Relating Hopfield Networks to Episodic Control
Authors: Hugo Chateau-Laurent, Frederic Alexandre
NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Finally, we empirically show that the dictionary outperforms the Max separation function, which had previously been argued to be optimal, and that performance can further be improved by replacing the Euclidean distance kernel by a Manhattan distance kernel. |
| Researcher Affiliation | Academia | Hugo Chateau-Laurent Inria centre of the University of Bordeaux, France IMN, CNRS UMR 5293, France La BRI, CNRS UMR 5800, France hugo.chateaulaurent@gmail.com Frederic Alexandre Inria centre of the University of Bordeaux, France IMN, CNRS UMR 5293, France La BRI, CNRS UMR 5800, France frederic.alexandre@inria.fr |
| Pseudocode | No | The paper does not contain explicit pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code is available at https://github.com/Hugo Chateau Laurent/DND_Associative Memory and is based on https://github.com/Beren Millidge/Theory_Associative_Memory (MIT license). |
| Open Datasets | Yes | In this section, the MNIST, CIFAR10 and Tiny Image Net datasets are used to test the robustness and capacity of DND as an associative memory model, using the same methods as for the other UHN instances (7) unless otherwise mentioned2. |
| Dataset Splits | No | The paper does not explicitly provide training/validation/test dataset splits (percentages or counts). It mentions 'sets of 100 images' for evaluation but not formal splits. |
| Hardware Specification | No | Experiments presented in this paper were carried out using the PlaFRIM experimental testbed, supported by Inria, CNRS (LABRI and IMB), Université de Bordeaux, Bordeaux INP and Conseil Régional d Aquitaine (see https://www.plafrim.fr). |
| Software Dependencies | No | The paper does not provide specific software names with version numbers for dependencies. |
| Experiment Setup | Yes | simi(K, q) = 1 δ + ||K i q||2 2 , (3) with K the observation keys, q the query, and δ = 10 3. ... For each dataset, 100 images are encoded. The noise is set to 1 for MNIST and 0.75 for CIFAR10 dataset and Tiny Image Net. |