Pareto-Optimality of Cognitively Preferred Polygonal Hulls for Dot Patterns Antony Galton University of Exeter UK.

Slides:



Advertisements
Similar presentations
Evaluating Classifiers
Advertisements

What is the Region Occupied by a Set of Points? Antony Galton University of Exeter, UK Matt Duckham University of Melbourne, Australia.
2003 International Congress of Refrigeration, Washington, D.C., August 17-22, 2003 Application of Multi-objective Optimization in Food Refrigeration Processes.
MAE 552 Heuristic Optimization Instructor: John Eddy Lecture #35 4/26/02 Multi-Objective Optimization.
QUANTITATIVE DATA ANALYSIS
Advisor: Yeong-Sung Lin Presented by Chi-Hsiang Chan 2011/5/23 1.
A TABU SEARCH APPROACH TO POLYGONAL APPROXIMATION OF DIGITAL CURVES.
1 Enviromatics Decision support systems Decision support systems Вонр. проф. д-р Александар Маркоски Технички факултет – Битола 2008 год.
CAD’12, CanadaDepartment of Engineering Design, IIT Madras P. Jiju and M. Ramanathan Department of Engineering Design Indian Institute of Technology Madras.
Quality Indicators (Binary ε-Indicator) Santosh Tiwari.
Chapter 2 Summarizing and Graphing Data
Link Reconstruction from Partial Information Gong Xiaofeng, Li Kun & C. H. Lai
Bayesian parameter estimation in cosmology with Population Monte Carlo By Darell Moodley (UKZN) Supervisor: Prof. K Moodley (UKZN) SKA Postgraduate conference,
Part 6: Graphics Output Primitives (4) 1.  Another useful construct,besides points, straight line segments, and curves for describing components of a.
Statistical Decision Making. Almost all problems in statistics can be formulated as a problem of making a decision. That is given some data observed from.
Basic Statistics for Engineers. Collection, presentation, interpretation and decision making. Prof. Dudley S. Finch.
Research Process Parts of the research study Parts of the research study Aim: purpose of the study Aim: purpose of the study Target population: group whose.
Multi-objective Optimization
Hypothesis Testing Introduction to Statistics Chapter 8 Feb 24-26, 2009 Classes #12-13.
CS6234 Advanced Algorithms - Convex hull. Terminologies – ◦ Convex hull of a set Q of points is the smallest convex polygon P for which each point in.
Evolutionary Computing Chapter 12. / 26 Chapter 12: Multiobjective Evolutionary Algorithms Multiobjective optimisation problems (MOP) -Pareto optimality.
A Multiobjective Evolutionary Algorithm Using Gaussian Process based Inverse Modeling Ran Cheng 1, Yaochu Jin 1, Kaname Narukawa 2 and Bernhard Sendhof.
June 23, Variational tetrahedral meshing of mechanical models for FEA Matthijs Sypkens Smit Willem F. Bronsvoort CAD ’08 Conference, Orlando, Florida.
Statistical Decision Making. Almost all problems in statistics can be formulated as a problem of making a decision. That is given some data observed from.
Class Six Turn In: Chapter 15: 30, 32, 38, 44, 48, 50 Chapter 17: 28, 38, 44 For Class Seven: Chapter 18: 32, 34, 36 Chapter 19: 26, 34, 44 Quiz 3 Read.
Slide 1 Copyright © 2004 Pearson Education, Inc.  Descriptive Statistics summarize or describe the important characteristics of a known set of population.
CLARANS: A Method for Clustering Objects for Spatial Data Mining IEEE Transactions on Knowledge and Data Enginerring, 2002 Raymond T. Ng et al. 22 MAR.
Estimating standard error using bootstrap
Sampling and Sampling Distribution
Where Are You? Children Adults.
Lecture 1.31 Criteria for optimal reception of radio signals.
Exploring Data: Summary Statistics and Visualizations
Chapter 2: The Research Enterprise in Psychology
INTRODUCTION TO STATISTICS
Image Representation and Description – Representation Schemes
Chapter 2 Summarizing and Graphing Data
L-Dominance: An Approximate-Domination Mechanism
March 9th, 2015 Maxime Lapalme Nazim Ould-Brahim
Lesson 6 Normal and Skewed Distribution Type one and Type two errors.
A paper on Join Synopses for Approximate Query Answering
Statistical Data Analysis
Lesson 6 Normal and Skewed Distribution Type one and Type two errors.
Subject Name: File Structures
Constructing Objects in Computer Graphics By Andries van Dam©
Introduction to Summary Statistics
Introduction to Summary Statistics
Chapter 12 Using Descriptive Analysis, Performing
Numerical Descriptive Measures
Introduction to Summary Statistics
Introduction to Summary Statistics
Introduction to Summary Statistics
Maths Information Evening 1 March 2017
Heuristic Optimization Methods Pareto Multiobjective Optimization
Outline H. Murase, and S. K. Nayar, “Visual learning and recognition of 3-D objects from appearance,” International Journal of Computer Vision, vol. 14,
Introduction to Summary Statistics
Chen-Yu Lee, Jia-Fong Yeh, and Tsung-Che Chiang
Introduction to Summary Statistics
Statistical Data Analysis
Measures of Position Section 3.3.
Introduction to Summary Statistics
Hamilton Paths and Hamilton Circuits
MOEA Testing and Analysis
DESIGN OF EXPERIMENT (DOE)
Introduction to Summary Statistics
Objectives 6.1 Estimating with confidence Statistical confidence
Objectives 6.1 Estimating with confidence Statistical confidence
Essentials of Statistics 4th Edition
Alan Girling University of Birmingham, UK
COMMON MISTAKES!.
Presentation transcript:

Pareto-Optimality of Cognitively Preferred Polygonal Hulls for Dot Patterns Antony Galton University of Exeter UK

The Problem To characterise the region occupied by a set of discrete point-like elements. Call the point-like elements ‘dots’, and the region their ‘footprint’. The footprint is a ‘higher-level’ entity: it is not the location of the dots as individuals, but of the aggregate or collective which has those dots as members.

