Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in [1].
Memory of recurrent networks: Do we compute it right?
Authors: Giovanni Ballarin, Lyudmila Grigoryeva, Juan-Pablo Ortega
JMLR 2024 | Venue PDF | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Numerical evaluations of the memory capacity (MC) of recurrent neural networks reported in the literature often contradict well-established theoretical bounds. ... Simulations show that the memory curves that are recovered using the proposed methods fully agree with the theory. ... All codes necessary to reproduce numerical results presented in the paper are publicly available at https://github.com/Learning-of-Dynamic-Processes/memorycapacity. |
| Researcher Affiliation | Academia | Giovanni Ballarin EMAIL Department of Economics University of Mannheim Germany Lyudmila Grigoryeva EMAIL Faculty of Mathematics and Statistics Universit at Sankt Gallen Switzerland Department of Statistics (Honorary Assoc. Prof.) University of Warwick United Kingdom Juan-Pablo Ortega EMAIL Division of Mathematical Sciences School of Physical and Mathematical Sciences Nanyang Technological University Singapore |
| Pseudocode | Yes | Algorithm 1: Averaged Orthogonalized Subspace Method (OSM+) |
| Open Source Code | Yes | All codes necessary to reproduce numerical results presented in the paper are publicly available at https://github.com/Learning-of-Dynamic-Processes/memorycapacity. |
| Open Datasets | No | The paper primarily deals with theoretical analysis and numerical simulations using generated stochastic processes (e.g., 'mean zero and variance one process (zt)T t=1', 'i.i.d. Gaussian with Var(zt) = γ(0) = 1') rather than specific external datasets with predefined access information. No named public datasets are utilized. |
| Dataset Splits | No | The paper primarily deals with theoretical analysis and numerical simulations using generated stochastic processes (e.g., 'mean zero and variance one process', 'i.i.d. Gaussian with Var(zt) = γ(0) = 1') rather than specific external datasets with predefined splits for training, validation, and testing. Therefore, explicit dataset split information is not applicable or provided. |
| Hardware Specification | No | The paper does not provide specific hardware details such as GPU/CPU models, processor types, or memory amounts used for running its experiments. It only mentions using 'MATLAB' for computations, which is software. |
| Software Dependencies | No | The paper states: 'Computations are performed in MATLAB with the standard double-precision of floating point numbers eps = 2 52 2.2 10 16 marked with the black horizontal solid line.' While MATLAB is mentioned, a specific version number is not provided, nor are any other software dependencies with their versions. |
| Experiment Setup | Yes | Figure 1 illustrates the case when N = 100, A is a scaled random orthogonal matrix, and C is a 2-norm-scaled random normal vector. Taking τmax = 500 to estimate MC... The results of our simulations with LESN models of size N = 100 are shown in Figure 3... For OSM+ the input mask C is resampled L = 1000 times to compute the average memory curve... |