Revealing interpretable object representations from human behavior

Authors: Charles Y. Zheng, Francisco Pereira, Chris I. Baker, Martin N. Hebart

ICLR 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We used Tensorflow (Abadi et al., 2015) to fit the model (3) to the 1,450,119 triplets collected, using a 90-10 train-validation split to pick the regularization parameter λ.
Researcher Affiliation Academia Charles Y. Zheng Section on Functional Imaging Methods National Institute of Mental Health charles.zheng@nih.gov Francisco Pereira Section on Functional Imaging Methods National Institute of Mental Health francisco.pereira@nih.gov Chris I. Baker Section on Learning and Plasticity National Institute of Mental Health martin.hebart@nih.gov Martin N. Hebart Section on Learning and Plasticity National Institute of Mental Health martin.hebart@nih.gov
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any explicit statements about making its source code available or links to a code repository.
Open Datasets No The collection of the Odd-one-out behavioral dataset is an ongoing project by the authors. ... We plan to collect additional triplets and release all data by the end of the study.
Dataset Splits Yes We randomly split the dataset into ntrain triplets for training and nval for choosing the sparsity regularization parameter λ. ... using a 90-10 train-validation split to pick the regularization parameter λ.
Hardware Specification No The paper mentions 'Portions of this study used the high-performance computational capabilities of the Biowulf Linux cluster at the National Institutes of Health, Bethesda, MD (biowulf.nih.gov)' but does not specify details such as GPU models, CPU models, or memory.
Software Dependencies No The paper mentions 'Tensorflow (Abadi et al., 2015)' and the 'Adam algorithm (Kingma & Ba, 2015)' but does not provide specific version numbers for these or other software dependencies.
Experiment Setup Yes We used the Adam algorithm (Kingma & Ba, 2015) with an initial learning rate of 0.001 to minimize the objective function, using a fixed number of 1,000 epochs over the training set, which was sufficient to ensure convergence.