851. |
852. User-defined trade-offs in LLM benchmarking : balancing accuracy, scale, and sustainabilityAna Gjorgjevikj, Ana Nikolikj, Barbara Koroušić-Seljak, Tome Eftimov, 2025, original scientific article Abstract: This paper presents xLLMBench, a transparent, decision-centric benchmarking framework that empowers decision-makers to rank large language models (LLMs) based on their preferences across diverse, potentially conflicting performance and non-performance criteria, e.g., domain accuracy, model size, energy consumption, CO emissions. Existing LLM benchmarking methods often rely on individual performance criteria (metrics) or human feedback, so methods systematically combining multiple criteria into a single interpretable ranking lack. Methods considering human preferences typically rely on direct human feedback to determine rankings, which can be resource-intensive and not fully aligned with application-specific requirements. Motivated by current limitations of LLM benchmarking, xLLMBench leverages multi-criteria decision-making methods to provide decision-makers with the flexibility to tailor benchmarking processes to their requirements. It focuses on the final step of the benchmarking process (robust analysis of benchmarking results) which in LLMs’ case often involves their ranking. The framework assumes that the selection of datasets, metrics, and LLMs involved in the experiment is conducted following established best practices. We demonstrate xLLMBench’s usefulness in two scenarios: combining LLM results for one metric across different datasets and combining results for multiple metrics within one dataset. Our results show that while some LLMs maintain stable rankings, others exhibit significant changes when correlated datasets are removed, when the focus shifts to contamination-free datasets or fairness metrics. This highlights that LLMs have distinct strengths/weaknesses, going beyond overall performance. Our sensitivity analysis reveals robust rankings, while the diverse visualizations enhance transparency. xLLMBench can be used with existing platforms to support transparent, reproducible, and contextually-meaningful LLM benchmarking.
Keywords: large language models, benchmarking, multi-criteria decision-making
Published in DiRROS: 01.10.2025; Views: 316; Downloads: 137
 Full text (33,40 MB)
This document has many files! More... |
853. |
854. |
855. Capacity building to support forest management in protective forests of SloveniaKristina Sever, Milan Kobal, Matjaž Guček, Andrej Breznikar, Aleš Poljanec, 2025, published scientific conference contribution abstract Keywords: protective forest, adaptive forest management, Forest Living Lab, natural hazard mitigation, marteloscope
Published in DiRROS: 01.10.2025; Views: 250; Downloads: 108
Full text (108,61 KB)
This document has many files! More... |
856. |
857. |
858. Construction of exceptional copositive matricesTea Štrekelj, Aljaž Zalar, 2025, original scientific article Abstract: An $n \times n$ symmetric matrix $A$ is copositive if the quadratic form $x^TAx$ is nonnegative on the nonnegative orthant ${\mathbb R}^n_{\ge 0}$. The cone of copositive matrices contains the cone of matrices which are the sum of a positive semidefinite matrix and a nonnegative one and the latter contains the cone of completely positive matrices. These are the matrices of the form $BB^T$ for some $n \times r$ matrix $B$ with nonnegative entries. The above inclusions are strict for $n\ge 5$. The first main result of this article is a free probability inspired construction of exceptional copositive matrices of all sizes $\ge 5$ i.e., copositive matrices that are not the sum of a positive semidefinite matrix and a nonnegative one. The second contribution of this paper addresses the asymptotic ratio of the volume radii of compact sections of the cones of copositive and completely positive matrices. In a previous work by Klep and the authors, it was shown that, by identifying symmetric matrices naturally with quartic even forms, and equipping them with the $L^2$ inner product and the Lebesgue measure, the ratio of the volume radii of sections with a suitably chosen hyperplane is bounded below by a constant independent of $n$ as $n$ tends to infinity. In this paper, we complement this result by establishing an analogous bound when the sections of the cones are unit balls in the Frobenius inner product.
Keywords: copositive matrix, completely positive matrix, positive polynomial, sum of squares, convex cone
Published in DiRROS: 01.10.2025; Views: 223; Downloads: 90
Full text (809,21 KB)
This document has many files! More... |
859. On reduced Hamilton walksAleksander Malnič, Rok Požar, 2026, original scientific article Abstract: A Hamilton walk in a finite graph is a walk, either open or closed, that traverses every vertex at least once. Here, we introduce Hamilton walks that are reduced in the sense that they avoid immediate backtracking: a reduced Hamilton walk never traverses the same edge forth and back consecutively. While every connected graph admits a Hamilton walk, existence of a reduced Hamilton walk is not guaranteed for all graphs. However, we prove that a reduced Hamilton walk does exist in a connected graph with minimal valency at least $2$. Furthermore, given such a graph on $n$ vertices, we present an $O(n^2)$-time algorithm that constructs a reduced Hamilton walk of length at most $n(n+3)/2$. Specifically, for a graph belonging to a family of regular expander graphs, we can find a reduced Hamilton walk of length at most $c(6n−2)\log n+2n$, where $c$ is a constant independent of $n$.
Keywords: algorithm, Hamilton walk, nonstandard metric, reduced walk
Published in DiRROS: 01.10.2025; Views: 164; Downloads: 74
Full text (1,91 MB)
This document has many files! More... |
860. Deformations of an affine Gorenstein toric pairMatej Filip, 2026, original scientific article Abstract: We consider deformations of a pair $(X,\partial X)$, where $X$ is an affine toric Gorenstein variety and $\partial X$ is its boundary. We compute the tangent and obstruction space for the corresponding deformation functor and for an admissible lattice degree $m$ we construct the miniversal deformation of $(X,\partial X)$ in degrees $-km$, for all $k\in{\mathbb N}$. This in particular generalizes Altmann's construction of the miniversal deformation of an isolated Gorenstein toric singularity to an arbitrary non-isolated Gorenstein toric singularity. Moreover, we show that the irreducible components of the reduced miniversal deformation are in one to one correspondence with maximal Minkowski decompositions of the polytope $P\cap(m=1)$, where $P$ is the lattice polytope defining $X$.
Keywords: deformation theory, toric singularities
Published in DiRROS: 01.10.2025; Views: 184; Downloads: 61
Full text (1,04 MB)
This document has many files! More... |
