The success of machine learning (ML) \u2014 particularly deep neural networks (NNs) \u2014 has revolutionized numerous industries. But these technologies have become increasingly complex, requiring ever-larger amounts of memory and energy for training. And as Moore\u2019s law falters, we are facing a looming limit to their computational power5.<\/p>\n\n\n\n

The underlying reason for this bottleneck is quantum mechanics: Each additional qubit doubles the computing power of a NN. To illustrate, a coin flipping experiment shows that more heads and tails can be simultaneously represented when a coin is entangled with its sister coin.<\/p>\n\n\n\n

Fortunately, research is underway on ways to solve this problem by harnessing the power of quantum mechanics. Many recent studies have explored a variety of methods for building a \u201cquantum NN,\u201d and some have even achieved impressive performance on noisy intermediate-scale quantum devices. But none of these attempts directly confronts the fundamental incompatibility between NNs and QC. Specifically, they do not allow neurons in a NN to share the outputs of the neurons in previous layers.<\/p>\n\n\n\n