We are interested in AI-driven design problems. Below are some areas of immediate interest:
This area investigates mathematical and computational frameworks for the automated discovery of novel designs. The goal is to develop AI methods (like Generative Adversarial Networks, Graph Neural Networks, etc.) which can learn from existing designs and discover novel designs which exceed the performance of all existing designs. Discovering new designs may require finding suitable representation scheme for heterogenous designs (sketches, CAD models, text), performing efficient search over such representation and learning human preferences. The methods will be applied to real-world design examples and they will help us identify what products can be introduced in the market.
Our goal in this area is to do fundamental research to enable the accelerated design and development of materials by developing and creatively integrating theory, manufacturing, and computational approaches with rigorous engineering design principles. Existing topology optimization methods cannot handle these requirements. To address this gap, we work on combining representation learning of material unit cells (metamaterials) with surrogate modeling and combinatorial search methods to solve very large inverse design problems. A typical example is designing a new prosthetic arm with thousands of unit cells, spatially varying properties, complex non-linear physics and manufacturing constraints.
We are interested in transforming the way people design products by democratizing design — where teams of globally distributed people work together to design physical products. This will enable designers to express complex ideas from anywhere in the world and firms to process thousands of ideas efficiently. To achieve this goal, we worked on principled methods for representation, learning, and optimization of discrete problems occurring in design. In the past, we investigated three main questions to help organizations and distributed teams: