MLP and CNN Architectures for Sparse Training

该数据集包含了用于评估高稀疏网络中Exact Orthogonal Initialization（EOI）方法性能的各种架构。具体包括对具有特定配置的1000层多层感知机（MLP）和1000层卷积神经网络（CNN）的实验。数据集涵盖了多种激活函数（如线性、双曲正切、硬双曲正切、ReLU），并评估了从0%到97%的多个稀疏度水平。这些实验均基于1000层网络的规模进行，旨在评估深度网络中稀疏初始化方法的性能。

This dataset contains various architectures for evaluating the performance of the Exact Orthogonal Initialization (EOI) method in highly sparse deep neural networks. Specifically, it includes experiments on 1000-layer multi-layer perceptrons (MLPs) and 1000-layer convolutional neural networks (CNNs) with specific configurations. The dataset covers a variety of activation functions, such as linear, hyperbolic tangent, hard hyperbolic tangent, and ReLU, and evaluates multiple sparsity levels ranging from 0% to 97%. All experiments are conducted on 1000-layer networks, aiming to assess the performance of sparse initialization methods in deep neural networks.