UC Berkeley alumna wins award from Association for Computing Machinery for dissertation

Related Posts

UC Berkeley alumna Chelsea Finn was announced as the winner of the 2018 Association for Computing Machinery, or ACM, Doctoral Dissertation Award for her work on approaches in machine learning and robotics.

Finn’s dissertation, “Learning to Learn with Gradients,” introduced meta-learning algorithms that allow deep networks to “solve new tasks from small datasets” and have applications in areas such as “computer vision, reinforcement learning and robotics,” according to the ACM website.

Finn is recognized as an expert and “has developed some of the most effective methods to teach robots skills to control and manipulate objects,” according to ACM.

According to Finn, her work was motivated by her desire to teach machines to do many different things and to apply meta-learning methods to robotics to allow machines to quickly learn new behavior.

“We were motivated by the fact that a lot of machine learning would train (machines) in one task. … The motivation behind our work was for machines to be more adaptable (and) to build more with experience — not trained in one individual task, but in many tasks,” Finn said. “My thesis was on algorithms that enabled systems to build upon previous data and experience to quickly learn new tasks. … (These are the) sort of capabilities that (are) in meta-learning algorithms.”

Finn’s methods are classified as model-agnostic meta-learning, or MAML. Highlighted in the dissertation, Finn’s MAML methods were able to use raw camera pixels from a single human demonstration in order to teach robots reaching and placing.

“By computing gradients and applying gradients, it’s basically how the neural network should change as a function of the task. You can kind of embed one optimization into another optimization,” Finn said.

According to ACM, “Finn’s MAML methods have had a tremendous impact on the field and have been widely adopted in reinforcement learning, computer vision and other fields of machine learning.”

Finn added that she was grateful for her advisers’ support in her dissertation and for the freedom they afforded her.

Finn received her doctorate in electrical engineering and computer sciences from UC Berkeley and has been a postdoctoral researcher at the Berkeley Artificial Intelligence Research Lab.

ACM annually presents the Doctoral Dissertation Award, which has an accompanying prize of $20,000, to the best doctoral dissertation in computer science and engineering.

Finn’s work will be published in the ACM Digital Library as part of the ACM Books series. In the fall, she will start a full-time appointment as an assistant professor at Stanford University after finishing her post at Google Brain.

“I’ve loved living in Berkeley over the past 4 1/2 years, and I think (it) has been amazing to be part of a collaborative group, and a lot of camaraderie, (and) a collection of really intelligent folks,” Finn said.

Contact Sarah Chung at [email protected] and follow her on Twitter at @sarahchungdc.