Pattern identification: hierarchical, relations, cycles,

disjoint graph components, and edge crossings

Anibal Montero Leguizamon
[email protected]
Diego Alejandro Babativa Melgarejo
[email protected]
Yeison David Sanchez Legarda
[email protected]
Maestria en ingenieria de sistemas y computación
Data science and visualization
Universidad Nacional de Colombia

Abstract—The purpose of this document is to understand the Known types of genes/patterns. Other examples is with Data
importance and objective of the patter recognition inside machine mining where the application is searching for meaningful
learning environment. Pattern as opposite of a chaos; it is an patterns m the input points in multidimensional space and
entity, vaguely defined, that could be given a name. One of
the main objectives of the pattern recognition is supervised or finally the patterns class will be compact and well-separated
unsupervised classification. The following issues should to have clusters. Rapidly growing and available large servers and
pay attention: a definition of pattern classes, feature extraction power computing enable the huge amount of data and improve
and selection, cluster analysis, and sensing environment. For the methods for data analysis and classification.
instance, face recognition, web search, gathering multimedia
data, require robust pattern recognition techniques. Along the
document the other popular techniques for pattern recognition
going to take account such as graphs components, cycles and
hierarchical relations.

I. I NTRODUCTION
There are many disciplines such as medicine, marketing,
computer vision, artificial intelligence, biology where the the
automatic (machine) recognition, description, classification
and clustering techniques are applied, there is a possible
definition for a ”pattern”. In [1] a patterns could be a human
face, handwritten cursive word or an entity. In this case, the
interest in the pattern recognition area has been growing in
Fig. 1. Examples of Pattern Recognition Applications
the last decades, where the correlation or finding the outliers,
document classification financial forecasting , retrieval of
document databases. Demands on automatic pattern recog-
nition systems are rising enormously due to the availability III. H IERARCHICAL RELATIONS
of large databases and stringent performance requirements.
The process to design of a pattern recognition system is:
data acquisition and pre processing after data representation
and lastly decision making. More ahead, whitin this approach,
The first approach to hierarchical relations is dividing the
we mentions another popular techniques to try separate the
problem in a hierarchical way where classes which are more
data through searching patterns, such as hierarchical relations,
similar to one another and grouped together. According with
cycles, graphs an its components and operations. Finally, some
[2] there are applications where the internal (non-leaf) nodes
algorithms applied to pattern recognition.
of the class hierarchy can be chosen based on data. One thing
II. A BOUT PATTERN R ECOGNITION APPLIED very important is that ”hierarchical classification” can be seen
as a particular type of structured classification problem, where
A very good example of pattern recognition applications
the output of the classification algorithm is defined over a class
is shown in figure 1: in this case the are several problem
taxonomy.
domain. First row the problem domain is Bioinformatics, and
its application could be Sequence analysis. As input pattern The figure 1 illustrates the tree based structure in a hierar-
is DNA/Protein sequence and finally the patterns class is chical relation. The R denotes the root node
 Fig. 2. A example of a tree-based hierarchical class structure Fig. 5. Example of a cycle based on [3]

V. G RAPH C OMPONENTS
A. Hierarchical relations - Differences
Graph is a set of vertices and edges This is a quickly
example of a Graph components. These components are vertex
Hierarchical classification method can be implemented in
or Node: Is the fundamental unit of which graphs are formed
a way that will always classify a leaf node. In hierarchical
and Edges: The ”line” that connect two nodes.
structure exploration:
A. Disjoint Graph Components
• Top-down (or local) classifiers: Here, the system em-
ploys a set of local classifiers As mentions [2] in this In graph theory, “Two graphs are disjoint if they do not
approach is not a full hierarchical classification by itself, have vertices (and therefore edges) in common.” In following
but this method avoid inconsistencies in class prediction example in Figure 6:
at different levels • Graph A has 4 elements, they are 0,1,2 and 3
• Graph B has 2 elements, they are 4 and 5

Fig. 3. Local classifier per node approach (circles represent classes and dashed
squares with round corners represent binary classifiers) Fig. 6. An example of disjoint graph components

