Current Year

PhD
[1]
P. Ernst, “Biomedical Knowledge Base Construction from Text and its Applications in Knowledge-based Systems,” Universität des Saarlandes, Saarbrücken, 2018.
Abstract
While general-purpose Knowledge Bases (KBs) have gone a long way in compiling comprehensive knowledgee about people, events, places, etc., domain-specific KBs, such as on health, are equally important, but are less explored. Consequently, a comprehensive and expressive health KB that spans all aspects of biomedical knowledge is still missing. The main goal of this thesis is to develop principled methods for building such a KB and enabling knowledge-centric applications. We address several challenges and make the following contributions: - To construct a health KB, we devise a largely automated and scalable pattern-based knowledge extraction method covering a spectrum of different text genres and distilling a wide variety of facts from different biomedical areas. - To consider higher-arity relations, crucial for proper knowledge representation in advanced domain such as health, we generalize the fact-pattern duality paradigm of previous methods. A key novelty is the integration of facts with missing arguments by extending our framework to partial patterns and facts by reasoning over the composability of partial facts. - To demonstrate the benefits of a health KB, we devise systems for entity-aware search and analytics and for entity-relationship-oriented exploration. Extensive experiments and use-case studies demonstrate the viability of the proposed approaches.
Export
BibTeX
@phdthesis{Ernstphd2017, TITLE = {Biomedical Knowledge Base Construction from Text and its Applications in Knowledge-based Systems}, AUTHOR = {Ernst, Patrick}, LANGUAGE = {eng}, URL = {urn:nbn:de:bsz:291-scidok-ds-271051}, DOI = {10.22028/D291-27105}, SCHOOL = {Universit{\"a}t des Saarlandes}, ADDRESS = {Saarbr{\"u}cken}, YEAR = {2018}, ABSTRACT = {While general-purpose Knowledge Bases (KBs) have gone a long way in compiling comprehensive knowledgee about people, events, places, etc., domain-specific KBs, such as on health, are equally important, but are less explored. Consequently, a comprehensive and expressive health KB that spans all aspects of biomedical knowledge is still missing. The main goal of this thesis is to develop principled methods for building such a KB and enabling knowledge-centric applications. We address several challenges and make the following contributions: -- To construct a health KB, we devise a largely automated and scalable pattern-based knowledge extraction method covering a spectrum of different text genres and distilling a wide variety of facts from different biomedical areas. -- To consider higher-arity relations, crucial for proper knowledge representation in advanced domain such as health, we generalize the fact-pattern duality paradigm of previous methods. A key novelty is the integration of facts with missing arguments by extending our framework to partial patterns and facts by reasoning over the composability of partial facts. -- To demonstrate the benefits of a health KB, we devise systems for entity-aware search and analytics and for entity-relationship-oriented exploration. Extensive experiments and use-case studies demonstrate the viability of the proposed approaches.}, }
Endnote
%0 Thesis %A Ernst, Patrick %Y Weikum, Gerhard %A referee: Verspoor, Karin %A referee: Berberich, Klaus %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations Databases and Information Systems, MPI for Informatics, Max Planck Society %T Biomedical Knowledge Base Construction from Text and its Applications in Knowledge-based Systems : %G eng %U http://hdl.handle.net/21.11116/0000-0001-1864-4 %U urn:nbn:de:bsz:291-scidok-ds-271051 %R 10.22028/D291-27105 %I Universität des Saarlandes %C Saarbrücken %D 2018 %8 20.02.2018 %P 147 p. %V phd %9 phd %X While general-purpose Knowledge Bases (KBs) have gone a long way in compiling comprehensive knowledgee about people, events, places, etc., domain-specific KBs, such as on health, are equally important, but are less explored. Consequently, a comprehensive and expressive health KB that spans all aspects of biomedical knowledge is still missing. The main goal of this thesis is to develop principled methods for building such a KB and enabling knowledge-centric applications. We address several challenges and make the following contributions: - To construct a health KB, we devise a largely automated and scalable pattern-based knowledge extraction method covering a spectrum of different text genres and distilling a wide variety of facts from different biomedical areas. - To consider higher-arity relations, crucial for proper knowledge representation in advanced domain such as health, we generalize the fact-pattern duality paradigm of previous methods. A key novelty is the integration of facts with missing arguments by extending our framework to partial patterns and facts by reasoning over the composability of partial facts. - To demonstrate the benefits of a health KB, we devise systems for entity-aware search and analytics and for entity-relationship-oriented exploration. Extensive experiments and use-case studies demonstrate the viability of the proposed approaches. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/26987
[2]
