Learning to design without prior data: Discovering generalizable design strategies using deep learning and tree search

ABSTRACT Building an AI agent that can design on its own has been a goal since the 1980s. Recently, deep learning has shown the ability to learn from large-scale data, enabling significant advances in data-driven design. However, learning over prior data limits us only to solve problems that have been solved before and biases data-driven learning towards existing solutions. The ultimate goal for a design agent is the ability to learn generalizable design behavior in a problem space without having seen it before. We introduce a self-learning agent framework in this work that achieves this goal. This framework integrates a deep policy network with a novel tree search algorithm, where the tree search explores the problem space, and the deep policy network leverages self-generated experience to guide the search further. This framework first demonstrates an ability to discover high-performing generative strategies without any prior data, and second, it illustrates a zero-shot generalization of generative strategies across various unseen boundary conditions. This work evaluates the effectiveness and versatility of the framework by solving multiple versions of two engineering design problems without retraining. Overall, this paper presents a methodology to self-learn high-performing and generalizable problem-solving behavior in an arbitrary problem space, circumventing the needs for expert data, existing solutions, and problem-specific learning. Published in ASME Journal of Mechanical Design. Published online November 11 2022. INTRODUCTION: Solving design problems is one of the most ubiquitous processes in engineering and arguably the most challenging [1,2]. The design automation research paradigm aims to augment the continually evolving design solving process by enabling machines to engage in design. Despite decades of research in the area, modern-day automated design synthesis is still heavily guided by handcrafted rules and prior expert data, making it susceptible to non-generalizability and errors resulting from human bias [3,4]. Developing a design agent that can learn from scratch is still a long-standing challenge. This paper addresses this challenge by introducing a generalizable design agent framework that integrates newly developed tree search and deep learning methods. The tree search enables exploration and information gathering, while the deep learning representation helps the agent leverage self-generated experience. Together, these methods provide a symbiotic integration of decision-making methods to effectively explore and learn in unknown design problem spaces. Learning problem-solving strategies from scratch has been achieved in multiple domains [5-8]. Some of these methods use a dual-process decision-making framework which is often compared to the slow and fast thinking ideology [9]s.