Learning Task-General Representations with Generative Neuro-Symbolic Modeling

Authors: Reuben Feinman, Brenden M. Lake

ICLR 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply our model to the Omniglot challenge of human-level concept learning, using a background set of alphabets to learn an expressive prior distribution over character drawings. In a subsequent evaluation, our GNS model uses probabilistic inference to learn rich conceptual representations from a single training image that generalize to 4 unique tasks, succeeding where previous work has fallen short.
Researcher Affiliation Academia Reuben Feinman & Brenden M. Lake New York University {reuben.feinman,brenden}@nyu.edu
Pseudocode Yes The psuedo-code of this sampling procedure is provided in Fig. A7.
Open Source Code Yes Concept learning experiments can be reproduced using our pre-trained generative model and source code: https://github.com/rfeinman/GNS-Modeling.
Open Datasets Yes We apply our model to the Omniglot challenge of human-level concept learning, using a background set of alphabets to learn an expressive prior distribution over character drawings.
Dataset Splits No The paper uses the Omniglot dataset and mentions training on the "background set" and evaluating on "held-out alphabets in the Omniglot evaluation set," but it does not explicitly provide specific percentages or counts for training, validation, and test splits used in its experiments, nor does it refer to a predefined split by citation for its specific usage.
Hardware Specification No The paper does not specify the hardware used for running the experiments (e.g., specific GPU or CPU models, memory).
Software Dependencies No The paper does not specify specific software versions for its dependencies (e.g., "Python 3.8, PyTorch 1.9").
Experiment Setup No The paper mentions a temperature parameter (T=0.5) for sampling at test time and general training objectives, but it lacks specific hyperparameters (e.g., learning rate, batch size, number of epochs) for training the neural networks.