Page 106 - Code & Click - 7

One especially important turning point around 2010 was developing effective methods for running
Machine Learning algorithms on Graphics Processing Units (GPUs) rather than on the Central
Processing Units (CPUs) that handle most computing workloads. Originally designed for video games
and computer graphics, GPUs are highly parallelized, which means they can perform large numbers
of similar calculations at the same time. For many types of Machine Learning, using GPUs can speed
up the training process by 10-20 times, while reducing computer hardware costs. Access to cloud
storage is also very helpful, since organisations can rapidly access massive computing resources on
demand and limit purchases of computing power to only what they need, when they need it.

Improved Machine Learning Algorithms
The first Machine Learning algorithms are decades old, and some decades-old algorithms remain
incredibly useful. In recent years, however, researchers have discovered many new algorithms that
have greatly sharpened up the field’s cutting-edge. These new algorithms have made Machine
Learning models more flexible, more robust, and more capable of solving different types of problems.

Open-source Code Libraries and Frameworks
For a long time, developing Machine Learning systems required a lot of specific expertise and custom
software development that made it out of reach for most organizations. Now, there are many open-
source code libraries and developer tools that allow organisations to use and build upon the work
of external communities. As a result, no team or organization has to start from scratch, and many
parts that used to require highly specialised expertise have been largely automated. The difficulty of
developing an AI model has reduced to the point where even non-experts and beginners can create
useful AI tools.

TYPES OF MACHINE LEARNING
Like Artificial Intelligence, Machine Learning is also an umbrella term, and there are four different
broad families of Machine Learning algorithms.
The four categories differ based on what types of
data their algorithms can work with. The important
distinction is not whether the data is audio, images,
text, or numbers. Rather, it is whether or not the
training data is labeled or unlabeled and how the
system receives its data inputs. The image given
illustrates labeled and unlabeled training data for a
classifier of images of cats and dogs.
Depending upon whether or not data is labeled, a different family of algorithms is applied. The four
main types of Machine Learning are Supervised Learning, Unsupervised Learning, Semi-supervised
Learning, and Reinforcment Learning.

Supervised Learning
‘Supervised’ means that, before the algorithm processes
the training data, a supervisor (which may be a human,
group of humans, or a different software system) has
accurately labeled each of the data inputs with its
correct associated output.

104