We introduce a hybrid machine-learning algorithm for designing quantum optics experiments to produce specific quantum states. Our algorithm successfully found experimental schemes to produce all 5 states we asked it to, including Schrödinger cat states and cubic phase states, all to a fidelity of over 96%. Here we specifically focus on designing realistic experiments, and hence all of the algorithm's designs only contain experimental elements that are available with current technology. The core of our algorithm is a genetic algorithm that searches for optimal arrangements of the experimental elements, but to speed up the initial search we incorporate a neural network that classifies quantum states. The latter is of independent interest, as it quickly learned to accurately classify quantum states given their photon-number distributions.As artificial intelligence (AI) and machine learning develop, their range of applicability continues to grow. They are now being utilised in the fast-growing field of quantum machine learning [1-3], with one particular application demonstrating that AI is an effective tool for designing quantum physics experiments [4][5][6][7][8]. In this vein, here we introduce a hybrid algorithm that designs and optimises quantum optics experiments for producing a range of useful quantum states, including Schrödinger cat states [9][10][11] and cubic phase states [12].The core of our algorithm, named AdaQuantum [13] and introduced in [14] and [4], uses a genetic algorithm to search for optimal arrangements of quantum optics experimental equipment. Any given arrangement will output a quantum state of light, and the algorithm's task is to optimise the arrangement to find states with specific properties. To assess the suitability of a given state, we require a fitness function that takes as input a quantum state and outputs a number -the fitness value -that quantifies whether the state has the properties we desire or not. Our previous works largely focused on quantum metrology, where our algorithm found quantum states with substantial improvements over the alternatives in the literature [4,14]. While in [4,14] the fitness function assessed the phase-measuring capabilities of the states, in this paper instead we look at producing a range of useful and interesting states (introduced below) to a high fidelity, and hence we use as our fitness function the fidelity to our target states.The search space for our genetic algorithm is huge, which means that typically the algorithm has to simulate and evaluate a vast number of quantum optics experiments in order to finally find strong solutions. This introduces a major challenge in our work: the speed, efficiency, and effectiveness of our algorithm depend in a large part on simulating and evaluating the experiments as quickly and accurately as possible. If short-cuts could be found that allow a given experimental set-up to be * Contact email address: Paul.Knott@nottingham.ac.uk evaluated approximately without the full simulation being performed, then this would...

No abstract