An introduction to heuristic algorithms

Natallia Kokash

Department of Informatics and Telecommunications

University of Trento, Italy
email: [email protected]

Abstract. Nowadays computers are used to solve incredibly complex

problems. But in order to manage with a problem we should develop an
algorithm. Sometimes the human brain is not able to accomplish this
task. Moreover, exact algorithms might need centuries to manage with
formidable challenges. In such cases heuristic algorithms that find ap-
proximate solutions but have acceptable time and space complexity play
indispensable role. In this paper heuristics, their areas of application and
the basic underlying ideas are surveyed. We also describe in more detail
some modern heuristic techniques, namely Evolutionary Algorithms and
Support Vector Machines.

1 Introduction
The most important among a variety of topics that relate to computation are
algorithm validation, complexity estimation and optimization. Wide part of the-
oretical computer science deals with these tasks. Complexity of tasks in general
is examined studying the most relevant computational resources like execution
time and space. The ranging of problems that are solvable with a given limited
amount of time and space into well-defined classes is a very intricate task, but
it can help incredibly to save time and money spent on the algorithms design.
A vast collection of papers were dedicated to algorithm development. A short
historical overview of the fundamental issues in theory of computation can be
found in [1]. We do not discuss precise definition of algorithm and complexity.
The interested reader can apply for the information to one of the fundamental
books on theory of algorithms, e.g. [2], [3].
Modern problems tend to be very intricate and relate to analysis of large
data sets. Even if an exact algorithm can be developed its time or space com-
plexity may turn out unacceptable. But in reality it is often sufficient to find an
approximate or partial solution. Such admission extends the set of techniques
to cope with the problem. We discuss heuristic algorithms which suggest some
approximations to the solution of optimization problems. In such problems the
objective is to find the optimal of all possible solutions, that is one that mini-
mizes or maximizes an objective function. The objective function is a function
used to evaluate a quality of the generated solution. Many real-world issues are
easily stated as optimization problems. The collection of all possible solutions for
a given problem can be regarded as a search space, and optimization algorithms,
in their turn, are often referred to as search algorithms.
 Approximate algorithms entail the interesting issue of quality estimation of
the solutions they find. Taking into account that normally the optimal solution
is unknown, this problem can be a real challenge involving strong mathematical
analysis. In connection with the quality issue the goal of the heuristic algorithm
is to find as good solution as possible for all instances of the problem. There are
general heuristic strategies that are successfully applied to manifold problems.
The paper is organized as follows. Section 2 presents some essential informa-
tion about algorithms and computational complexity. Section 3 describes preva-
lent heuristic techniques, Support Vector Machines and Evolutionary Algorithms
are presented. Some intractable problems that could help to understand deeper
importance of heuristics are also mentioned. Finally, the last section is devoted
to the conclusion.

2 Algorithms and complexity

It is difficult to imagine the variety of existing computational tasks and the

number of algorithms developed to solve them. Algorithms that either give nearly
the right answer or provide a solution not for all instances of the problem are
called heuristic algorithms. This group includes a plentiful spectrum of methods
based on traditional techniques as well as specific ones. For the beginning we
sum up the main principles of traditional search algorithms.
The simplest of search algorithms is exhaustive search that tries all possible
solutions from a predetermined set and subsequently picks the best one.
Local search is a version of exhaustive search that only focuses on a limited
area of the search space. Local search can be organized in different ways. Pop-
ular hill-climbing techniques belong to this class. Such algorithms consistently
replace the current solution with the best of its neighbors if it is better than
the current. For example, heuristics for the problem of intragroup replication
for multimedia distribution service based on Peer-to-Peer network is based on
hill-climbing strategy [4].
Divide and conquer algorithms try to split a problem into smaller problems
that are easier to solve. Solutions of the small problems must be combinable to
a solution for the original one. This technique is effective but its use is limited
because there is no a great number of problems that can be easily partitioned
and combined in a such way.
Branch-and-bound technique is a critical enumeration of the search space.
It enumerates, but constantly tries to rule out parts of the search space that
cannot contain the best solution.
Dynamic programming is an exhaustive search that avoids re-computation
by storing the solutions of subproblems. The key point for using this technique
is formulating the solution process as a recursion.
A popular method to construct successively space of solutions is greedy
technique, that is based on the evident principle of taking the (local) best choice
at each stage of the algorithm in order to find the global optimum of some
objective function.
 Usually heuristic algorithms are used for problems that cannot be easily
solved. Classes of time complexity are defined to distinguish problems according
to their ”hardness”. Class P consists of all those problems that can be solved
on a deterministic Turing machine in polynomial time from the size of the in-
put. Turing machines are an abstraction that is used to formalize the notion of
algorithm and computational complexity. A comprehensive description of them
can be found in [3]. Class NP consists of all those problems whose solution can
be found in polynomial time on a non-deterministic Turing machine. Since such
a machine does not exist, practically it means that an exponential algorithm
can be written for an NP-problem, nothing is asserted whether a polynomial
algorithm exists or not. A subclass of NP, class NP-complete includes prob-
lems such that a polynomial algorithm for one of them could be transformed
to polynomial algorithms for solving all other NP problems. Finally, the class
NP-hard can be understood as the class of problems that are NP-complete or
harder. NP-hard problems have the same trait as NP-complete problems but
they do not necessary belong to class NP, that is class NP-hard includes also
problems for which no algorithms at all can be provided.
In order to justify application of some heuristic algorithm we prove that the
problem belongs to the classes NP-complete or NP-hard. Most likely there are
no polynomial algorithms to solve such problems, therefore, for sufficiently great
inputs heuristics are developed.

