Similarity-Based Pattern Recognition: First International Workshop, SIMBAD 2011, Venice, Italy, September 28-30, 2011, Proceedings (Lecture Notes in Computer Science, 7005)

✍ Scribed by Marcello Pelillo (editor), Edwin R. Hancock (editor)

• Publisher: Springer
• Year: 2011
• Tongue: English
• Leaves: 348
• Edition: 2011
• Category: Library

✦ Synopsis

This book constitutes the proceedings of the First International Workshop on Similarity Based Pattern Recognition, SIMBAD 2011, held in Venice, Italy, in September 2011. The 16 full papers and 7 poster papers presented were carefully reviewed and selected from 35 submissions. The contributions are organized in topical sections on dissimilarity characterization and analysis; generative models of similarity data; graph-based and relational models; clustering and dissimilarity data; applications; spectral methods and embedding.

✦ Table of Contents

Cover
Lecture Notes in Computer Science 7005
Similarity-Based
Pattern Recognition
ISBN 9783642244704
Preface
Organization
Table of Contents
Dissimilarity Characterization and Analysis
On the Usefulness of Similarity Based Projection Spaces for Transfer Learning
Introduction
Notations
Learning with Good Similarity Functions
Domain Adaptation
Modifying the Projection Space for Domain Adaptation
A Normalization of a Similarity Function
An Additional Regularization Term for Moving Closer the Two Distributions
Experiments
Synthetic Toy Problem
Image Classification
Conclusion
References
Appendix
Metric Anomaly Detection via Asymmetric Risk Minimization
Introduction
Theoretical Results
Preliminaries
Known Separation Distance
Definition of Risk
Classification Rule
No Explicit Prior on
Experiments
Methodology
Data Sets
Results
Discussion and Future Work
References
One Shot Similarity Metric Learning for Action Recognition
Introduction
Related Work
One-Shot-Similarity Metric Learning (OSSML)
The Free-Scale LDA-Based, Symmetric OSS Score
Deriving the OSSML
Objective Function
Free-Scale LDA-Based OSS Gradient
Application to Action Recognition
ASLAN Data Set
Same/Not-Same Benchmark
Experimental Setup
Experimental Results
Conclusion
References
On a Non-monotonicity Effect of Similarity Measures
Introduction
Construction Principles of Similarity Measures Induced by the Aggregation of Element-Wise Operating Functions
$f$-Divergence Measures
The Monotonicity Property of the Discrepancy Measure
Impact of the Non-monotonicity Effect on Applications
Image Tracking
Stereo Matching
Defect Detection in Textured Surfaces
Conclusion and Future Work
References
Section-Wise Similarities for Clustering and Outlier Detection of Subjective Sequential Data
Introduction
CABINTEC Database
Data Acquisition
Trend Segmentation Algorithm
Similarity Definitions
Mean Level Based Similarity
Angle Based Similarity
Experiments
Clustering
Outlier Detection
Conclusions
References
Generative Models of Similarity Data
Hybrid Generative-Discriminative Nucleus Classification of Renal Cell Carcinoma
Introduction
Background: The Probabilistic Latent Semantic Analysis
The Tissue Microarray (TMA) Pipeline
Tissue Micro Arrays
Image Normalization and Patching
Segmentation
Feature Extraction
Nuclei Classification
Experiments
Discussion
Conclusion
References
Multi-task Regularization of Generative Similarity Models
Introduction
Background on Local Similarity Discriminant Analysis
Multi-task Regularization of Mean Similarity Estimates
Closed-Form Solution
Choice of Task Relatedness A
Related Work in Multi-Task Learning
Benchmark Classification Results
Iraqi Insurgent Rhetoric Analysis
Discussion and Open Questions
References
A Generative Dyadic Aspect Model for Evidence Accumulation Clustering
Introduction
Generative Model for Evidence Accumulation Clustering
Clustering Ensembles and Evidence Accumulation
Generative Model
The Expectation Maximization Algorithm
The E-Step
The M-Step
Summary of the Algorithm and Interpretation of the Estimates
Related Work
Experimental Results and Discussion
Conclusions and Future Work
References
Graph Based and Relational Models
Supervised Learning of Graph Structure
Introduction
Generative Graph Model
Correspondence Sampler
Estimating the Model
Model Selection
Experimental Evaluation
Shock Graphs
3D Shapes
Synthetic Data
Edge-Weighted Graphs
Conclusions
References
An Information Theoretic Approach to Learning Generative Graph Prototypes
Introduction
Probabilistic Framework
