MNIST-1D

经典深度学习基准测试的极简、低内存和低计算替代方案。训练示例比 MNIST 示例小 20 倍，但它们更清楚地区分线性、非线性和卷积模型，分别达到 32、68 和 94% 的准确度（这些模型在 MNIST 上获得 94、99+ 和 99+%） .A minimalist, low-memory and low-computation alternative to classic deep learning benchmarks. The training examples are 20 times smaller than their MNIST counterparts, yet they more clearly differentiate the performance of linear, nonlinear, and convolutional models, achieving accuracies of 32%, 68%, and 94% respectively (these models attain 94%, 99+%, and 99+% on MNIST).

搜集汇总

背景与挑战

背景概述

MNIST-1D是经典深度学习基准测试的极简替代方案，具有低内存和低计算需求。其训练示例比MNIST小20倍，能更清晰地区分线性、非线性和卷积模型，准确度分别达到32%、68%和94%。