3 Heuristic techniques

Branch-and-bound technique and dynamic programming are quite effective but

their time-complexity often is too high and unacceptable for NP-complete tasks.
Hill-climbing algorithm is effective, but it has a significant drawback called pre-
mature convergence. Since it is ”greedy”, it always finds the nearest local optima
of low quality. The goal of modern heuristics is to overcome this disadvantage.
Simulated annealing algorithm [5], invented in 1983, uses an approach
similar to hill-climbing, but occasionally accepts solutions that are worse than
the current. The probability of such acceptance is decreasing with time.
Tabu search [6] extends the idea to avoid local optima by using memory
structures. The problem of simulated annealing is that after ”jump” the algo-
rithm can simply repeat its own track. Tabu search prohibits the repetition of
moves that have been made recently.
Swarm intelligence [7] was introduced in 1989. It is an artificial intel-
ligence technique, based on the study of collective behavior in decentralized,
self-organized, systems. Two of the most successful types of this approach are
Ant Colony Optimization (ACO) and Particle Swarm Optimization (PSO). In
ACO artificial ants build solutions by moving on the problem graph and chang-
ing it in such a way that future ants can build better solutions. PSO deals with
problems in which a best solution can be represented as a point or surface in
an n-dimensional space. The main advantage of swarm intelligence techniques is
that they are impressively resistant to the local optima problem.
 Evolutionary Algorithms succeed in tackling premature convergence by
considering a number of solutions simultaneously. Later we discuss this group of
algorithms more elaborately.
Neural Networks are inspired by biological neuron systems. They consist of
units, called neurons, and interconnections between them. After special training
on some given data set Neural Networks can make predictions for cases that are
not in the training set. In practice Neural Networks do not always work well
because they suffer greatly from problems of underfitting and overfitting [8].
These problems correlate with the accuracy of prediction. If a network is not
complex enough it may simplify the laws which the data obey. From the other
point of view, if a network is too complex it can take into account the noise that
usually assists at the training data set while inferring the laws. The quality of
prediction after training is deteriorated in both cases. The problem of premature
convergence is also critical for Neural Networks.
Support Vector Machines (SVMs) extend the ideas of Neural Networks.
They successfully overcome premature convergence since convex objective func-
tion is used, therefore, only one optimum exists. Classical divide and conquer
technique gives elegant solution for separable problems. In connection with
SVMs, that provide effective classification, it becomes an extremely powerful
instrument. Later we discuss SVM classification trees, which applications cur-
rently present promising object for research.
Description and comparative analysis of simulated annealing, tabu search,
neural networks and evolutionary algorithms can be found in [9].

