Compositional inductive biases in function learning
How do people recognize and learn about complex functional structure? Taking inspiration from other areas of cognitive science, we propose that this is achieved by harnessing compositionality: complex structure is decomposed into simpler building blocks. We formalize this idea within the framework o...
| Main Authors: | , , , , |
|---|---|
| Other Authors: | |
| Format: | Article |
| Language: | English |
| Published: |
Elsevier BV
2021
|
| Online Access: | https://hdl.handle.net/1721.1/133410 |
| _version_ | 1826204038137380864 |
|---|---|
| author | Schulz, Eric Tenenbaum, Joshua B Duvenaud, David Speekenbrink, Maarten Gershman, Samuel J |
| author2 | Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences |
| author_facet | Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences Schulz, Eric Tenenbaum, Joshua B Duvenaud, David Speekenbrink, Maarten Gershman, Samuel J |
| author_sort | Schulz, Eric |
| collection | MIT |
| description | How do people recognize and learn about complex functional structure? Taking inspiration from other areas of cognitive science, we propose that this is achieved by harnessing compositionality: complex structure is decomposed into simpler building blocks. We formalize this idea within the framework of Bayesian regression using a grammar over Gaussian process kernels, and compare this approach with other structure learning approaches. Participants consistently chose compositional (over non-compositional) extrapolations and interpolations of functions. Experiments designed to elicit priors over functional patterns revealed an inductive bias for compositional structure. Compositional functions were perceived as subjectively more predictable than non-compositional functions, and exhibited other signatures of predictability, such as enhanced memorability and reduced numerosity. Taken together, these results support the view that the human intuitive theory of functions is inherently compositional. |
| first_indexed | 2024-09-23T12:47:50Z |
| format | Article |
| id | mit-1721.1/133410 |
| institution | Massachusetts Institute of Technology |
| language | English |
| last_indexed | 2024-09-23T12:47:50Z |
| publishDate | 2021 |
| publisher | Elsevier BV |
| record_format | dspace |
| spelling | mit-1721.1/1334102023-12-12T16:01:41Z Compositional inductive biases in function learning Schulz, Eric Tenenbaum, Joshua B Duvenaud, David Speekenbrink, Maarten Gershman, Samuel J Massachusetts Institute of Technology. Department of Brain and Cognitive Sciences Massachusetts Institute of Technology. Computer Science and Artificial Intelligence Laboratory Center for Brains, Minds, and Machines How do people recognize and learn about complex functional structure? Taking inspiration from other areas of cognitive science, we propose that this is achieved by harnessing compositionality: complex structure is decomposed into simpler building blocks. We formalize this idea within the framework of Bayesian regression using a grammar over Gaussian process kernels, and compare this approach with other structure learning approaches. Participants consistently chose compositional (over non-compositional) extrapolations and interpolations of functions. Experiments designed to elicit priors over functional patterns revealed an inductive bias for compositional structure. Compositional functions were perceived as subjectively more predictable than non-compositional functions, and exhibited other signatures of predictability, such as enhanced memorability and reduced numerosity. Taken together, these results support the view that the human intuitive theory of functions is inherently compositional. 2021-10-27T19:52:42Z 2021-10-27T19:52:42Z 2017 2019-09-26T16:18:33Z Article http://purl.org/eprint/type/JournalArticle https://hdl.handle.net/1721.1/133410 en 10.1016/J.COGPSYCH.2017.11.002 Cognitive Psychology Creative Commons Attribution-NonCommercial-NoDerivs License http://creativecommons.org/licenses/by-nc-nd/4.0/ application/pdf Elsevier BV bioRxiv |
| spellingShingle | Schulz, Eric Tenenbaum, Joshua B Duvenaud, David Speekenbrink, Maarten Gershman, Samuel J Compositional inductive biases in function learning |
| title | Compositional inductive biases in function learning |
| title_full | Compositional inductive biases in function learning |
| title_fullStr | Compositional inductive biases in function learning |
| title_full_unstemmed | Compositional inductive biases in function learning |
| title_short | Compositional inductive biases in function learning |
| title_sort | compositional inductive biases in function learning |
| url | https://hdl.handle.net/1721.1/133410 |
| work_keys_str_mv | AT schulzeric compositionalinductivebiasesinfunctionlearning AT tenenbaumjoshuab compositionalinductivebiasesinfunctionlearning AT duvenauddavid compositionalinductivebiasesinfunctionlearning AT speekenbrinkmaarten compositionalinductivebiasesinfunctionlearning AT gershmansamuelj compositionalinductivebiasesinfunctionlearning |