[2402.09146] ResQuNNs:Towards Enabling Deep Learning in Quantum Convolution Neural Networks

PDF display of the paper entitled RESQUNNS: Towards Enabling deep learning in Nerve Network

PDF HTML (experimental) view

a summary:In this paper, we offer a new framework to enhance the performance of the QUNNS (Qunns) by introducing training -to -revolutionary layers and addressing the critical challenges associated with it. The traditional fixed layers, although they are useful for extracting features, were largely fixed, providing limited adaptation. Unlike the latest latest, our research overcomes this restriction by enabling training within these layers, which greatly increases the flexibility and Qunns capabilities. However, the insertion of multiple -trained -to -training layers leads to complications in the gradual improvement, due primarily due to the difficulty of reaching gradients across these layers. To solve this, we suggest a new structure, the RESqunns, and take advantage of the concept of the remaining learning, facilitating the flow of gradients by adding skipping connections between the layers. By inserting the remaining blocks between the fixed layers, we guarantee access to augmented gradient throughout the network, which improves training performance. Moreover, we provide experimental evidence of the strategic position of these remaining blocs within the Qunns. Through an intensive experience, we define an effective composition of the remaining blocks, which enables gradients across all layers in the network that ultimately leads to effective training. The results we have found indicate that the exact location of the remaining blocs plays a decisive role in maximizing performance gains in Qunns. Our results define a big step forward in the development of quantum deep learning, providing new ways for both theoretical development and practical quantum computing applications.