Scaling Laws for Associative Memories

Authors: Vivien Cabannes, Elvis Dohmatob, Alberto Bietti

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We provide extensive numerical experiments to validate and interpret theoretical results, including fine-grained visualizations of the stored memory associations.
Researcher Affiliation Industry Vivien Cabannes FAIR, Meta Elvis Dohmatob FAIR, Meta Alberto Bietti Flatiron Institute
Pseudocode No The paper does not contain any clearly labeled pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating that the source code for its methodology is publicly available.
Open Datasets No The paper states: “The data follows a Zipf law with α = 0.5, N = 100, M = 5 and f(x) = x mod. M.” indicating that synthetic data was generated according to a specified distribution, not obtained from a publicly available dataset with concrete access information.
Dataset Splits No The paper studies generalization error as a function of the number of data (T) and model capacity (d), rather than providing specific training, validation, or test dataset splits from a fixed dataset.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, or cloud computing resources) used for running experiments.
Software Dependencies No The paper does not provide specific version numbers for any software libraries or dependencies used in the experiments. While PyTorch is cited in the bibliography, its version is not specified as a dependency for reproducibility.
Experiment Setup Yes All the experiments are conducted with N = 100, M = 5, α = 2, f (x) = x mod M, averaged over ten runs. We initialized parameters and rescale learning rates to ensure maximal feature updates, as explained in Appendix B.1.