These two subgraph dont have vertices in common. Graph

• Big-bang (or global) classifiers: Here, a single classifier theory is awell-studied discipline as are the fields of natural
coping with the entire class hierarchy is used. language processing and information retrieval. More ahead,
We are going to show some applications of these graph
properties.
“Two subgraphs are edge disjoint if they share no edges“
For this first property, like that, two subgraphs are edge disjoint
if they share no edges For the left figure, from S to T, path
S1 to a to b to t and path S2 to c to d to t as follow:

Fig. 4. Big-bang classification approach using a classification algorithm that

learns a global classification model about the whole class hierarchy

IV. C YCLES

A pattern is often denoted in the form of an IF-THEN Fig. 7. An example of Edge disjoint
rule (IF antecedent THEN consequent, where the antecedent
and consequent are logical conjunctions of predicates. Fur- B. Where can we apply Graph components ?
thermore, Knowledge can take on more complex forms than In social media conversations, what type users can be
a simple implication as a causal chain or network by inter- organized in several clusters: In text analysis, we can find
connecting the consequent of one rule to the antecedent of semantic relations, relations within groups and difference
another. According with [3] ”cycles” is a path in a directed between groups.
graph that begins and ends at the same node. Chain of causal Graph theory is awell-studied discipline as are the fields of
dependencies with feedback. Finally, cycles are powerful sets natural language processing and information retrieval. Tradi-
of connected IF-THEN rules. May be detected in more than tionally, these areas of study have been perceived as distinct,
one database instance. As database instances change, new with different algorithms, different applications, and different
patterns are learned that may be cycle participants. potential end-users. However, as recent research work has
shown, these disciplines in fact are intimately connected, with Many such algorithms have been produced [1], all of which
much variety in the way that natural language processing and typicallytake into account one or more aesthetic criteria, with
information retrieval applications find efficient solutions within the assumption that bydoing so, the readabilityof the drawing
graph-theoretical frameworks. In a cohesive text, language is increased. Such aesthetic criteria include, for example,
units – whether they are words, phrases, or entire sentences – minimizing the number of edge crossing. Following are some
are connected through various relationships, which contribute of application of graph components.
to the overall meaning and maintain the cohesive structure • Planar graphs
and discourse unity of the text. Since the early stages of arti- • Topological graph theory
ficial intelligence, associative or semantic networks have been • Discrete geometry
proposed as representations that enable the storage of such • Computer Science
language units and their interconnecting relationships,which
allow for a variety of inference and reasoning processes that R EFERENCES
simulate functionalities of the human mind. The symbolic [1] A. K. Jain, R. P. W. Duin, and Jianchang Mao, “Statistical pattern recog-
structures that emerge from these representations correspond nition: a review,” IEEE Transactions on Pattern Analysis and Machine
Intelligence, vol. 22, no. 1, pp. 4–37, 2000.
naturally to graphs – in which text constituents are represented [2] C. N. Silla and A. A. Freitas, “A survey of hierarchical classification
as vertices and their interconnecting relationships form the across different application domains,” Data Mining and Knowledge
edges in the graph. Discovery, vol. 22, pp. 31–72, 2010.
[3]
Many text-processing applications can be modeled by means
of a graph. These data structures have the capability to
encode naturally the meaning and structure of a cohesive text
and to follow closely the associative or semantic memory
representations.
C. Edge Crossing
How many edges can be used to join all the vertices ?
The number of crossings in a graph drawing is the number
of pairs of edges which cross. For the bottom figure, there isn’t
crossing between all vertices. For the top figure, was needed
6 crossing edges to join all vertex or nodes.
For this task, we could be used 7 crossing edges to join all
nodes, but is not efficient. “Minimizing Edge crossing” is one
of the most known graph aesthetic measures.

Fig. 8. N crossing

Fig. 9. 6 crossings

D. Why are they important ?

AUTOMATIC GRAPH layout algorithms take as input
an abstract graph structure comprising relational information
about objects and the associations between them, and pro-
duce a visual representation of the graph, where objects are
typically represented as circles on a two-dimensional plane,
and the associations represented as lines between the circles.