Gradient-Based Learning Applied to Document Recognition

By Yann LeCun et al

Published on Nov. 10, 1998

Read the original document by opening this link in a new tab.

Table of Contents

Abstract
Introduction
Learning from Data
Gradient-Based Learning

Summary

Multilayer Neural Networks trained with the backpropagation algorithm constitute the best example of a successful Gradient-Based Learning technique. Real-life document recognition systems are composed of multiple modules including field extraction, segmentation, recognition, and language modeling. The study considers the tasks of handwritten character recognition and compares the performance of several learning techniques on a benchmark dataset for handwritten digit recognition. The authors discuss the importance of automatic learning and the shift towards relying more on learning machines rather than hand-designed heuristics. They introduce the concept of Graph Transformer Networks for a unified and well-principled design paradigm. Various neural network architectures specialized for pattern recognition tasks are explored along with the benefits of global training methods. The document also covers the application of Graph Transformer Networks in reading bank checks, demonstrating record accuracy in commercial use. The text delves into the principles of learning from data, gradient-based learning, and optimization algorithms used in pattern recognition systems.