Let $\sigma$ be a first-order signature and let $\mathbf{W}_n$ be the set of all $\sigma$-structures with domain $\{1, \ldots, n\}$. By an inference framework we mean a class $\mathbf{F}$ of pairs $(\mathbb{P}, L)$, where $\mathbb{P} = (\mathbb{P}_n : n = 1, 2, 3, \ldots)$ and $\mathbb{P}_n$ is a probability distribution on $\mathbf{W}_n$, and $L$ is a logic with truth values in the unit interval $[0, 1]$. An inference framework $\mathbf{F}'$ is asymptotically at least as expressive as another inference framework $\mathbf{F}$ if for every $(\mathbb{P}, L) \in \mathbf{F}$ there is $(\mathbb{P}', L') \in \mathbf{F}'$ such that $\mathbb{P}$ is asymptotically total-variation-equivalent to $\mathbb{P}'$ and for every $\varphi(\bar{x}) \in L$ there is $\varphi'(\bar{x}) \in L'$ such that $\varphi'(\bar{x})$ is asymptotically equivalent to $\varphi(\bar{x})$ with respect to $\mathbb{P}$. This relation is a preorder and we describe a partial order on the equivalence classes of some inference frameworks that seem natural in the context of machine learning and artificial intelligence. Several previous results about asymptotic (or almost sure) equivalence of formulas or convergence in probability can be formulated in terms of relative asymptotic strength of inference frameworks. We incorporate these results in our classification of inference frameworks and prove two new results. Both concern sequences of probability distributions defined by directed graphical models that use ``continuous'' aggregation functions. The first considers queries expressed by a logic with truth values in $[0, 1]$ which employs continuous aggregation functions. The second considers queries expressed by a two-valued conditional logic that can express statements about relative frequencies.
翻译:设 $\sigma$ 为一个一阶签名,$\mathbf{W}_n$ 为所有定义在 $\{1, \ldots, n\}$ 上的 $\sigma$-结构的集合。我们称一个推理框架为一个由成对元 $(\mathbb{P}, L)$ 组成的类 $\mathbf{F}$,其中 $\mathbb{P}=(\mathbb{P}_n: n=1,2,3,\ldots)$ 为 $\mathbf{W}_n$ 上的概率分布,而 $L$ 是带值域在 $[0, 1]$ 上的逻辑。如果对于每个 $(\mathbb{P}, L) \in \mathbf{F}$,存在 $(\mathbb{P}', L') \in \mathbf{F}'$ 使得 $\mathbb{P}$ 在总变差距离下渐近等价于 $\mathbb{P}'$,并且对于每个在 $L$ 中的 $\varphi(\bar{x})$,都有一个在 $L'$ 中的 $\varphi'(\bar{x})$ 使得 $\varphi'(\bar{x})$ 关于 $\mathbb{P}$ 渐近等价于 $\varphi(\bar{x})$,那么我们称推理框架 $\mathbf{F}'$ 相对渐近表现力至少不劣于另一个推理框架 $\mathbf{F}$。这种关系是一个偏序关系,并且我们通过一些自然的机器学习和人工智能背景下的推理框架的等价类之间描述了一部分偏序关系。在这个框架中,我们还能归纳之前一些关于渐近等价性或概率收敛性的结果。我们将这些结果纳入我们对推理框架的分类中,证明了两个新结果。它们涉及到使用“连续”聚合函数定义的有向图模型所定义的概率分布序列的查询。第一个结果考虑采用带有值域在 $[0, 1]$ 上的连续聚合函数的逻辑所表示的查询。第二个结果关注于采用二值条件逻辑的查询,该逻辑可以表达关于相对频率的语句。