W. Li, “From Perception over Anticipation to Manipulation,” Universität des Saarlandes, Saarbrücken, 2018.
Abstract
From autonomous driving cars to surgical robots, robotic system has enjoyed significant growth over the past decade. With the rapid development in robotics alongside the evolution in the related fields, such as computer vision and machine learning, integrating perception, anticipation and manipulation is key to the success of future robotic system. In this thesis, we explore different ways of such integration to extend the capabilities of a robotic system to take on more challenging real world tasks. On anticipation and perception, we address the recognition of ongoing activity from videos. In particular we focus on long-duration and complex activities and hence propose a new challenging dataset to facilitate the work. We introduce hierarchical labels over the activity classes and investigate the temporal accuracy-specificity trade-offs. We propose a new method based on recurrent neural networks that learns to predict over this hierarchy and realize accuracy specificity trade-offs. Our method outperforms several baselines on this new challenge. On manipulation with perception, we propose an efficient framework for programming a robot to use human tools. We first present a novel and compact model for using tools described by a tip model. Then we explore a strategy of utilizing a dual-gripper approach for manipulating tools – motivated by the absence of dexterous hands on widely available general purpose robots. Afterwards, we embed the tool use learning into a hierarchical architecture and evaluate it on a Baxter research robot. Finally, combining perception, anticipation and manipulation, we focus on a block stacking task. First we explore how to guide robot to place a single block into the scene without collapsing the existing structure. We introduce a mechanism to predict physical stability directly from visual input and evaluate it first on a synthetic data and then on real-world block stacking. Further, we introduce the target stacking task where the agent stacks blocks to reproduce a tower shown in an image. To do so, we create a synthetic block stacking environment with physics simulation in which the agent can learn block stacking end-to-end through trial and error, bypassing to explicitly model the corresponding physics knowledge. We propose a goal-parametrized GDQN model to plan with respect to the specific goal. We validate the model on both a navigation task in a classic gridworld environment and the block stacking task.
Export
BibTeX
@phdthesis{Wenbinphd2018, TITLE = {From Perception over Anticipation to Manipulation}, AUTHOR = {Li, Wenbin}, LANGUAGE = {eng}, DOI = {10.22028/D291-27156}, SCHOOL = {Universit{\"a}t des Saarlandes}, ADDRESS = {Saarbr{\"u}cken}, YEAR = {2018}, DATE = {2018}, ABSTRACT = {From autonomous driving cars to surgical robots, robotic system has enjoyed significant growth over the past decade. With the rapid development in robotics alongside the evolution in the related fields, such as computer vision and machine learning, integrating perception, anticipation and manipulation is key to the success of future robotic system. In this thesis, we explore different ways of such integration to extend the capabilities of a robotic system to take on more challenging real world tasks. On anticipation and perception, we address the recognition of ongoing activity from videos. In particular we focus on long-duration and complex activities and hence propose a new challenging dataset to facilitate the work. We introduce hierarchical labels over the activity classes and investigate the temporal accuracy-specificity trade-offs. We propose a new method based on recurrent neural networks that learns to predict over this hierarchy and realize accuracy specificity trade-offs. Our method outperforms several baselines on this new challenge. On manipulation with perception, we propose an efficient framework for programming a robot to use human tools. We first present a novel and compact model for using tools described by a tip model. Then we explore a strategy of utilizing a dual-gripper approach for manipulating tools -- motivated by the absence of dexterous hands on widely available general purpose robots. Afterwards, we embed the tool use learning into a hierarchical architecture and evaluate it on a Baxter research robot. Finally, combining perception, anticipation and manipulation, we focus on a block stacking task. First we explore how to guide robot to place a single block into the scene without collapsing the existing structure. We introduce a mechanism to predict physical stability directly from visual input and evaluate it first on a synthetic data and then on real-world block stacking. Further, we introduce the target stacking task where the agent stacks blocks to reproduce a tower shown in an image. To do so, we create a synthetic block stacking environment with physics simulation in which the agent can learn block stacking end-to-end through trial and error, bypassing to explicitly model the corresponding physics knowledge. We propose a goal-parametrized GDQN model to plan with respect to the specific goal. We validate the model on both a navigation task in a classic gridworld environment and the block stacking task.}, }
