Learning abstract structure for drawing by efficient motor program induction

Authors: Lucas Tian, Kevin Ellis, Marta Kryven, Josh Tenenbaum

NeurIPS 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Our behavioral data provides evidence that humans indeed perform few-shot updating of their inductive biases by learning program-like drawing procedures that guide generalization behavior. We describe a program-induction algorithm that discovers new abstract, compositional, drawing routines given the same limited training data given to humans. Moreover, the model s learned drawing behavior captures certain diagnostic features of how humans generalize. These results suggest that abstraction and compositionality are key principles for explaining how humans rapidly learn program-like structure that guides reasoning and planning in drawing.
Researcher Affiliation Academia 1Brain and Cognitive Sciences, MIT 2Center for Brains, Minds and Machines 3Laboratory of Neural Systems, Rockefeller University
Pseudocode No The paper describes algorithms conceptually and refers to a figure (Figure 2) for an overall model, but does not present pseudocode in a structured, code-like block.
Open Source Code No The paper does not provide a direct link to open-source code for the methodology.
Open Datasets No The paper mentions generating stimuli for training sets but does not provide concrete access information (link, DOI, specific repository, or formal citation) for the dataset used for training.
Dataset Splits No The paper mentions 250 stimuli generated for each training set and 18 test images, but does not specify validation splits or other detailed dataset split information.
Hardware Specification No The paper states that the experiment was presented "on a touchscreen device (phone, tablet, or laptop)", but this describes the device used by subjects, not the hardware used to run the models or experiments. No specific GPU/CPU or other experimental hardware details are provided.
Software Dependencies No The paper mentions "Psi Turk [34] and Raphael Sketch Pad2" and provides a URL for the latter, but does not include specific version numbers for any software dependencies.
Experiment Setup No The paper describes the experimental procedure for human subjects and conceptual aspects of the model but lacks specific hyperparameters or system-level training settings for the computational model in the main text.