People Infer Recursive Visual Concepts from Just a Few Examples

Lake, Brenden M.; Piantadosi, Steven T.

doi:10.1007/s42113-019-00053-y

Cited by 23 publications

(22 citation statements)

References 42 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We allow these to take absolute feature values, like color(r')⇐blue, but also values relative to the agent or recipient’s pre-interaction features, such as color(r') ⇐color(a) or edge(r') ⇐edge(r)+ 1. These causal functions natively capture many kinds of causal theories people may entertain, as confirmed by their self-reports and our model fits (see also Bramley et al, 2018 ; Goodman et al, 2008 ; Lake & Piantadosi 2020 ). Moreover, by grounding causal functions in such object-based representations, these causal functions naturally generalize to novel objects.…”

Section: Discussionsupporting

confidence: 71%

“…Symbolic approaches enable compositionality and systematicity, while the sub-symbolic techniques, especially the fast, incremental approximations, make this more scalable to real-world data (Bramley et al, 2017 ). This framework also draws a close link with probabilistic program induction models (e.g., Bramley et al, 2018 ; Ellis et al, 2021 ; Lake et al, 2015 ; Lake & Piantadosi 2020 ), where causal beliefs and concepts can be viewed as programs, and accurate generalizations can be viewed as a evidence for successful program synthesis whereby these programs increasingly reflect the true causal laws of nature. We believe our modeling framework can be extended to broader generalization cases beyond causal cognition, and contributes to the collective effort for a hybrid approach in understanding human cognition (Lake et al, 2017 ; Oaksford et al, 2007 ; Valentin et al, 2021 ).…”

Section: Discussionmentioning

confidence: 95%

See 1 more Smart Citation

How Do People Generalize Causal Relations over Objects? A Non-parametric Bayesian Account

2021

View full text Add to dashboard Cite

How do people decide how general a causal relationship is, in terms of the entities or situations it applies to? What features do people use to decide whether a new situation is governed by a new causal law or an old one? How can people make these difficult judgments in a fast, efficient way? We address these questions in two experiments that ask participants to generalize from one (Experiment 1) or several (Experiment 2) causal interactions between pairs of objects. In each case, participants see an agent object act on a recipient object, causing some changes to the recipient. In line with the human capacity for few-shot concept learning, we find systematic patterns of causal generalizations favoring simpler causal laws that extend over categories of similar objects. In Experiment 1, we find that participants’ inferences are shaped by the order of the generalization questions they are asked. In both experiments, we find an asymmetry in the formation of causal categories: participants preferentially identify causal laws with features of the agent objects rather than recipients. To explain this, we develop a computational model that combines program induction (about the hidden causal laws) with non-parametric category inference (about their domains of influence). We demonstrate that our modeling approach can both explain the order effect in Experiment 1 and the causal asymmetry, and outperforms a naïve Bayesian account while providing a computationally plausible mechanism for real-world causal generalization.

show abstract

Section: Discussionsupporting

confidence: 71%

Section: Discussionmentioning

confidence: 95%

How Do People Generalize Causal Relations over Objects? A Non-parametric Bayesian Account

2021

View full text Add to dashboard Cite

show abstract

“…The learning as programming approach, however, is importantly different in providing learners the full expressive power of symbolic programs both theoretically (i.e., Turing completeness) and practically (i.e., freedom to adopt any formal syntax). This approach applies broadly to developmental phenomena, including counting [52], concept learning [13,53], function words [54], kinship [55], theory learning [56,57], lexical acquisition [23], question answering [15], semantics and pragmatics [25,58,59], recursive reasoning [60], sequence transformation [61], sequence prediction [18,62], structure learning [63], action concepts [64], perceptual understanding [14,65], and causality [66]. These applications build on a tradition of studying agents who understand the world by inferring computational processes that could have generated observed data, which is optimal in a certain sense [67,68], and aligns with rational constructivist models of development [69][70][71][72].…”

Section: Trends Trends In In Cognitivementioning

confidence: 99%

The Child as Hacker

Rule

Tenenbaum

Piantadosi

2020

Trends in Cognitive Sciences

Self Cite

View full text Add to dashboard Cite

The scope of human learning and development poses a radical challenge for cognitive science. We propose that developmental theories can address this challenge by adopting perspectives from computer science. Many of our best models treat learning as analogous to computer programming because symbolic programs provide the most compelling account of sophisticated mental representations. We specifically propose that children's learning is analogous to a particular style of programming called hacking, making code better along many dimensions through an open-ended set of goals and activities. By contrast to existing theories, which depend primarily on local search and simple metrics, this view highlights the many features of good mental representations and the multiple complementary processes children use to create them. Hacking as a Metaphor for Learning in Cognitive DevelopmentHuman cognitive development is qualitatively unique. Though humans are born unusually helpless, they amass an unparalleled cognitive repertoire, including: intuitive theories for domains like physics and biology, formal theories in mathematics and science, language comprehension and production, and complex perceptual and motor skills. Children also learn to learn, producing higher-order knowledge that enriches existing concepts and enhances future learning. Humanlike performance in any one of these domains seems substantially beyond current efforts in artificial intelligence. Yet, human children essentially acquire these abilities simultaneously and universally. They may even discover new ideas that radically alter humanity's understanding of the world. The foundational fields of cognitive science, including philosophy, psychology, neuroscience, and computer science, face a radical challenge in explaining the richness of human development.

show abstract

“…It also draws a close link with probabilistic program induction models (e.g. Bramley et al, 2018;Ellis et al, 2020;Lake et al, 2015;Lake & Piantadosi, 2020), where causal beliefs and concepts can be viewed as programs, and accurate generalizations can be viewed as a evidence for successful program synthesis whereby these programs increasingly reflect the true causal laws of nature.…”

Section: Constructive Cognitionmentioning

confidence: 96%

How do people generalize causal relations over objects? A non-parametric Bayesian account

Zhao¹,

Lucas²,

Bramley³

2021

Preprint

View full text Add to dashboard Cite

How do people decide how general a causal relationship is, in terms of the entities or situations it applies to? What features do people use to decide whether a new situation is governed by a new causal law or an old one? How can people make these difficult judgments in a fast, efficient way? We address these questions in two experiments that ask participants to generalize from one (Experiment 1) or several (Experiment 2) causal interactions between pairs of objects. In each case, participants see an agent object act on a recipient object, causing some changes to the recipient. In line with the human capacity for few-shot concept learning, we find systematic patterns of causal generalizations favoring simpler causal laws that extend over categories of similar objects. In Experiment 1, we find that participants’ inferences are shaped by the order of the generalization questions they are asked. In both experiments, we find an asymmetry in the formation of causal categories: participants preferentially identify causal laws with features of the agent objects rather than recipients. To explain this, we develop a computational model that combines program induction (about the hidden causal laws) with non-parametric category inference (about their domains of influence). We demonstrate that our modeling approach can both explain the order effect in Experiment 1 and the causal asymmetry, and outperforms a naive Bayesian account while providing a computationally plausible mechanism for real world causal generalization.

show abstract

People Infer Recursive Visual Concepts from Just a Few Examples

Cited by 23 publications

References 42 publications

How Do People Generalize Causal Relations over Objects? A Non-parametric Bayesian Account

How Do People Generalize Causal Relations over Objects? A Non-parametric Bayesian Account

The Child as Hacker

How do people generalize causal relations over objects? A non-parametric Bayesian account

Contact Info

Product

Resources

About