Semantically-correlated memories in a dense associative model
Authors: Thomas F Burns
ICML 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Experimental demonstrations showcase CDAM s efficacy in handling realworld data, replicating a classical neuroscience experiment, performing image retrieval, and simulating arbitrary finite automata. |
| Researcher Affiliation | Academia | 1Institute for Computational and Experimental Research in Mathematics, Brown University, USA 2Sci AI Center, Cornell University, USA 3Neural Coding and Brain Computing Unit, OIST Graduate University, Japan. |
| Pseudocode | No | The paper does not contain a structured pseudocode or algorithm block. |
| Open Source Code | Yes | A copy of the code used is available at https://github.com/tfburns/CDAM. |
| Open Datasets | Yes | Here I use a directed cycle graph # C50 where the patterns are sparsely sampled frames of videos (see Appendix A.11 for details). The two videos used were sourced from Wikimedia Commons and were uploaded by User:Raul654 on 24 January 2006. They can found at the below URLs: https://commons.wikimedia.org/wiki/File: Gorilla_gorilla_gorilla1.ogv https://commons.wikimedia.org/wiki/File: Gorilla_gorilla_gorilla4.ogv... I test CDAM on the Fashion MNIST dataset (Xiao et al., 2017) |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits. The experiments focus on demonstrating model dynamics and recall, not typical supervised learning evaluation with predefined splits. |
| Hardware Specification | Yes | All numerical simulations were performed on a Lenovo x260 Think Pad laptop computer using the Python 3 programming language. |
| Software Dependencies | No | The paper mentions 'Python 3 programming language' but does not specify version numbers for any key libraries, frameworks, or solvers used. |
| Experiment Setup | Yes | Unless otherwise stated, in the following numerical analyses I used n = 1, 000, β = 1, and η = 0.1. ...To initialise the network state, I chose a memory pattern µ and set σ(0) = ξµ + cζ, where ζ is a random vector with elements independently drawn from the interval [ 0.5, 0.5] and c R+ is the amplitude of the additive random noise, here c = 1. |