Model Coding Using MDL
Encoding Sample Graphs
Encoding the Supergraph Model
Expectation-Maximization
Weighted Code-Length Function
Maximization
Expectation
Experiments
Conclusion
References
Graph Characterization via Backtrackless Paths
Introduction
Backtrackless Walks on Graphs
Kernels for Labeled Graphs
Pattern Vectors for Unlabeled Graphs
Experiments
Synthetic Data
Real-World Dataset
Timing Analysis
Strengths and Weaknesses
Conclusion
References
Impact of the Initialization in Tree-Based Fast Similarity Search Techniques
Introduction
The MDF-Tree
Building an MDT-Tree
The Search Algorithm
Initialization Methods
Random Method
Outlier Method
Median Method
Experiments
Non Balanced Trees
Conclusions
References
Clustering and Dissimilarity Data
Multiple-Instance Learning with Instance Selection via Dominant Sets
Introduction
Background
Instance-Selection Based MIL
Clustering with Dominant Sets
Proposed Method
Notations
Instance Selection with Dominant Sets
Classification
Extension to Multi-class MIL
Computational Complexity
Experimental Results
Benchmark Data Sets
Image Classification
Sensitivity to Labeling Noise
Summary and Future Work
References
Min-sum Clustering of Protein Sequences with Limited Distance Information
Introduction
Preliminaries
Algorithm Overview
Algorithm Analysis
Algorithm Description
Structure of the Clustering Instance
Proof of Theorem 1 and Additional Analysis
Experimental Results
Conclusion
References
Model-Based Clustering of Inhomogeneous Paired Comparison Data
Introduction
Relevant Work
Modeling Paired Comparison Data
Model-Based Clustering
Missings
Model Inference
Selection of Comparisons
Application: Preference Prediction
Experimental Results
Synthetic Data
Political Goals German Data
Conclusion
References
Bag Dissimilarities for Multiple Instance Learning
Introduction
Bag Dissimilarities
Bag Distribution Dissimilarities
Pairwise Instance Dissimilarities
Linear Assignment Dissimilarity
Standard MIL Classifiers
Experiments
Conclusions
References
Mutual Information Criteria for Feature Selection
Introduction
Dominant-Set Clustering Algorithm
Concept of Dominant Set
Dominant-Set Clustering Algorithm
Feature Similarity Measure
Feature Selection Using Dominant-Set Clustering
Computing the Similarity Matrix
Dominant-Set Clustering
Selecting Key Features
Classification
Experiments and Comparisons
Cluster Performance Evaluation Using Different Similarity Measures
Classification Results Using Selected Feature Subset
Conclusions
References
Applications
Combining Data Sources Nonlinearly for Cell Nucleus Classification of Renal Cell Carcinoma
Introduction
Data Set
Tissue Micro Arrays
Image Normalization and Patching
Segmentation
Feature Extraction
Methodology
Linear Multiple Kernel Learning
Nonlinear Multiple Kernel Learning
Experiments
Experimental Methodology
Results
Discussion
Conclusion
References
Supervised Segmentation of Fiber Tracts
Introduction
Methods
Basic Definitions and Notation
Evaluation Criteria
Distances
Classification Algorithms and Feature Space
Dissimilarity Space
Fiber Tract Segmentation
Experiments and Results
Dataset: PBCC2009 Spring Edition
Single Subject Segmentation
Predictions Cross-Subjects
Discussion
References
Exploiting Dissimilarity Representations for Person Re-identification
Introduction
Background
Previous Works on Person Re-identification
The Multiple Component Matching Framework
The Multiple Component Dissimilarity Framework for Person Re-identification
Application of MCD
Conclusions and Future Work
References
Spectral Methods and Embedding
A Study of Embedding Methods under the Evidence Accumulation Framework
Introduction
Embedding Methods
Nonlinear Methods
Linear Methods
Evidence Accumulation: The Co-association Matrix
Dimensionality Reduction in Evidence Accumulation Clustering
Quality Measures
Experimental Results
Data
Experiment 1: Feature Space
Experiment 2: Similarity Space
Discussion
Conclusions
References
A Study on the Influence of Shape in Classifying Small Spectral Data Sets
Introduction
Introduction to Dissimilarity Representation Approach
1D and 2D Dissimilarity Measures for Spectral Data
Experimental Section
Data sets
Experiments and Discussion
Discussion and Conclusions
References
Feature Point Matching Using a Hermitian Property Matrix
Introduction
Complex Laplacian (Hermitian) Matrix
Expectation Maximization
E-Step
M-Step
Experimental Results
Conclusions
References
Author Index