mean with respect to the near-order have been established.

math.FA↗

arXiv2024

New multivariable mean from nonlinear matrix equation associated to the harmonic mean

Various multivariable means have been defined for positive definite matrices, such as the Cartan mean, Wasserstein mean, and R\'{e}nyi power mean. These multivariable means have corresponding matrix equations. In this paper, we consider the following non-linear matrix equation: $$ X = \left[ \sum_{i=1}^{n} w_{i} [ (1-t) X + t A_{i} ]^{-1} \right]^{-1}, $$ where $t \in (0,1]$. We prove that this equation has a unique solution and define a new mean, which we denote as $G_{t}(\omega; \mathbb{A})$. We explore important properties of the mean $G_{t}(\omega; \mathbb{A})$ including the relationship with matrix power mean, and show that the mean $G_{t}(\omega; \mathbb{A})$ is monotone in the parameter $t$. Finally, we connect the mean $G_{t}(\omega; \mathbb{A})$ to a barycenter for the log-determinant divergence.

math.FA↗

arXiv2023

Weak log-majorization and inequalities of power means

As non-commutative versions of the quasi-arithmetic mean, we consider the Lim-P\'{a}lfia's power mean, R\'{e}nyi right mean and R\'{e}nyi power means. We prove that the Lim-P\'{a}lfia's power mean of order $t \in [-1,0)$ is weakly log-majorized by the log-Euclidean mean and fulfills the Ando-Hiai inequality. We establish the log-majorization relationship between the R\'{e}nyi relative entropy and the product of square roots of given variables. Furthermore, we show the norm inequalities among power means and provide the boundedness of R\'{e}nyi power mean in terms of the quasi-arithmetic mean.

math.QA↗

arXiv2023

Weak log-majorization between the geometric and Wasserstein means

There exist lots of distinct geometric means on the cone of positive definite Hermitian matrices such as the metric geometric mean, spectral geometric mean, log-Euclidean mean and Wasserstein mean. In this paper, we prove the log-majorization relation on the singular values of the product of given two positive definite matrices and their (metric and spectral) geometric means. We also establish the weak log-majorization between the spectra of two-variable Wasserstein mean and spectral geometric mean. In particular, we verify with certain condition on variables that two-variable Wasserstein mean converges decreasingly to the log-Euclidean mean with respect to the weak log-majorization.

math.FA↗

arXiv2022

Revisit on spectral geometric mean

In this paper we introduce the limit, unique solution of the nonlinear equations, geodesic property, tolerance relations and pinch on the spectral geometric mean for two positive definite operators. We show that the spectral geometric mean is a geodesic with respect to some semi-metric. We also prove that the tolerance relation on determinant one matrices can be characterized by the spectral geometric mean. Moreover, two positive tuples can be pinched by the spectral geometric mean.

math.FA↗

arXiv2022

Right mean for the $\alpha-z$ Bures-Wasserstein quantum divergence

A new quantum divergence induced from the $\alpha-z$ Renyi relative entropy, called the $\alpha-z$ Bures-Wasserstein quantum divergence, has been recently introduced. We investigate in this paper properties of the right mean, which is a unique minimizer of the weighted sum of $\alpha-z$ Bures-Wasserstein quantum divergences to each points. Many interesting operator inequalities of the right mean with the matrix power mean including the Cartan mean are presented. Moreover, we verify the trace inequality with the Wasserstein mean and provide bounds for the Hadamard product of two right means.

math-ph↗

arXiv2020

Operator inequalities and gyrolines of the weighted geometric means

We consider in this paper two different types of the weighted geometric means of positive definite operators. We show the component-wise bijection of these geometric means and give a geometric property of the spectral geometric mean as a metric midpoint. Moreover, several interesting inequalities related with the geometric means of positive definite operators will be shown. We also see the meaning of weighted geometric means in the gyrogroup structure with finite dimension and find the formulas of weighted geometric means of 2-by-2 positive definite matrices and density matrices.

