Collaborative: Systems for Quantum Learning: Perceptrons and Networks

量子学习系统：感知器和网络两个激进的想法已经从自然科学渗透到计算机科学和工程中：计算机可以像生物系统一样学习，计算可以像物理系统一样量子力学。这些想法分别导致了人工神经网络和机器学习理论，以及量子计算。这两种计算模型的实际优势是正交的：前者用于解决没有明显算法方法的复杂系统的图像分类和控制等实际问题；后者（不考虑大规模量子计算机尚未建成的事实）只有一个已知的算法，用于解决一个具体的实际问题——分解大数——但该算法比已知的最好的经典算法运行得快得多。该项目结合了这两种计算模型。它将出于实际原因解决量子学习的可能性。最终目标是开发量子学习模型，使其具有与经典模型类似的优势——能够为没有明显量子算法解决方案的问题找到量子解决方案。在这个方向上只采取了最基本的步骤，因此该项目将自下而上地实现这一目标。PI将研究一个特定的架构（受生物学启发）：量子感知器，更一般地说，量子神经网络。然后将仔细定义一个完全量子感知器，它可以连接到一个网络中，分析它的能力，为它推导学习规则，将这些规则应用于各种任务，并将这些目标扩展到量子中性网络。这项工作将强调设计具有相应经典系统无法实现的性能的量子系统的重要性。

AbstractCollaborative(0202087) David Meyer, UCSD & (0202055) Mitchell Rothstein, U of GeorgiaSystems for Quantum Learning: Perceptrons and NetworksTwo radical ideas have infiltrated computer science and engineering from the natural sciences: that computers might learn, like biological systems, and that computation might he quantum mechanical, like physical systems. These ideas have led to artificial neural networks and machine learning theory, and to quantum computation, respectively. The practical advantages of these two computational models are orthogonal: the former are used to attack practical problems like image classification and control of complex systems for which there are no obvious algorithmic approaches; the latter (putting aside the fact that large scale quantum computers have not yet been built) has a known algorithm for only one specific practical problem-factoring large numbers--but that algorithm runs immensely faster than the best classical algorithm known.The project combines both of these computational models. It will address the possiblility of quantum learning for practical reasons. The ultimate goal is to develop quantum learning models which have strengths analogous to classical ones-the ability to find quantum solutions to problems for which there is no apparent quantum algorithmic solution. Only the most rudimentary steps have been taken in this direction-hence the project will approach this goal from the bottom up. The PI's will investigate a specific architecture (inspired by biology): quantum perceptrons, and more generally, quantum neural networks. Then will carefully define a completely quantum perceptron which can be connected into a network, analyze its capacity, derive learning rules for it, apply these rules to various tasks, and extend each of these objectives to quantum neutral networks. This work will emphasize the importance of designing quantum systems with performance not achievable by corresponding classical systems.