Learning physical parameters from dynamic scenes

Abstract

Humans acquire their most basic physical concepts early in development, and continue to enrich and expand their intuitive physics throughout life as they are exposed to more and varied dynamical environments. We introduce a hierarchical Bayesian framework to explain how people can learn physical parameters at multiple levels. In contrast to previous Bayesian models of theory acquisition (Tenenbaum, Kemp, Griffiths, & Goodman, 2011), we work with more expressive probabilistic program representations suitable for learning the forces and properties that govern how objects interact in dynamic scenes unfolding over time. We compare our model to human learners on a challenging task of estimating multiple physical parameters in novel microworlds given short movies. This task requires people to reason simultaneously about multiple interacting physical laws and properties. People are generally able to learn in this setting and are consistent in their judgments. Yet they also make systematic errors indicative of the approximations people might make in solving this computationally demanding problem with limited computational resources. We propose two approximations that complement the top-down Bayesian approach. One approximation model relies on a more bottom-up feature-based inference scheme. The second approximation combines the strengths of the bottom-up and top-down approaches, by taking the feature-based inference as its point of departure for a search in physical-parameter space.

A counterfactual simulation model of causal judgments for physical events

“…Further, we assume that our participants already understand how the physical world works. We do not model the learning process by which people arrive at this understanding Goodman, Ullman, & Tenenbaum, 2011;Kemp, Goodman, & Tenenbaum, 2010;Lake, Ullman, Tenenbaum, & Gershman, 2016;Piantadosi, Tenenbaum, & Goodman, 2012;Tenenbaum, Kemp, Griffiths, & Goodman, 2011;Ullman, Goodman, & Tenenbaum, 2012;Ullman, Stuhlmüller, Goodman, & Tenenbaum, 2014;Wellman & Gelman, 1992). The CSM operates in two steps: first, it uses a fine-grained test of difference-making to identify all candidate causes.…”

Intuitive Theories

“…This work has demonstrated how qualitative transitions in people's knowledge can be explained in terms of transitions between programs of different complexity (Piantadosi et al, 2012). While some first attempts have been made (Fragkiadaki, Agrawal, Levine, & Malik, submitted;Ullman, Stuhlmüller, Goodman, & Tenenbaum, 2014), further work is required to explain how people arrive at their rich intuitive theories of how the world works.…”

An integrative computational architecture for object-driven cortex

“…Humans can perform these tasks with little or no training, and the ability to do so is a key advantage of generative models over pattern recognition approaches such as neural network classifiers [17]. Subsequent work has shown how the framework extends more broadly across many aspects of intuitive physics, including predictions of future motion for rigidly colliding objects [18,19], predictions about the behavior of liquids (e.g., water, honey) [20,21] and granular materials (e.g., sand) [22,23], and judgments about objects' dynamic properties and interactions from how they move under gravity as well as latent forces such as magnetism [24,25].…”