Page 69 - Data Science Algorithms in a Week

Machine Learning Applied to Autonomous Vehicles 53

Many authors have described different taxonomies about learning processes which
only include the basic learner and teacher problem. However, Camastra & Vinciarelli
(2007) provided a more focused definition based on the application of audio, images and
video analysis to machine learning. They identify four different learning types: rote
learning, learning from instruction, learning by analogy, and learning from examples,
which are briefly explained below.

 Rote Learning: This type consists of directly implanting new knowledge in the
learner. This method includes (1) Learning processes using programs and
instructions implemented by external entities, and (2) Learning processes using
memorization of a given data with no inferences drawn from the incoming
information.
 Learning from instruction: This learning consists of a learner acquiring
knowledge from the instructor and/or other source and transforming it into
internal representations. The new information is integrated with prior knowledge
for effective use. One of the objectives is to keep the knowledge in a way that
incrementally increases the learner’s actual knowledge (Camastra & Vinciarelli,
2007).
 Learning by analogy: This type of learning consists of acquiring new facts or
skills based on “past situations that bear strong similarity to the present problem
at different levels of abstraction" (Carbonell, 2015). Learning by analogy
requires more inferencing by the learner than rote learning and learning from
instruction. Carbonell (2015) gives a good definition: “A fact or skill analogous
in relevant parameters must be retrieved from memory. Then, the retrieved
knowledge must be transformed, applied to the new situation, and stored for
future use."
 Learning from examples: This can simply be called learning: if given a set of
concept’s examples, the learner builds a general concept representation based on
the examples. The learning problem is described as the search for a general rule
that could explain the examples even if only a limited size of examples is given.
Learning techniques can be grouped into four main types: supervised learning,
unsupervised learning, reinforcement learning, and semi-supervised learning.
 Supervised Learning: the learning process is based on examples with inputs
and desired outputs, given by a “teacher”. The data is a sample of input-
output patterns. The goal is to learn a general rule about how the output can
be generated, based on the given input. Some common examples are
predictions of stock market indexes and recognition of handwritten digits and
letters. The training set is a sample of input-output pairs, the task of learning
problem is to find a deterministic function that maps an input to the
respective output to predict future input-output observations and therefore