math.FA↗

arXiv2020

Mixture and interpolation of the parameterized ordered means

Loewner partial order plays a very important role in metric topology and operator inequality on the open convex cone of positive invertible operators. In this paper we consider a family G of the ordered means for positive invertible operators equipped with homogeneity and properties related to the Loewner partial order such as the monotonicity, joint concavity, and arithmetic-G-harmonic weighted mean inequalities. Similar to the resolvent average, we construct a parameterized ordered mean and compare two types of the mixture of parameterized ordered means in terms of the Loewner order. We also show the relation between two families of parameterized ordered means associated with the power mean, monotonically interpolating given two parameterized ordered means.

math.FA↗

arXiv2019

Tensor product and Hadamard product for the Wasserstein means

As one of the least squares mean, we consider the Wasserstein mean of positive definite Hermitian matrices. We verify in this paper the inequalities of the Wasserstein mean related with a strictly positive and unital linear map, the identity of the Wasserstein mean for tensor product, and some inequalities of the Wasserstein mean for Hadamard product.

math.FA↗

arXiv2019

Parameterized Wasserstein mean with its properties

A new least squares mean of positive definite matrices for the divergence associated with the sandwiched quasi-relative entropy has been introduced. It generalizes the well-known Wasserstein mean for covariance matrices of Gaussian distributions with mean zero, so we call it the parameterized Wasserstein mean. We investigate in this article norm inequality of the parameterized Wasserstein mean, give its bounds with respect to the Loewner order, and show the extended version of Lie-Trotter-Kato formula for the parameterized Wasserstein mean. Finally we show the log-majorzation properties of the parameterized Wasserstein mean by using the Cartan mean.

math.FA↗

arXiv2018

Geometric Mean of Partial Positive Definite Matrices with Missing Entries

In this paper the geometric mean of partial positive definite matrices with missing entries is considered. The weighted geometric mean of two sets of positive matrices is defined, and we show whether such a geometric mean holds certain properties which the weighted geometric mean of two positive definite matrices satisfies. Additionally, counterexamples demonstrate that certain properties do not hold. A Loewner order on partial Hermitian matrices is also defined. The known results for the maximum determinant positive completion are developed with an integral representation, and the results are applied to the weighted geometric mean of two partial positive definite matrices with missing entries. Moreover, a relationship between two positive definite completions is established with respect to their determinants, showing relationship between their entropy for a zero-mean,multivariate Gaussian distribution. Computational results as well as one application are shown.

math.FA↗

arXiv2018

Bounds for the Wasserstein mean with applications to the Lie-Trotter mean

As the least squares mean for the Riemannian trace metric on the cone of positive definite matrices, the Riemannian mean with its computational and theoretical approaches has been widely studied. Recently the new metric and the least squares mean on the cone of positive definite matrices, which are called the Wasserstein metric and the Wasserstein mean, respectively, have been introduced. In this paper, we explore some properties of Wasserstein mean such as determinantal inequality and find bounds for the Wasserstein mean. Using bounds for the Wasserstein mean, we verify that the Wasserstein mean is the multivariate Lie-Trotter mean.

math.FA↗

arXiv2018

Structure for $g$-Metric Spaces and Related Fixed Point Theorems

In this paper, we propose a generalized notion of a distance function, called a $g$-metric. The $g$-metric with degree $n$ is a distance of $n+1$ points, generalizing the ordinary distance between two points and $G$-metric between three points. Indeed, it is shown that the $g$-metric with degree 1 (resp. degree 2) is equivalent to the ordinary metric (resp. the $G$-metric). Fundamental properties and several examples for the $g$-metric are also given. Moreover, topological properties on the $g$-metric space including the convergence of sequences and the continuity of mappings on the $g$-metric space are studied. Finally, we generalize some well-known fixed point theorems including Banach contraction mapping principle and \'Ciri\'c fixed point theorem in the $g$-metric space.

math.GN↗