The recent adaptation of deep neural network-based methods to reinforcement learning and planning domains has yielded remarkable progress on individual tasks. Nonetheless, progress on task-to-task transfer remains limited. In pursuit of efficient and robust generalization, we introduce the Schema Network, an object-oriented generative physics simulator capable of disentangling multiple causes of events and reasoning backward through causes to achieve goals. The richly structured architecture of the Schema Network can learn the dynamics of an environment directly from data. We compare Schema Networks with Asynchronous Advantage Actor-Critic and Progressive Networks on a suite of Breakout variations, reporting results on training efficiency and zero-shot generalization, consistently demonstrating faster, more robust learning and better transfer. We argue that generalizing from limited data and learning causal relationships are essential abilities on the path toward generally intelligent systems.
Ken Kansky (Vicarious AI)
Tom Silver (Vicarious AI)
David A Mély (Vicarious AI)
Mo Eldawy (Vicarious AI)
Miguel Lazaro-Gredilla (Vicarious AI)
Xinghua Lou (Vicarious AI)
Nimrod Dorfman (Vicarious AI)
Szymon Sidor (OpenAI)
Scott Phoenix (Vicarious AI)
Scott Phoenix is the cofounder and CEO of Vicarious, an AI company that is building artificial general intelligence for robots. Vicarious has raised over $135 million in funding from luminaries including Mark Zuckerberg, Elon Musk, Peter Thiel, and Jeff Bezos. Prior to co-founding Vicarious, Mr. Phoenix was entrepreneur-in-residence at Founders Fund, and a CXO at OnlySecure (acquired byNetShops) and at MarchingOrder (Ben Franklin Partners). He is an advisor to Felicis Ventures and 8VC in the areas of AI and hard technology investments and advises the nonprofit Thorn, led by Ashton Kutcher and Demi Moore, on how to apply AI technology to fight child abuse and sex trafficking. He is a global advocate for the development of safe AI. He earned his BAS in Computer Science and Entrepreneurship from the University of Pennsylvania.
Dileep George (Vicarious AI)
Related Events (a corresponding poster, oral, or spotlight)
2017 Talk: Schema Networks: Zero-shot Transfer with a Generative Causal Model of Intuitive Physics »
Mon Aug 7th 08:09 -- 08:27 AM Room C4.6 & C4.7