Deep Neural Networks in Medical Image Classification

Opis

Articial intelligence and machine learning have become buzzwords which have for
years been attracting the attention of researchers and software engineers around
the world. New ideas within the eld of articial intelligence have appeared together
with the development of relevant equipment. Owing to the symbolic approach,
the rst intelligent systems appeared in the mid-20th century. The solutions
based on the symbolic approach were algorithms made up of a set of rules,
Although initially there were high hopes for the symbolic approach (Chollet 2017a),
nowadays its methods are being abandoned in favor of sub-symbolic ones, such as
articial neural networks. However, a complete departure from symbolic methods
seems to be rather unlikely, as on the one hand, they are still eective and being
developed and on the other hand, the use of neural networks is not always fully
justied. Also, the symbolic approach includes such advanced methods as genetic
algorithms and fuzzy inference.
Nowadays, articial neural networks represent the landscape of everyday life.
They are not a new invention. In the history of science, there had been several
instances of their renaissance before they were abandoned en masse in favor of other
approaches. The current wave of renewed interest in neural networks emerged
around 2012 with AlexNet’s (Krizhevsky, Sutskever & Hinton 2012) breakthrough
performance in the ImageNet competition. Nowadays, models composed of at
least several hidden (deep) layers are known as deep neural networks.

[frg. tekstu]