Endnote
%0 Thesis %A Li, Wenbin %Y Fritz, Mario %A referee: Leonardis, Aleš %A referee: Slussalek, Philip %+ Computer Vision and Multimodal Computing, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Computer Vision and Multimodal Computing, MPI for Informatics, Max Planck Society External Organizations External Organizations %T From Perception over Anticipation to Manipulation : %G eng %U http://hdl.handle.net/21.11116/0000-0001-4193-F %R 10.22028/D291-27156 %I Universität des Saarlandes %C Saarbrücken %D 2018 %P 165 p. %V phd %9 phd %X From autonomous driving cars to surgical robots, robotic system has enjoyed significant growth over the past decade. With the rapid development in robotics alongside the evolution in the related fields, such as computer vision and machine learning, integrating perception, anticipation and manipulation is key to the success of future robotic system. In this thesis, we explore different ways of such integration to extend the capabilities of a robotic system to take on more challenging real world tasks. On anticipation and perception, we address the recognition of ongoing activity from videos. In particular we focus on long-duration and complex activities and hence propose a new challenging dataset to facilitate the work. We introduce hierarchical labels over the activity classes and investigate the temporal accuracy-specificity trade-offs. We propose a new method based on recurrent neural networks that learns to predict over this hierarchy and realize accuracy specificity trade-offs. Our method outperforms several baselines on this new challenge. On manipulation with perception, we propose an efficient framework for programming a robot to use human tools. We first present a novel and compact model for using tools described by a tip model. Then we explore a strategy of utilizing a dual-gripper approach for manipulating tools – motivated by the absence of dexterous hands on widely available general purpose robots. Afterwards, we embed the tool use learning into a hierarchical architecture and evaluate it on a Baxter research robot. Finally, combining perception, anticipation and manipulation, we focus on a block stacking task. First we explore how to guide robot to place a single block into the scene without collapsing the existing structure. We introduce a mechanism to predict physical stability directly from visual input and evaluate it first on a synthetic data and then on real-world block stacking. Further, we introduce the target stacking task where the agent stacks blocks to reproduce a tower shown in an image. To do so, we create a synthetic block stacking environment with physics simulation in which the agent can learn block stacking end-to-end through trial and error, bypassing to explicitly model the corresponding physics knowledge. We propose a goal-parametrized GDQN model to plan with respect to the specific goal. We validate the model on both a navigation task in a classic gridworld environment and the block stacking task. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/27026
[3]
A. Mishra, “Leveraging Semantic Annotations for Event-focused Search & Summarization,” Universität des Saarlandes, Saarbrücken, 2018.
Abstract
Today in this Big Data era, overwhelming amounts of textual information across different sources with a high degree of redundancy has made it hard for a consumer to retrospect on past events. A plausible solution is to link semantically similar information contained across the different sources to enforce a structure thereby providing multiple access paths to relevant information. Keeping this larger goal in view, this work uses Wikipedia and online news articles as two prominent yet disparate information sources to address the following three problems: • We address a linking problem to connect Wikipedia excerpts to news articles by casting it into an IR task. Our novel approach integrates time, geolocations, and entities with text to identify relevant documents that can be linked to a given excerpt. • We address an unsupervised extractive multi-document summarization task to generate a fixed-length event digest that facilitates efficient consumption of information contained within a large set of documents. Our novel approach proposes an ILP for global inference across text, time, geolocations, and entities associated with the event. • To estimate temporal focus of short event descriptions, we present a semi-supervised approach that leverages redundancy within a longitudinal news collection to estimate accurate probabilistic time models. Extensive experimental evaluations demonstrate the effectiveness and viability of our proposed approaches towards achieving the larger goal.