3.1 Evolutionary algorithms

Evolutionary algorithms are methods that exploit ideas of biological evolution,
such as reproduction, mutation and recombination, for searching the solution
of an optimization problem. They apply the principle of survival on a set of
potential solutions to produce gradual approximations to the optimum. A new
set of approximations is created by the process of selecting individuals according
to their objective function, which is called fitness for evolutionary algorithms,
and breeding them together using operators inspired from genetic processes. This
process leads to the evolution of populations of individuals that are better suited
to their environment than their ancestors.
The main loop of evolutionary algorithms includes the following steps:
1. Initialize and evaluate the initial population.
2. Perform competitive selection.
3. Apply genetic operators to generate new solutions.
4. Evaluate solutions in the population.
5. Start again from point 2 and repeat until some convergence criteria is satis-
fied.
Sharing the common idea, evolutionary techniques can differ in the details
of implementation and the problems to which they are applied. Genetic pro-
gramming searches for solutions in the form of computer programs. Their fitness
is determined by the ability to solve a computational problem. The only dif-
ference from evolutionary programming is that the latter fixes the structure of
the program and allows their numerical parameters to evolve. Evolution strat-
egy works with vectors of real numbers as representations of solutions, and uses
self-adaptive mutation rates.
The most successful among evolutionary algorithms are Genetic Algorithms
(GAs). They have been investigated by John Holland in 1975 and demonstrate
essential effectiveness. GAs are based on the fact that the role of mutation im-
proves the individual quite seldom and, therefore, they rely mostly on applying
recombination operators. They seek solutions of the problems in the form of
strings of numbers, usually binary.
The prevalent area for applying genetic algorithms are optimization prob-
lems requiring large scale high performance computing resources. For instance,
the problem of effective resource allocation in conjunction with Plane Cover
Multiple-Access (PCMA) scheme has been examined in [10]. Its objective is to
maximize the attainable capacity of packet-switching wireless cellular networks.
The main issue to consider for resource allocation is to minimize the number of
Unit of Bandwidth (UB) that should be allocated. The problem has been proven
to be in the class NP-hard. Authors applied genetic algorithm instead of greedy
search used before. Computer simulation has been performed for the idealized
cellular system with one base station that can support m connections required
one UB per second, constructed in a cluster of B cells. As result, it is stated
that the genetic algorithm can improve the system capacity utilization.
Substantial increase of the consumer demand on computers and mobile phones
made network optimization problems extremely relevant. Being general tech-
nique that can be easily modified under the various conditions, evolutionary
algorithms are widely used in this area.
To give an example let us mention Adaptive Mesh Problem (AMP), that has
a goal to minimize the required number of base stations of cellular network to
cover a region. Like the previous discussed problem, AMP is NP-hard. One of
the evolutionary techniques, called Hybrid Island Evolutionary Strategy (HIES)
has been applied to tackle this problem [11]. It represents evolutionary algorithm
borrowing features from two types of genetic algorithms, namely island model
GA and fine-grained or cellular GA. The original problem has been transformed
into a geometric meshing generation problem and specific genetic operators,
micromutation, macromutation and crossover, have been altering an array of
hexagonal cells. Initially regular honeycomb was transformed to irregular mesh
that fits better to the real-life conditions. The desired result, the reduction of
the number of base stations, has been achieved.
Some other examples of evolutionary algorithms can be found in [12], which
is devoted to machine learning theory problems.

3.2 Support Vector Machines

Statistical learning theory concerns the problem of choosing desired functions
on the basis of empirical data. Its fundamental problem is generalization, which
consists in inferring laws for future observations, given only a finite amount of
data. Support Vector Machines is the most prominent approach among modern
results in this field.
The basic principles of SVMs have been developed by Vapnik in 1995 [13].
Owing to their attractive qualities they immediately gained a wide range of
applications. SVMs diverged from Empirical Risk Minimization (ERM) principle
embodied by conventional neural networks, which minimizes the error on the
training data, and use Structural Risk Minimization (SRM) principle, which
minimizes an upper bound on the expected risk [14]. SVMs support classification
and regression tasks based on the concept of optimal separator.
The classification problem can be stated as a problem of data set separation
into classes by the functions which are induced from available instances. We will
refer to such functions as classifiers.
In a regression task we have to estimate the functional dependence of the
dependent variable y on a set of independent variables x. It is assumed that
the relationship between the independent and dependent variables is given by a
deterministic function f with some additive noise.
Consider the problem of separating the set of vectors belonging to two classes,

{(x1 , y 1 ), ..., (xl , y l )}, x ∈ Rn , y ∈ {−1, 1},

with a hyperplane,
hw, xi + b = 0,
where w, b are parameters, hw, xi denotes inner product (fig. 1).

Fig. 1. Classification problem

The objective is to separate classes by the hyperplane without errors and

maximize the distance between the closest vector to the hyperplane. Such a
hyperplane is called optimal separating hyperplane. According to the results [14],
the optimal separating hyperplane minimizes
1
Φ(w) = kwk2 , (1)
2
under the constraints
y i [hw, xi + b] ≥ 1. (2)
The solution of optimization problem (1), (2) is given by the saddle point of the
Lagrangian functional. Points that have non-zero Lagrange multipliers are called
Support Vectors and they are used to describe the generated classifier. Usually
the support vectors is a small subset of the training data set. This fact provides
the main attractive quality of the SVMs, namely low computational complexity.
In the case when the training data is not linearly separable there are two
approaches. The first is to introduce an additional function associated with mis-
classification. Another one is to use a more complex function to describe the
boundary. The optimization problem in general is posed to minimize the clas-
sification error as well as the bound on the VC dimension of the classifier. The
VC dimension of a set of functions is p if exists a set of p points, such that these
points can be separated in all 2p possible configurations using this functions,
and there is no set of q points, q > p, satisfying this property.
Whole complex and vast theory of SVMs cannot be covered in this article.
A lot of methods based on the ideas of SVMs have been developed in the last
years. Among them is SVM classification tree algorithm, successfully applied in
text and image classification.
A classification tree consists of internal and external nodes connected by
branches. Each internal node performs a split function that partitions a training
data into two disjoint subsets and each external node contains a label indicat-
ing the predicted class of a given feature vector. Recently this method has been
applied to deal with the classification complexity of membership authentication
problem [15], a typical issue in digital security schemes. Its objective is to distin-
guish membership class M from the nonmembership class G − M in the human
group G. An SVM classifier is trained using two partitioned subgroups, and,
finally, the trained SVM tree is used for identifying the membership of an un-
known person. Experimental results have shown that the proposed method has
a better performance and robustness than previous approaches.

