Towards a human-like understanding of the physical world
Abstract/Contents
- Abstract
- Humans have a rich understanding of the physical world, which enables us to effectively interact with and navigate in our environment. A core component of this understanding is the ability to decompose the visual environment into objects, and reason effectively about the dynamic interactions between these objects. We introduce an end-to-end differentiable graph neural network that learns to predict the physical dynamics of a wide variety of three-dimensional objects using a hierarchical particle-based object representation. Additionally, we present a physical prediction benchmark that compares a broad suite of state-of-the-art models to humans on a diverse set of physical phenomena. We show that graph neural networks with access to the physical state best capture human behavior, whereas models that receive only visual input fall far short of human accuracy. This suggests that endowing models with more physically explicit state representations is a promising path towards human-like physical understanding.
Description
Type of resource | text |
---|---|
Form | electronic resource; remote; computer; online resource |
Extent | 1 online resource. |
Place | California |
Place | [Stanford, California] |
Publisher | [Stanford University] |
Copyright date | 2022; ©2022 |
Publication date | 2022; 2022 |
Issuance | monographic |
Language | English |
Creators/Contributors
Author | Wang, Elias Ziyu |
---|---|
Degree supervisor | Yamins, Daniel |
Thesis advisor | Yamins, Daniel |
Thesis advisor | Boahen, Kwabena (Kwabena Adu) |
Thesis advisor | Finn, Chelsea |
Degree committee member | Boahen, Kwabena (Kwabena Adu) |
Degree committee member | Finn, Chelsea |
Associated with | Stanford University, Department of Electrical Engineering |
Subjects
Genre | Theses |
---|---|
Genre | Text |
Bibliographic information
Statement of responsibility | Elias Wang. |
---|---|
Note | Submitted to the Department of Electrical Engineering. |
Thesis | Thesis Ph.D. Stanford University 2022. |
Location | https://purl.stanford.edu/jy197dt9938 |
Access conditions
- Copyright
- © 2022 by Elias Ziyu Wang
- License
- This work is licensed under a Creative Commons Attribution Non Commercial 3.0 Unported license (CC BY-NC).
Also listed in
Loading usage metrics...