09.05.2023 Views

pdfcoffee

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Convolutional Neural Networks

In general, deep networks require more training data available to fully express their

power (see Figure 11):

Figure 11: Accuracy for different amount of data

A list of state-of-the-art results (for example, the highest performance available) for

MNIST is available online (http://rodrigob.github.io/are_we_there_yet/

build/classification_datasets_results.html). As of March 2019, the best

result has an error rate of 0.21% [2].

Recognizing CIFAR-10 images with deep

learning

The CIFAR-10 dataset contains 60,000 color images of 32×32 pixels in 3 channels,

divided in 10 classes. Each class contains 6,000 images. The training set contains

50,000 images, while the test sets provides 10,000 images. This image taken from

the CIFAR repository (https://www.cs.toronto.edu/~kriz/cifar.html)

shows a few random examples from each of the 10 classes:

[ 122 ]

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!