Page 292 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

Chapter 10 Deep neural network in medical image processing 283

Different type of tasks that can be performed using unsuper-
vised learning are as follows:
• Robotics and industrial automation
• Video game agents
• Chatbots
• Text summarization

2.12 Artificial neural network
The development of machine learning algorithms in AI is
dominated by neural networks (NNs). Inspired by biological neu-
rons working in the human brain, the NNs were created to mimic
the working of a human brain. An interconnection of these
neurons induces a large amount of computational power that
leads to solving complex tasks, which are otherwise extremely
difficult to solve using normal computational algorithms. An
interconnection of these neurons induces a large amount of
computational power that leads to solving complex tasks. Warren
McCulloch and Walter Pitts (1943) [47] developed a computa-
tional model based on theoretical model for the NNs. This model
laid the groundwork for two approaches in NN research: one for
study of biological NN of the brain and other is application of arti-
ficial neural network (ANN).
ANNs are computer systems that are based on the neural bio-
logical networks of humans’ and animals’ brains. An ANN
consists of an artificial neurons array of attached units. A signal
can be transmitted from each link between the neurons. The
neuron that is obtained will interpret and then transfer the sig-
nal(s) to linked downstream neurons. Neurons usually have
states between 0 and 1, which are represented by real numbers.
The weights of neurons differ even with increasing learning,
which can increase or decrease the strength of the message to
other neurons.

3. Deep learning
Deep learning is a type of machine learning where complex
data are modeled using a structure, which tends to mimic the hu-
man brain; it uses multiple layers of simple processing units
called neurons much like the neurons in a human brain. It en-
ables these algorithms to have multiple levels of abstraction.
These structures are called ANN. The primary difference between
deep learning algorithms and older learning algorithms is that the
older algorithms tend to reach a plateau after a certain period of