Machine Learning
Martin Sewell
Department of Computer Science
University College London
April 2007 (last updated January 2009)
1 Introduction
Machine learning is an area of articial intelligence concerned with the study
of computer algorithms that improve automatically through experience. In
practice, this involves creating programs that optimize a performance criterion
through the analysis of data. Machine learning can be viewed as an attempt to
automate `doing science'. For introductory texts, see Langley (1996), Mitchell
(1997), Alpaydin (2004) and Bishop (2006). Mitchell has long been considered
the `bible', but is now slightly dated. Bishop (2006) is the rst textbook on pat-
tern recognition to present the Bayesian viewpoint. For introductory books on
computational learning theory (which emphasizes the `probably approximately
correct' (PAC) model of learning (Valiant 1984)), see Anthony and Biggs (1992)
and Kearns and Vazirani (1994). As the following taxonomy shows, machine
learning algorithms may be categorised in at least six ways.
2 Taxonomy
 Model type
probabilistic Build a full or partial probability model.
non-probabilistic Find a discriminant/regession function directly.
 Type of reasoning
Induction Reasoning from observed training cases to general rules, which
are then applied to the test cases.
Transduction Reasoning from observed, specic (training) cases to spe-
cic (test) cases. Figure 1 (page 2) shows the relationship between
induction and transduction.
 Type of machine learning
1

Figure 1: Two types of inference: induction-deduction and transduction
(Cherkassky and Mulier 1998)
2

Supervised learning The algorithm is rst presented with training data
which consists of examples which include both the inputs and the
desired outputs, thus enabling it to learn a function. The learner
should then be able to generalize from the presented data to unseen
examples. In situations where there is a cost to labeling data, a
method known as active learning may be used, where the learner
chooses which data to label.
Unsupervised learning The algorithm is presented with examples from
the input space only and a model is t to these observations. For
example, a clustering algorithm would be a form of unsupervised
learning.
Reinforcement learning An agent explores an environment and at the
end receives a reward, which may be either positive or negative. In
eect, the agent is told whether he was right or wrong, but is not
told how. Examples include playing a game of chess (you don't know
whether you've won or lost until the very end) or a waitress in a
restaurant (she has to wait for the end of the meal before she nds
out whether or not she receives a tip).
 The manner in which the training data are presented to the
learner
Batch All the data are given to the learner at the start of learning.
On-line The learner receives one example at a time, and gives an esti-
mate of the output, before receiving the correct value. The learner
updates its current hypothesis in response to each new example and
the quality of learning is assessed by the total number of mistakes
made during learning.
 Task
Classication May be binary or multi-class.
Regression Real-valued targets (generalizes classication).
 Classification model type
generative model Denes the joint probability of the data and latent
variables of interest, and therefore explicitly states how the observa-
tions are assumed to have been generated.
discriminative model Focuses only on discriminating one class from
another.
3 Desirable Features
Desirable features of a machine learning algorithm:
3

 Simple solutions are appropriately favoured over complicated ones.
 Powerful enough to learn the solution to a given problem.
 Stable to parameter variations.
 Converges in nite time.
 Scales reasonably with the number of training examples, the number of
input features and the number of test examples.
4

References
ALPAYDIN, Ethem, 2004. Introduction to Machine Learning. Adaptive Com-
putation and Machine Learning. Cambridge, MA: The MIT Press.
ANTHONY, Martin, and Norman BIGGS, 1992. Computational Learning The-
ory. Volume 30 of Cambridge Tracts in Theoretical Computer Science. Cam-
bridge: Cambridge University Press.
BISHOP, Christopher M., 2006. Pattern Recognition and Machine Learning.
Information Science and Statistics. New York: Springer.
CHERKASSKY, Vladimir, and Filip MULIER, 1998. Learning from Data: Con-
cepts, Theory, and Methods. Adaptive and Learning Systems for Signal Pro-
cessing, Communications and Control Series. New York: Wiley.
KEARNS, Michael J., and Umesh V. VAZIRANI, 1994. An Introduction to
Computational Learning Theory. Cambridge, MA: The MIT Press.
LANGLEY, Pat, 1996. Elements of Machine Learning. San Francisco, CA:
Morgan Kaufmann.
MITCHELL, Tom M., 1997. Machine Learning. New York: McGraw-Hill.
VALIANT, L. G., 1984. A Theory of the Learnable. Communications of the
ACM, 27(11), 1134{1142.
5