> ML_LITERATURE // LECUN-1998-GRADIENT-BASED-LEARNING-APPLIED-TO-DOCUMENT-RECOGNITION_v1.0
Gradient-Based Learning Applied to Document Recognition
Yann LeCun, Léon Bottou, Yoshua Bengio, Patrick Haffner · Proceedings of the IEEE (1998)
foundational1998foundationalthirdPartyReproduced
Principal Contribution
Introduced LeNet-5, establishing convolutional layers, weight sharing, and spatial subsampling for 2D visual recognition.
Operational Relevance
Architectural foundation for modern CNNs in optical character recognition, check processing, and embedded computer vision.
Assumptions
- Local receptive fields, shared weights, and spatial subsampling enforce shift, scale, and distortion invariance in images
Limitations
- Constrained by 1990s CPU compute and small image resolutions (32x32 pixels); struggled on complex real-world scenes