Results 251 to 260 of about 6,431,216 (308)
Some of the next articles are maybe not open access.
Continuity of Best Hankel Approximation and Convergence of Near-Best Approximants
SIAM Journal on Control and Optimization, 1994Summary: Consider a bounded Hankel operator \(\Gamma\) with \(s\)-numbers \(s_ 0\geq s_ 1\geq\cdots\) and a sequence of bounded Hankel operators \(\Gamma_ n\) converging to \(\Gamma\) in the operator norm. In this paper, it is shown that for each \(k\) with \(s_{k-1}> s_ k\geq s_{k+1}\geq\cdots\), the rational symbols of the best rank-\(k\) Hankel ...
Chui, Charles K., Li, Xin
openaire +1 more source
Theoretical guarantees on the best-of-n alignment policy
International Conference on Machine LearningA simple and effective method for the inference-time alignment and scaling test-time compute of generative models is best-of-$n$ sampling, where $n$ samples are drawn from a reference policy, ranked based on a reward function, and the highest ranking one
Ahmad Beirami +6 more
semanticscholar +1 more source
Variational Best-of-N Alignment
International Conference on Learning RepresentationsBest-of-N (BoN) is a popular and effective algorithm for aligning language models to human preferences. The algorithm works as follows: at inference time, N samples are drawn from the language model, and the sample with the highest reward, as judged by a
Afra Amini, Tim Vieira, Ryan Cotterell
semanticscholar +1 more source
An algorithm for best rational approximation based on barycentric rational interpolation
Numerical Algorithms, 2021C. Hofreither
semanticscholar +1 more source
EMBEDDING THEOREMS AND BEST APPROXIMATIONS
Mathematics of the USSR-Sbornik, 1975We establish necessary and sufficient conditions, in terms of best approximations, for a function in () to belong to (p$ SRC=http://ej.iop.org/images/0025-5734/26/2/A04/tex_sm_2477_img4.gif/>). The proofs depend on the properties of equimeasurable functions, which were applied by Ul'janov in the theory of the embedding of certain classes for (MR 37
openaire +2 more sources
The Best Interpolating Approximation is a Limit of Best Weighted Approximations
Canadian Mathematical Bulletin, 1982AbstractUnder appropriate conditions it is shown that the best interpolating approximation to a given function in the uniform norm is a limit of best unconstrained approximations with respect to a certain sequence of discontinuous weight functions.
openaire +1 more source
Best approximations by rational functions
Mathematical Notes of the Academy of Sciences of the USSR, 1971Description of a general class of real continuous functions cn a segment Δ of the real line for which a best rational approximation with complex coefficients is not unique.
openaire +3 more sources
Best Approximation with Walsh Atoms
Constructive Approximation, 1997The author connects the theoretical algorithmic approximation rate in \(L^2 (\overline R)\) for the model case where \(W\) consist of Walsh atoms. The main result stated implies that a linear combination of \(K\) atoms can be approximated to relative error \(e\) with linear combination of \(O(K^2 \log(1/E)\)-orthogonal atoms. In finite dimension \(N\),
openaire +2 more sources
Connectedness and solarity in problems of best and near-best approximation
, 2016A. Alimov, I. G. Tsar'kov
semanticscholar +1 more source
SSRN Electronic Journal, 1971
The central theme of this paper is the minimax approximation of a function rotated by an angle about the origin and the study of the optimal angle of rotation of least minimax error. Existence theorems indicate validity for any choice of norm. In the following formulation of the problem we will include, with obvious modifications, the cases when the ...
openaire +1 more source
The central theme of this paper is the minimax approximation of a function rotated by an angle about the origin and the study of the optimal angle of rotation of least minimax error. Existence theorems indicate validity for any choice of norm. In the following formulation of the problem we will include, with obvious modifications, the cases when the ...
openaire +1 more source