4 Conclusion
This paper has presented an overview of heuristics, that are approximate tech-
niques to solve optimization problems. Usually heuristic algorithms are devel-
oped to have low time complexity and applied to the complex problems. We
briefly defined basic traditional and modern heuristic strategies. Evolutionary al-
gorithms and Support Vector Machines were considered more comprehensively.
Due to their eminent characteristics they gained a great popularity. Recently
appeared research results confirm the fact that their applications can be signifi-
cantly enlarged in the future.
The current paper does not pretend to be complete. It would be interest-
ing to carry out a more profound survey of heuristics, compare implementation
complexity and accuracy of the different approximate algorithms. But this task
cannot be easily accomplished because of the enormous bulk of information. We
even did not touch upon such a prominent area for heuristic algorithms as plan-
ning and scheduling theory. But we hope that our work makes clear the extreme
importance of heuristics in modern computer science.

References
1. S. A. Cook. ”An overview of computational complexity”, in Communication of the
ACM, vol. 26, no. 6, June 1983, pp. 401–408.
2. T. Cormen, Ch. Leiserson, R. Rivest. Introduction to algorithms. MIT Press, 1989.
3. M. R. Garey, D. S. Johnson. Computers and Intractability. Freeman&Co, 1979.
4. Z. Xiang, Q. Zhang, W. Zhu, Z. Zhang, Y. Q. Zhang. ”Peer-to-Peer Based Mul-
timedia Distribution Service”, in IEEE Transactions on Multimedia, vol. 6, no. 2,
Apr. 2004, pp. 343–355.
5. M. E. Aydin, T. C. Fogarty. ”A Distributed Evolutionary Simulated Annealing
Algorithm for Combinatorial Optimization Problems”, in Journal of Heuristics,
vol. 24, no. 10, Mar. 2004, pp. 269–292.
6. R. Battiti. ”Reactive search: towards self-tuning heuristics”, in Modern heuristic
search methods. Wiley&Sons, 1996, pp. 61-83.
7. R. Eberhart, Y. Shi, and J. Kennedy. Swarm intelligence. Morgan Kaufmann, 2001.
8. B. Kröse, P. Smagt. An introduction to Neural Networks. University of Amsterdam,
Nov. 1996.
9. D. Karaboga, D. Pham. Intelligent Optimisation Techniques: Genetic Algorithms,
Tabu Search, Simulated Annealing and Neural Networks. Springer Verlag, 2000.
10. X. Wu, B. S. Sharif, O. R. Hinton. ”An Improved Resource Allocation Scheme for
Plane Cover Multiple Access Using Genetic Algorithm”, in IEEE Transactions on
Evolutionary Computation, vol. 9, no. 1, Feb. 2005, pp.74–80.
11. J.C. Crput, A. Koukam, T. Lissajoux, A. Caminada. ”Automatic Mesh Genera-
tion for Mobile Network Dimensioning Using Evolutionary Approach”, in IEEE
Transactions on Evolutionary Computation, vol. 9, no. 1, Feb. 2005, pp. 18–30.
12. F. Divina, E. Marchiori. ”Handling Continuous Attributes in an Evolutionary In-
ductive Learner”, in IEEE Transactions on Evolutionary Computation, vol. 9, no.
1, Feb. 2005, pp. 31–43.
13. V. N. Vapnik. The Nature of Statistical Learning Theory. Springer Verlag, 1995.
14. S. Gunn. ”Support Vector Machines for Classification and Regression”, Technical
Report, May 1998, http://www.ecs.soton.ac.uk/∼srg/publications/pdf/SVM.pdf.
15. S. Pang, D. Kim, S. Y. Bang. ”Face Membership Authentication Using SVM Clas-
sification Tree Generated by Membership-Based LLE Data Partition”, in IEEE
Transactions on Neural Networks, vol. 16, no 2, Mar. 2005, pp. 436–446.