Along with mobile payments, augmented reality, advanced
biometrics, and enhanced online meetings, among other subjects, machine
learning is one of the many trends that can be observed in the world of digital
information technology.
اضافة اعلان
What is machine learning?
It is a way to make machines (i.e. computers) learn to accomplish
specific tasks, somewhat like human beings learn and acquire skills, without
programming it in a traditional way. To date, the latter is still the
predominant method used in the industry, but machine learning is well on its
way to becoming much more common.
It is interesting to note, that the concept was initially introduced
in the early 1950s by Canadian psychologist Donald Hebb as a theory based
partly on “brain cells interaction”. Back then, very little could be achieved or
implemented practically speaking, given the limited memory, storage capacity,
and processing power that computers had in those days.
To put it simply, machine learning takes over when traditional
programming is too long, complex, or just not powerful enough. For years,
specialists and researchers have tried to improve the efficiency of computer programming,
and indeed modern programming languages are significantly faster, more powerful,
and easier to apply than the early languages, but they still have their limits.
Machine learning, on the other hand opens new horizons. Take
language translation for instance, a field where machine learning is being put
to better and better use.
If you program a computer the traditional way to do automated
language translation, say from Arabic to English, the program will have to take
into consideration not only all the vocabulary of the source and the target
languages, but also their grammar, the context of the contents, and a few
linguistic elements that make the task a real mission impossible. To this day,
computerized translation based on pure programming has not been able to provide
acceptable results that would match human translation.
With machine learning, the approach is different. You keep
feeding the computer (i.e. the machine) with the largest possible amount of finished
human translations, in the largest possible variety of domains and contexts, so
that the machine can build itself an idea of how to translate, until it reaches
a level where it is able to translate new content by itself. The “already finished”
translations that are fed into the machine are referred to as “training data”.
In the end, the machine would not have been programmed to translate, but would
have learnt to do so.
Naturally, the larger the amount of training data you feed
into a system, the larger the variety of domains and contexts used, and the
closer the machine will get to doing a good job. Because of that, machine
learning can take a very long time before tangible results are seen.
There are fields of application where programming will still
be the proper approach: accounting, scientific calculation, straightforward
processing of large databases such as population censuses, and the like. On the
other hand, there are countless other kinds of applications where only machine
learning will do — computerized language translation is one of them.
Others include learning how to play chess or tennis (virtually).
There is already a virtual tennis game on PlayStation 3 where you can play with
or against a virtual Roger Federer or Rafael Nadal. The models were trained,
with machine learning, to play like champions after being fed with training
data from available matches that covered all aspects of their behavior: style,
response, body action, height, weight, and such.
It is expected that, in a few years, machine learning will enable
computers to come up with automated language translations that will be very
close to what a human being can provide. After all, in real life, human beings
do “learn” too. Machine learning is now a solid reality, and no longer a
fantasy.
Read more Opinion & Analysis
.jpg)
.jpg)
