Work

Training and Evaluating a Binary MNIST CNN with PyTorch

Year: 2026

Type: book

Abstract: A complete PyTorch implementation of a 180-parameter convolutional neural network for classifying handwritten MNIST digits 0 and 1. The program includes image normalisation, a stratified training-validation split, mini-batch stochastic gradient descent, automatic differentiation, and numerically stable binary cross-entropy loss. It reports loss and accuracy after each epoch, saves and restores the checkpoint with the lowest validation loss, and evaluates the selected model on held-out test images. Execution is supported on CPU and compatible CUDA devices.

Source: Zenodo (CERN European Organization for Nuclear Research)

DOI: 10.5281/zenodo.23970450

Author: Majumdar, Partha

Institution: Kalinga University

Language: English

Topic: Advanced Neural Network Applications

Subfield: Computer Vision and Pattern Recognition

Field: Computer Science

Domain: Physical Sciences

Open Access status: green

OpenAlex ID: W7225317285

FWCI:

Cites:

Cited by:

Related to: 10