Designing complex industrial scale deep neural network models requires extensive engineering expertise in multidisciplinary domains ,is time consuming and since it is a labor intensive work, is prone to human error. Handcrafted neural models are the hallmark of today’s advancements inAI . However, there has been an idea that such design processes for optimal architectures can be automated ,giving rise to the thriving research area of automatic deep neural network design. As opposed to hyperparameter optimization, automatic DL methods such as Neural Architecture search have been devised that from scratch, given data from a given domain,one can search from the search space of model motifs for the most optimal architecture ,taking advantage of routines such as reinforcement learning. Most of these algorithms have traditionally been implemented on high end classical ASICS such as TPUs, GPUs and however advanced such hardware can be, there is still computational overheads since neural architecture search algorithms are very greedy. Quantum computing has thrived the past decade and nascent technologies have paved the way to designing various platforms on which to run quantum algorithms including Superconducting qubits, quantum optics, ion traps and crystal topologies. Here, we design a neural architecture search algorithm for a neural model for predicting protein folding problems and run it on the existing quantum photonic chip accessed through cloud service . In the process we seek to demonstrate quantum computational advantage through exponential speed up of the above neural architecture search algorithm, thus ushering in the era of Quantum ASICS specific for automated artificial intelligence.
Hey @Yego, welcome to the forum. Sounds interesting!
thank you very much @isaacdevlugt for your kind reply; I am still trying to figure out how to attack this one from the ground up… someone wish me luck