Brown CS PhD student Kai Wang has just received an Adobe Research Fellowship for his research in automating design of structures and layouts. His work currently centers on bridging computer graphics and machine learning to create data-driven algorithms to achieve this automation.
The Adobe Research Fellowship program was created to recognize outstanding graduate students carrying out exceptional research. This research spans a broad array of subjects, including Computer Graphics, Computer Vision, Machine Learning, Natural Language Processing, and Programming Languages. The fellowship includes a monetary award, a chance to interview for an internship at Adobe, and an opportunity to work with an Adobe Research mentor. This year, Kai is one of ten winners of the award, joining fellow graduate students from MIT, Cornell, UC Berkeley, and other universities around the world.
Kai’s research thus far, he explains, has focused on the problem of creating algorithms that can automatically design indoor scenes. “We spend a lot of time in indoor spaces designed to support functionalities such as sleeping, cooking, meeting, etc,” Kai says. “To create the layouts of such spaces, a lot of human expertise and labor is often needed.” Kai's algorithms use deep neural networks to learn indoor scene layout patterns from large collections of existing layouts, enabling them to create plausible new scenes without ever being explicitly told how to do so. These algorithms can rapidly generate many new scene layouts, making them useful for design inspiration or to provide novel virtual training environments for autonomous robots. Going forward, Kai is interested in extending these machine learning algorithms to other types of layout problems (floor plans, for example) and other design problems that involve making sequences of decisions.
Over the past two years, Kai’s lab –led by Brown CS Professor Daniel Ritchie– has made significant inroads into this field of automatic indoor scene layout via machine learning. “We call this the scene synthesis problem,” Daniel explains. “Given the architectural specification of a room –its walls, doors, and windows– the goal is to create a plausible instance of a particular type of room, say a bedroom, by filling it with the right types of objects in the right arrangements. In doing so, we aim to build a system that can support a range of use cases.” Coming off the heels of a recent grant award, the researchers in Daniel’s lab continue to be recognized for their groundbreaking work.
For more information, click the link that follows to contact Brown CS Communications Outreach Specialist Jesse Polhemus.