Density modelling is the task of learning an unknown probability density function from samples, and is one of the central problems of unsupervised machine learning. In this work, we show that there exists a density modelling problem for which fault-tolerant quantum computers can offer a super-polynomial advantage over classical learning algorithms, given standard cryptographic assumptions. Along the way, we provide a variety of additional results and insights, of potential interest for proving future distribution learning separations between quantum and classical learning algorithms. Specifically, we (a) provide an overview of the relationships between hardness results in supervised learning and distribution learning, and (b) show that any weak pseudo-random function can be used to construct a classically hard density modelling problem. The latter result opens up the possibility of proving quantum-classical separations for density modelling based on weaker assumptions than those necessary for pseudo-random functions.
翻译:密度建模的任务是从样本中学习未知的概率密度函数,这是不受监督的机器学习的中心问题之一。 在这项工作中,我们表明存在一个密度建模问题,根据标准的加密假设,过错耐用量子计算机可以提供比经典学习算法的超级多极性优势。 沿途,我们提供了各种额外的结果和洞察力,这些结果和洞察力对证明量子和经典学习算法之间未来分布性学习分离具有潜在兴趣。 具体而言,我们(a) 提供了监督学习和分配学习的硬性结果之间的关系概览,以及(b) 表明任何薄弱的伪随机功能都可以用来构建典型的硬性密度建模问题。 后一种结果开启了根据比伪随机功能所需的假设更弱的假设来证明密度建模的量子级分离的可能性。