3
Examples for ML could be a system that could predict whether
a student will fail or pass in a test by learning from the historical test
results and student attributes. Here, the system is not encoded with a
comprehensive list of all possible rules that can decide whether a student
will pass or fail; instead, the system learns on its own based on the patterns
it learned from the historical data.
So, where does DL stand within this context? It happens that while ML
works very well for a variety of problems, it fails to excel in some specific
cases that seem to be very easy for humans: say, classifying an image as a
cat or dog, distinguishing an audio clip as of a male or female voice, and
so on. ML performs poorly with image and other unstructured data types.
Upon researching the reasons for this poor performance, an inspiration
led to the idea of mimicking the human brain’s biological process, which
is composed of billions of neurons connected and orchestrated to adapt
to learning new things. On a parallel track, neural networks had already
been a research topic for several years, but only limited progress had been
made due to the computational and data limitations at the time. When
researchers reached the cusp of ML and neural networks, there came the
field of DL, which was framed by developing deep neural networks (DNNs),
that is, improvised neural networks with many more layers. DL excelled at
the new frontiers where ML was falling behind. In due course, additional
research and experimentation led to the understanding of where we could
leverage DL for all ML tasks and expect better performance, provided there
was surplus data availability. DL, therefore, became a ubiquitous field
to solve predictive problems rather than just being confined to areas of
computer vision, speech, and so on.
Today, we can leverage DL for almost all use cases that were earlier
solved using ML and expect to outperform our previous achievements,
provided that there is a surplus of data. This realization has led to
distinguishing the order of the fields based on data. A new rule of thumb
was established: ML would not be able to improve performance with
increased training data after a certain threshold, whereas DL was able to
Chapter 1 an IntroduCtIon todeep LearnIng andKeras