A Compositional Object-Based Approach to Learning Physical Dynamics
Name
1612.00341.pdf
Size
4.26 MB
Format
Adobe PDF
Checksum (MD5)
659c6945e4b7197c2cb5055a19ac3fdc
Author(s)
Chang, Michael B.
Ullman, Tomer David
Torralba, Antonio
Tenenbaum, Joshua B
Date Issued
2017
Journal
5th International Conference on Learning Representations (ICLR 2017)
Publisher
ICLR
Citation
Chang, Michael B. et al. "A Compositional Object-Based Approach to Learning Physical Dynamics." 5th International Conference on Learning Representations (ICLR 2017), April 24-26 2017, Toulon, France, ICLR, 2017
Version
Original manuscript
Abstract
We present the Neural Physics Engine (NPE), an object-based neural network architecture for learning predictive models of intuitive physics. We propose a factorization of a physical scene into composable object-based representations and also the NPE architecture whose compositional structure factorizes object dynamics into pairwise interactions. Our approach draws on the strengths of both symbolic and neural approaches: like a symbolic physics engine, the NPE is endowed with generic notions of objects and their interactions, but as a neural network it can also be trained via stochastic gradient descent to adapt to specific object properties and dynamics of different worlds. We evaluate the efficacy of our approach on simple rigid body dynamics in two-dimensional worlds. By comparing to less structured architectures, we show that our model's compositional representation of the structure in physical interactions improves its ability to predict movement, generalize to different numbers of objects, and infer latent properties of objects such as mass.
MIT Department
Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences
Massachusetts Institute of Technology. Department of Electrical Engineering and Computer Science
Terms of Use
Creative Commons Attribution-Noncommercial-Share Alike
Persistent DSpace Link
DOI of Published Version
http://www.iclr.cc/doku.php?id=iclr2017:main
