Page 204 - Handbook of Deep Learning in Biomedical Engineering Techniques and Applications

194 Chapter 7 Early detection and diagnosis using deep learning

3. Automatic machine translation
Neural networks are helpful in identifying images containing
visible letters. Upon identification, they can be turned into text,
which can be further translated into an appropriate image; this
procedure is termed as instant visual translation. Automatic
translations into another language are possible by giving sets of
phrases, sentences, and words.
4. Natural language processing
NLP (natural language processing) helps machines under-
stand linguistic nuances and forms suitable responses using DL.
Certain aspects of any language such as tone, sarcasm, semantics,
and expressions are hard to learn, which is what NLP is trying to
achieve. Activities such as language modeling, text classification,
and analysis of tweets come under NLP.
5. Visual recognition
Visual recognition uses DL to sort images into groups based
on various factors such as location, people, dates, faces, events,
and more. Before the development of such technology, sorting
out photos and videos had to be done manually, which are
proved to be a cumbersome task. In today's age, this task is
automatically performed for a large number of photos in every
user's gallery.

1.1.2 Challenges faced by deep learning
Advancements in AI and DL continue to be at the forefront of the
technological world. With such a wide array of applications, DL is
developing at a fast rate with scientists all across the world work-
ing on algorithms. While its importance in today's world cannot
be ignored, DL and its development face several challenges:
1. The requirement of quality data
DL works best when there is a large amount of quality data
available to assist its operations. There is a direct correlation
between increased performance and availability of quality data.
Disingenuous data can lead to wrong predictions and alter
results, which can prove to be extremely damaging in some cases.
Furthermore, in certain organizations, lack of data leads to
hampering in their DL efforts.
2. Expectations
There is an unrealistic expectation from AI that it will replace
human roles. In reality, DL simply enhances productivity by auto-
mating mundane tasks and performing optimization.
3. Ready for production
While more than 80% of enterprises are investing in research
in DL and AI, there needs to be a transition from modeling and