Browse

Nonparametric probabilistic models for learning spatiotemporal patterns of stream data
스트림 데이터에서 시공간적 패턴 학습을 위한 비모수적 확률 모델

DC Field Value Language
dc.contributor.advisor장병탁-
dc.contributor.author석호식-
dc.date.accessioned2017-07-13T08:58:17Z-
dc.date.available2017-07-13T08:58:17Z-
dc.date.issued2012-08-
dc.identifier.other000000002916-
dc.identifier.urihttps://hdl.handle.net/10371/119987-
dc.description학위논문 (박사)-- 서울대학교 대학원 : 컴퓨터공학과, 2012. 8. 장병탁.-
dc.description.abstractThe complexity in temporal domains requires introduction of the unobservable inherent dependencies. However, this task is very challenging due to difficulties such as hierarchical characteristics of the domain and abrupt changes in modalities. This thesis addresses the problem of extracting inherent dependencies in temporal domains. We develop algorithms for modeling and inference: inference with changing underlying distributions, temporal sequence learning with summarized problem space, and inference in a multichannel sequential environment.
There exist domains where it is not possible to select a stable underlying distribution due to environmental causes. We develop an inference algorithm based on the feature relevance network in order to tackle the distribution change. The idea of feature relevance network is inspired by the existence of stable relations between features irrespective of environmental change. The proposed algorithm is verified on an indoor location estimation task. We show that the non-parametric approach making no assumption on feature distributions is capable of discovering hidden relations among features.
The most obvious instances of temporal domain are dynamic streams such as TV drama. In order to estimate hidden dependencies in an episode of TV drama and represent a whole episode in a more succinct form, this thesis introduces a temporal learning scheme based on a set of particles. Instead of assuming a prior distribution, particles capture prominent characteristics of a given stream in a collaborative manner. The proposed method evolves particles through two-stage learning. At the first stage, a segment (scene) is estimated using evolutionary particle filtering (PF). At the second stage, a transitional probability matrix representing dependencies between estimated segments is computed and stored. We demonstrate performance by comparing to human-evaluated ground truth and regenerating images of expected sequences succeeding a given seed image.
For inference in a multichannel sequential environment, the sequential hierarchical Dirichlet process (sHDP) is developed. For seamless processing of multichannel data, sHDP divides a multichannel stream into sub-channel of single modality and builds a latent model for a sub-channel. Changes in sub-channels are linked though a dynamic channel merging scheme. The proposed method is applied for semantic segmentation in real TV drama episodes and the performance is compared to human evaluated ground-truth.
By incorporating inherent dependencies, we present successful algorithms to deal with applications in temporal domains. The three areas of this thesis are promising realizations of inherent dependencies learning, and the algorithms presented here inform the range of possibility of applicability in temporal domains.
-
dc.description.tableofcontentsAbstract
1 Introduction
1.1 Background and Motivation
1.2 Problems to be Tackled
1.3 Our Approach and Its Contributions
1.4 Thesis Organization
2 Probabilistic Models
2.1 Particle Filtering
2.1.1 Particle Filters
2.1.2 Evolutionary Particle Filtering
2.2 Nonparametric Bayesian Framework
2.2.1 Dirichlet Process
2.2.2 Hierarchical Dirichlet Process
2.3 General Framework
3 Inference with Changing Underlying Distributions
3.1 Nonparametric Approach for Changing Distributions
3.2 Related Works
3.3 Feature Relevance Network Learning
3.3.1 Indoor Location Estimation Problem
3.3.2 The Proposed Method
3.4 Experimental Results
3.4.1 Quality of Candidate Recommendation
3.4.2 Prediction Performance and Comparisons
3.5 Discussion and summarization
4 Temporal Stream Learning with Evolutionary Particle Filtering
4.1 Collaborative Particles and Temporal Stream Analysis
4.2 Related Works
4.3 Evolutionary Particle Filtering and Dynamical Sequence Modelling
4.3.1 Representation
4.3.2 Evolutionary Particle Filtering
4.3.3 Sequential Dependency Learning and Volatility Measure
4.3.4 Image Regeneration
4.4 Experimental Results
4.4.1 Data and Human Evaluations
4.4.2 Segmentation and Dependency Learning
4.4.3 Comparison with Other Method
4.5 Discussion and Summarization
5 Multiple Stream Learning
5.1 Multichannel based Approach
5.2 Related Works
5.2.1 Approaches for Temporal Stream Analysis
5.2.2 Hierarchical Dirichlet Process
5.2.3 Sticky HDP-HMM and Dynamic HDP
5.2.4 Speaker Recognition
5.3 Semantic Segmentation Scheme
5.3.1 Sequential HDP
5.3.2 Posterior sampling and story change estimation
5.3.3 Speaker Recognition
5.3.4 Dynamic Channel Merging
5.4 Experimental Results
5.4.1 Data and Representation
5.4.2 Story Change Estimation Results
5.4.3 Comparison with Other Method
5.5 Discussion and Summarization
6 Concluding Remarks
6.1 Summary of Methods and Contributions
6.2 Suggestions for Future Research
-
dc.formatapplication/pdf-
dc.format.extent3010891 bytes-
dc.format.mediumapplication/pdf-
dc.language.isoen-
dc.publisher서울대학교 대학원-
dc.subjectProbabilistic Model-
dc.subjectNonparametric-
dc.subjectSpatiotemporal Patterns-
dc.subjectEvolutionary Particle Filtering-
dc.subjectStream Analysis-
dc.subject.ddc621-
dc.titleNonparametric probabilistic models for learning spatiotemporal patterns of stream data-
dc.title.alternative스트림 데이터에서 시공간적 패턴 학습을 위한 비모수적 확률 모델-
dc.typeThesis-
dc.contributor.AlternativeAuthorHo-Sik Seok-
dc.description.degreeDoctor-
dc.citation.pages135-
dc.contributor.affiliation공과대학 컴퓨터공학과-
dc.date.awarded2012-08-
Appears in Collections:
College of Engineering/Engineering Practice School (공과대학/대학원)Dept. of Computer Science and Engineering (컴퓨터공학부)Theses (Ph.D. / Sc.D._컴퓨터공학부)
Files in This Item:
  • mendeley

Items in S-Space are protected by copyright, with all rights reserved, unless otherwise indicated.

Browse