Page 94 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

82 Chapter 3 Application, algorithm, tools directly related to deep learning

Due to the specific structures of RBMs, visible and hidden
units are conditionally independent. Using this property, it can
be written as Eq.3.7 [20]
Y
pðhjyÞ¼ pðh i jyÞ
i
Y (3.7)
pðyjhÞ¼ pðy i jhÞ
i

4. Applications of deep learning
1. Virtual assistants
The most popular application of DL is virtual assistant, which
may range from Alexa to Siri to Google Assistant. Interaction with
these assistants helps to gain knowledge more about your voice
and accent [21]. Virtual assistants use DL to know the subjects
ranging from your preferences to your most visited spots or favor-
ite songs. They can understand your commands by evaluating
most common natural human language to execute them. Another
capability that virtual assistants are endowed with is to perform
translation of speech to text, make notes for you, and create ap-
pointments. Virtual assistants can do everything from running er-
rands to autoresponding specific calls to coordinate tasks among
team members [22].
2. Entertainment (VEVO, Netflix, Film Making, Sports Highlights,
etc.)
Wimbledon 2018 used IBM Watson to analyze entire player
emotions and expressions via thousands of hours of footage to
autogenerate highlights for whole telecast to save their effort
and cost. They were able to factor in audience response and
match or game player to come up with a more accurate model.
VEVO has been using DL to make the next generation of meta
data services for not only personalized experiences for its users
and subscribers, artists, and companies. They can record labels
to generate their insights based on overall performance and
popularity [23].
3. Visual recognition
Images can be sorted based on geographical locations
detected in photographs, faces, a combination of people, or ac-
cording to dating of events. Searching for a specific photo from
a library that is a data set as large as Google's picture library
needs state-of-the-art visual recognition systems consisting of
several layers from basic layer to advance layer to recognize ele-
ments [24].