Export
BibTeX
@phdthesis{Mishraphd2018, TITLE = {Leveraging Semantic Annotations for Event-focused Search \& Summarization}, AUTHOR = {Mishra, Arunav}, LANGUAGE = {eng}, URL = {urn:nbn:de:bsz:291-scidok-ds-271081}, DOI = {10.22028/D291-27108}, SCHOOL = {Universit{\"a}t des Saarlandes}, ADDRESS = {Saarbr{\"u}cken}, YEAR = {2018}, ABSTRACT = {Today in this Big Data era, overwhelming amounts of textual information across different sources with a high degree of redundancy has made it hard for a consumer to retrospect on past events. A plausible solution is to link semantically similar information contained across the different sources to enforce a structure thereby providing multiple access paths to relevant information. Keeping this larger goal in view, this work uses Wikipedia and online news articles as two prominent yet disparate information sources to address the following three problems: \mbox{$\bullet$} We address a linking problem to connect Wikipedia excerpts to news articles by casting it into an IR task. Our novel approach integrates time, geolocations, and entities with text to identify relevant documents that can be linked to a given excerpt. \mbox{$\bullet$} We address an unsupervised extractive multi-document summarization task to generate a fixed-length event digest that facilitates efficient consumption of information contained within a large set of documents. Our novel approach proposes an ILP for global inference across text, time, geolocations, and entities associated with the event. \mbox{$\bullet$} To estimate temporal focus of short event descriptions, we present a semi-supervised approach that leverages redundancy within a longitudinal news collection to estimate accurate probabilistic time models. Extensive experimental evaluations demonstrate the effectiveness and viability of our proposed approaches towards achieving the larger goal.}, }
Endnote
%0 Thesis %A Mishra, Arunav %Y Berberich, Klaus %A referee: Weikum, Gerhard %A referee: Hauff, Claudia %+ Databases and Information Systems, MPI for Informatics, Max Planck Society International Max Planck Research School, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society Databases and Information Systems, MPI for Informatics, Max Planck Society External Organizations %T Leveraging Semantic Annotations for Event-focused Search & Summarization : %G eng %U http://hdl.handle.net/21.11116/0000-0001-1844-8 %U urn:nbn:de:bsz:291-scidok-ds-271081 %R 10.22028/D291-27108 %I Universität des Saarlandes %C Saarbrücken %D 2018 %8 08.02.2018 %P 252 p. %V phd %9 phd %X Today in this Big Data era, overwhelming amounts of textual information across different sources with a high degree of redundancy has made it hard for a consumer to retrospect on past events. A plausible solution is to link semantically similar information contained across the different sources to enforce a structure thereby providing multiple access paths to relevant information. Keeping this larger goal in view, this work uses Wikipedia and online news articles as two prominent yet disparate information sources to address the following three problems: • We address a linking problem to connect Wikipedia excerpts to news articles by casting it into an IR task. Our novel approach integrates time, geolocations, and entities with text to identify relevant documents that can be linked to a given excerpt. • We address an unsupervised extractive multi-document summarization task to generate a fixed-length event digest that facilitates efficient consumption of information contained within a large set of documents. Our novel approach proposes an ILP for global inference across text, time, geolocations, and entities associated with the event. • To estimate temporal focus of short event descriptions, we present a semi-supervised approach that leverages redundancy within a longitudinal news collection to estimate accurate probabilistic time models. Extensive experimental evaluations demonstrate the effectiveness and viability of our proposed approaches towards achieving the larger goal. %U https://publikationen.sulb.uni-saarland.de/handle/20.500.11880/26995