Map Generalisation

Region Approximation

Location of Collectives

What’s wrong with the Convex Hull? The convex hull of a set of dots is the smallest convex region which contains all of the dots. It has well-known mathematical and computational properties. BUT it does not always give a highly representative footprint.

A ‘C’-shape and an ‘S’-shape

Their Convex Hulls

What we actually see

Some existing work Edelsbrunner et al (1983) Chaudhuri et al (1997) Garai and Chaudhuri (1999) Melkemi and Djebali (2000) Alani et al (2001) Arampatzis (2006) Galton and Duckham (2006) Moreira and Santos (2007) Duckham et al (2008?)

What is missing? A typical paper in this area –Proposes an algorithm for generating footprints for dot patterns –Explores its mathematical and computational characteristics –Examines its behaviour when applied to various dot patterns. What is missing is a principled way of evaluating that behaviour. ‘The shape produced by the algorithm is a good approximation to the perceived shape of the dot pattern’.

But what is ‘the perceived shape’ of the dot pattern? There is no unique solution. It is highly subjective. It is influenced by both the actual geometry of the dots and a multitude of subtle cognitive factors. Nobody seems to have investigated this.

Polygonal Hulls We shall restrict the investigation to shapes having the following properties: –It is a polygon whose vertices are members of the dot pattern –Any member of the dot pattern which is not a vertex lies in the interior of the polygon –The boundary of the polygon is a Jordan curve. A shape of this kind will be called a polygonal hull.

A dot pattern

Its Convex Hull

A very unlikely footprint

A more ‘reasonable’ footprint

What makes a good footprint? The convex hull can include large areas devoid of dots (e.g., perceived concavities) Of all the polygonal hulls, the convex hull has maximum area and minimum perimeter. The very jagged hull reduces the area but has a much longer perimeter. The ‘reasonable’ hull achieves a compromise between reducing the area and increasing the perimeter.

Area vs Perimeter AreaPerimeter Convex hull ‘Reasonable’ hull Jagged hull

Conflicting objectives A cognitively acceptable outline should –Not contain too much empty space –Not be too long and sinuous. To produce an optimal outline we should seek to simultaneously minimise both the area and the perimeter. But these are conflicting objectives, since the perimeter can only be minimised by maximising the area (convex hull).

Multi-objective Optimisation A polygonal hull with area A 1 and perimeter P 1 dominates a hull with area A 2 and perimeter P 2 so long as either A 1 < A 2 & P 1 < P 2 or A 1 < A 2 & P 1 < P 2. In seeking to minimise both area and perimeter we are looking for non- dominated hulls.

Pareto optimisation The non-dominated hulls form the Pareto set. When plotted in area-perimeter space (‘objective space’) these hulls lie along a line called the Pareto front. The Pareto front is the ‘south-western’ frontier of the set of points corresponding to all the hulls for a given dot pattern.

Example

HYPOTHESIS Our hypothesis is The points in area-perimeter space corresponding to polygonal hulls which best capture a perceived shape of a dot pattern lie on or close to the Pareto front.

Pilot Study A small pilot study was carried out to gain an initial estimation of the plausibility of the hypothesis. 8 dot patterns were presented to 13 subjects, who were asked to draw a polygonal outline which best captures the shape formed by each pattern of dots.

Dot Patterns Used in Pilot Study

Area-perimeter plots

Evaluating the Results For each outline drawn, the relative domination RD was computed. RD is the ratio of the number of hulls which dominate it to the maximum number of hulls which dominate any one hull for that dot pattern. By definition, 0 < RD < 1 Our hypothesis predicts that subjects should draw hulls with RD close to 0.

Results Pattern (# dots) HullsPareto Distinct responses Pareto responses Mean RD 1 (12) (12) (11) (12) (13) (11) (11) (11)

Pareto fronts, with selections

Summary of results 57 out of the 104 responses were Pareto optimal. The highest individual value for RD was The mean value for RD over all 104 responses was Therefore, on average, subjects hit the Pareto front with an error of 0.18%. A chi-squared test indicates statistical significance at the 0.1% level (in fact much better than this). The hypothesis is strongly supported by the results of the pilot study.

What Next? Many possible variations to explore: –Choice of dot patterns –Choice of experimental procedure –Application context –Other objective criteria –Evaluation of algorithms –Algorithm design –Extension to three dimensions

The immediate goal … … is to find ways of handling larger dot patterns. Computing the full set of polygonal hulls is computationally expensive, especially when a ‘brute force’ algorithm is used. Two plans of attack: –Look for a more efficient algorithm for computing the full hull-set –Estimate the distribution of the hulls in objective space by some form of sampling, e.g., using an evolutionary algorithm to home in on the Pareto front.

THANK YOU FOR LISTENING ANY QUESTIONS ?