A Random Matrix Approach to Echo-State Neural Networks

Authors: Romain Couillet, Gilles Wainrib, Hafiz Tiomoko Ali, Harry Sevi

ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In order to validate the results of Section 3.1 and Section 3.2, we provide in Figure 2 an example of simulated versus asymptotic performance for a prediction task over the popular Mackey Glass model (Glass & Mackey, 1979). ... Comparison between Monte Carlo simulations (Monte Carlo) and theory from Propositions 1–2 (Th. (fixed W)) or Corollary 2 (Th. (limit)).
Researcher Affiliation Academia Romain Couillet ROMAIN.COUILLET@CENTRALESUPELEC.FR Centrale Sup elec, 91192 Gif-sur-Yvette, FRANCE Gilles Wainrib GILLES.WAINRIB@ENS.FR Ecole Normale Suprieure, 75005 Paris, FRANCE Hafiz Tiomoko Ali HAFIZ.TIOMOKOALI@CENTRALESUPELEC.FR Centrale Sup elec, 91192 Gif-sur-Yvette, FRANCE Harry Sevi HARRY.SEVI@ENS-LYON.FR ENS Lyon, 69000 Lyon, FRANCE
Pseudocode No The paper does not contain structured pseudocode or algorithm blocks.
Open Source Code No The paper does not provide any statement or link indicating that source code for the methodology described is publicly available.
Open Datasets Yes In order to validate the results of Section 3.1 and Section 3.2, we provide in Figure 2 an example of simulated versus asymptotic performance for a prediction task over the popular Mackey Glass model (Glass & Mackey, 1979).
Dataset Splits No The paper mentions "training phase" and "testing phase" and refers to T and ˆT as training/testing durations (e.g., "The ESN will be trained for a period T and tested for a period ˆT"), but it does not specify exact dataset split percentages, sample counts, or refer to predefined splits with citations for the dataset partitioning needed for reproduction. There is no explicit mention of a separate validation set.
Hardware Specification No The paper does not provide specific hardware details (like GPU/CPU models, processor types, or memory) used for running its experiments or simulations. It only refers to "Monte Carlo simulations" and "simulated versus asymptotic performance."
Software Dependencies No The paper does not specify any software dependencies with version numbers (e.g., programming languages, libraries, or specific simulation environments) used in the experiments.
Experiment Setup Yes The paper provides specific parameters for its simulations, such as reservoir size n, training and testing durations T and ˆT, and internal noise variance η^2. For instance, Figure 2 states: "n = 200, T = ˆT = 400 (top) and n = 400, T = ˆT = 800 (bottom)." Figure 4 describes "W Haar with σ = .99, σ = .9, σ = .5". Figure 7 uses "1% or 10% impulsive N(0, .01) noise pollution" and "W Haar with σ = .9, n = 400, T = ˆT = 1000". These are explicit details about the experimental setup.