Early Neuron Alignment in Two-layer ReLU Networks with Small Initialization

Authors: Hancheng Min, Enrique Mallada, Rene Vidal

ICLR 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical experiments on the MNIST dataset illustrate our theoretical findings.
Researcher Affiliation Academia Hancheng Min University of Pennsylvania hanchmin@seas.upenn.edu Enrique Mallada Johns Hopkins University mallada@jhu.edu René Vidal University of Pennsylvania vidalr@seas.upenn.edu
Pseudocode No The paper does not contain any explicitly labeled 'Pseudocode' or 'Algorithm' blocks.
Open Source Code No The paper does not contain any explicit statement about releasing source code or a link to a code repository for the methodology described.
Open Datasets Yes Numerical experiments on the MNIST dataset illustrate our theoretical findings.
Dataset Splits No The paper mentions using the MNIST dataset for numerical experiments but does not provide specific details on training, validation, or test splits (e.g., percentages or sample counts).
Hardware Specification No The paper does not specify any hardware details (e.g., GPU, CPU models, memory) used for running the experiments.
Software Dependencies No The paper does not provide specific software names with version numbers (e.g., programming languages, libraries, frameworks, or solvers) used for the experiments.
Experiment Setup Yes We build a two-layer Re LU network with h = 50 neurons and initialize all entries of the weights as [W]ij i.i.d. N 0, α2 , vj i.i.d. N 0, α2 , i [n], j [h] with α = 10 6. Then we run gradient descent on both W and v with step size η = 2 10 3.