Machine Learning for Developers

Lenet 5

As we saw in the historical introduction of the convolutional neural networks, convolutional layers were discovered during the 1980s. But the available technology wasn't powerful enough to build complex combinations of layers until the end of the 1990s.

Around 1998, in Bell Labs, during research around the decodification of handwritten digits, Ian LeCun formed a new approach—a mix of convolutional, pooling, and fully connected layers—to solve the problem of recognizing handwritten digits.

At this time, SVM and other much more mathematically defined problems were used more or less successfully, and the fundamental paper on CNNs shows that neural networks could perform comparatively well with the then state-of-the-art methods.

In the ...