精确实数搜索:构造性单价数学中的形式化优化与回归
Exact Real Search: Formalised Optimisation and Regression in Constructive Univalent Mathematics
AI总结:
本文在构造性单价类型论中提出一个基于可搜索类型和接近空间的形式化框架,用于在精确实数编码类型上实现并证明全局优化与参数回归算法的收敛性。
摘要:
实数在数学和计算理论中都很重要。在计算上,实数可以用多种方式表示;最常用的是不精确的浮点数据类型,但也可以使用满足实数预期数学性质的精确任意精度数据类型。本论文关注于形式化某些精确实数算术类型的性质,并将其用于搜索、优化和回归的计算目的。我们在构造性和单价类型论数学基础中,开发了一个形式化框架,用于在广泛的类型上执行搜索、优化和回归。该框架利用 Martín Escardó 先前关于可搜索类型的工作,以及一种方便的超度量空间版本——我们称之为接近空间——以便使用函数式编程语言和证明助手 Agda 一致地搜索某些无限类型。我们正式定义并证明了全局优化和参数回归的类型论变体的收敛性质,这些问题是分析文献中与搜索相关的问题。由于我们在构造性环境中工作,这些收敛定理为在我们框架的类型上进行正确的优化和回归提供了计算算法。重要的是,我们可以将我们的框架实例化到精确实数算术文献中的数据类型上,使我们能够在实数的三进制有符号数字编码上,以及 Hans-J. Boehm 的实数函数编码的简化版本上,执行我们的搜索、优化和回归变体。此外,我们通过使用 Escardó-Simpson 紧区间的区间对象规范,正式验证了某些精确实数算术运算的定义,为关于三进制有符号数字的广泛工作做出了贡献。
英文摘要:
The real numbers are important in both mathematics and computation theory. Computationally, real numbers can be represented in several ways; most commonly using inexact floating-point data-types, but also using exact arbitrary-precision data-types which satisfy the expected mathematical properties of the reals. This thesis is concerned with formalising properties of certain types for exact real arithmetic, as well as utilising them computationally for the purposes of search, optimisation and regression. We develop, in a constructive and univalent type-theoretic foundation of mathematics, a formalised framework for performing search, optimisation and regression on a wide class of types. This framework utilises Martín Escardó's prior work on searchable types, along with a convenient version of ultrametric spaces -- which we call closeness spaces -- in order to consistently search certain infinite types using the functional programming language and proof assistant Agda. We formally define and prove the convergence properties of type-theoretic variants of global optimisation and parametric regression, problems related to search from the literature of analysis. As we work in a constructive setting, these convergence theorems yield computational algorithms for correct optimisation and regression on the types of our framework. Importantly, we can instantiate our framework on data-types from the literature of exact real arithmetic, allowing us to perform our variants of search, optimisation and regression on ternary signed-digit encodings of the real numbers, as well as a simplified version of Hans-J. Boehm's functional encodings of real numbers. Furthermore, we contribute to the extensive work on ternary signed-digits by formally verifying the definition of certain exact real arithmetic operations using the Escardó-Simpson interval object specification of compact intervals.
机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。