The video discusses the evolution of convolutional neural networks (CNNs), starting with the pioneering work of John Lee Cohn at AT&T Bell Labs, who developed LeNet for handwritten digit recognition. LeNet's architecture, including its layers and non-linearities, is detailed, highlighting its significance in the 1990s. The video also covers LeNet's impact on machine learning, its adoption in ATMs, and the transition to more advanced CNN architectures like AlexNet, VGG, and ResNet.

Who developed the first convolutional neural network for handwritten digit recognition?

• Yann LeCun

• Geoffrey Hinton

• Andrew Ng

• John Lee Cohn

What was the input image size used in the LeNet architecture?

• 32 by 32

• 28 by 28

• 64 by 64

• 16 by 16

How many filters were used in the first convolutional layer of LeNet?

• 8

• 10

• 12

• 6

What type of nonlinearity was initially used in LeNet?

• ReLU

• Sigmoid

• Tanh

• Leaky ReLU

What is the size of the resultant volume after the second average pooling in LeNet?

• 7 by 7 by 16

• 3 by 3 by 16

• 5 by 5 by 16

• 10 by 10 by 16

How many units are in the first fully connected layer of LeNet?

• 100

• 64

• 84

• 120

What was one of the practical applications of LeNet in the 1990s?

• Image classification

• Digit recognition for ATM deposits

• Speech recognition

• Facial recognition