General Sequential Episodic Memory Model
Authors: Arjun Karuvally, Terrence Sejnowski, Hava T Siegelmann
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We demonstrate its dense capacity under polynomial activation functions. We validate this for the case of n = 1 with simulation in Figure 2. |
| Researcher Affiliation | Academia | 1College of Information and Computer Sciences, University of Massachusetts Amherst 2Computational Neurobiology Laboratory, The Salk Institute for Biological Studies. |
| Pseudocode | No | The paper provides mathematical equations and descriptions of the model's dynamics, but it does not include any clearly labeled pseudocode or algorithm blocks. |
| Open Source Code | Yes | The code for the simulations are available in the repository: https://github.com/arjun23496/gsemm. |
| Open Datasets | No | The paper describes the generation of synthetic memory sequences for simulations: 'Each memory in the model is a random binary vector such that Pr h ξ(i) j = +1 i = Pr h ξ(i) j = 1 i = 1/2. These memories are organized as 2 separate cyclical episodes: ξ1 ξ2 ξ3 ξ1 and ξ4 ξ5 ξ6 ξ7 ξ4 with their sequential relationships stored as an adjacency matrix in Φ.' It does not use or provide access to a publicly available, pre-existing dataset. |
| Dataset Splits | No | The paper describes how synthetic memory sequences are created and used in simulations (e.g., 'These memories are organized as 2 separate cyclical episodes...'). However, it does not specify traditional training, validation, or test dataset splits with percentages or counts, as the data is generated for the model's simulation rather than being a pre-existing dataset. |
| Hardware Specification | No | The paper does not provide any specific details about the hardware used for running the simulations, such as CPU or GPU models, memory specifications, or cloud computing environments. |
| Software Dependencies | No | The paper mentions using 'the fourth order Runge-Kutta numerical procedure' for its simulations but does not specify any particular software libraries, frameworks (e.g., PyTorch, TensorFlow), or their version numbers that were utilized. |
| Experiment Setup | Yes | We simulated Dense GSEMM with Nf = 100, αs = 0.05, αc = 0.007, Tf = 1.0, and Td = 20.0. The power of the polynomial non-linearity is 1 for the energy surface shown in the figure. The parameters Nf = 100, Tf = 1.0, Td = 20.0, αs = 1.0 are fixed. The parameter αc {0.01, 0.02, 0.03, 0.04, 0.05, 0.06, 0.07, 0.08, 0.09, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9, 1.0} is treated as hyperparameter and optimized to find the maximum number of memories that can be stored for each [seed, polynomial] pair. |