Discriminative learning from partially annotated examples
Hlaváč, Václav; Franc, Vojtěch; Antoniuk, Kostiantyn
2016 - anglický
A number of algorithms and its applications for automatic classi ers learning from examples
is ever growing. Most of existing algorithms require a training set of completely annotated
examples, which are often hard to obtain. In this thesis, we tackle the problem of learning
from partially annotated examples, which means that each training input comes with a set
of admissible labels only one of which is correct. We contributed to two di erent cases of
this scenario. In the rst case, we studied the problem of learning the ordinal classi ers from
examples with interval annotation of labels. We designed a convex learning algorithm for this
case and demonstrated its advantage on real data empirically. At the same time, we made
several contributions to the supervised learning of the ordinal classi ers, namely, we proposed
new parametrization of the ordinal classi er, we introduced more
exible piece wise version
of the ordinal classi er, and we proposed a generic cutting plane solver with convergence
guarantees. In the second case, we studied the problem of learning the structured output
classi ers from examples with missing annotation of a subset of labels. We have de ned
the concept of a surrogate classi cation calibrated partial loss, the minimization of which
guarantees that learning is statistical consistent under fairly general conditions on the data
generating process. We proved the existence of a convex classi cation calibrated surrogate loss
for learning from partially annotated examples. We showed which existing surrogate losses
are classi cation calibrated and which are not. Our work thus provides a missing theoretical
justi cation for so far heuristic methods which have been successfully used in practice.
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Discriminative learning from partially annotated examples
A number of algorithms and its applications for automatic classi ers learning from examples is ever growing. Most of existing algorithms require a training set of completely annotated examples, ...
Design and Integration of Simulation Models for Industrial Systems
Šindelář, Radek; Novák, Petr
2016 - anglický
Industrial systems are becoming complex and large-scale. Optimization of their operation
and testing of their control systems are done on simulation models frequently, because
simulated experiments are faster, cheaper, and repeatable compared to experiments done
on real industrial plants. However, design and re-design of simulation models are difficult
and time-consuming tasks. In addition, integration of simulation models within industrial
automation systems is not satisfactory nowadays. This thesis is aimed at improving the
design and integration phases of the simulation model life-cycle.
In the area of the simulation model design, especially a component-based approach for
simulation model creation is investigated and improved in this thesis. It assumes that engineering
systems consist of atomic components that are connected into topologies of real
industrial plants. The proposed method supports assembling simulation models from simulation
components, which can be reused from previous simulation projects. Each real device
can be simulated by one of the available implementations of the component, representing
this device. The proposed solution is based on the utilization of the bond-graph theory
to guarantee the compatibility of the interfaces of the connected component implementations
and to support their selection. In addition, the bond-graph theory is used to support
splitting a simulation model into a set of simulation modules and their integration into a
simulation workflow. For all of these types of tasks, the bond-graph theory was enhanced
with an explicit description of component interfaces and a new causality assignment algorithm
was designed. This algorithm can be used not only for generation of simulation
models, but also for verifications on a conceptual planning level, whether specific sets of
simulation component implementations are sufficient to model particular plants.
In the area of the simulation model integration, two research threads are followed. The
first one is related to formalizing, capturing, and integrating knowledge about the real industrial
plant, input and output tags, parameters of devices, and mappings of all these entities
to simulation model components, variables, and parameters. Such engineering knowledge
is used to support simulation model design and maintenance of existing simulation models
when a real plant is changed. The second thread in the integration area is focused on
interoperability of simulation modules on the level of the supervisory control and data acquisition
of the automation pyramid. This task covers the access of simulations to runtime
data, improved parameter setting, and version-control of simulation modules.
This thesis contributes to the areas of the simulation modeling, knowledge representation,
and distributed system integration. The most important results are (i) adaptation
of the bond graph theory for non-traditional applications including selection of explicitly
specified component implementations as well as a new causality assignment algorithm supporting
this approach, (ii) utilization of ontologies for supporting simulation model design
and integration, and (iii) improved simulation model integration.
ii
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Design and Integration of Simulation Models for Industrial Systems
Industrial systems are becoming complex and large-scale. Optimization of their operation and testing of their control systems are done on simulation models frequently, because simulated experiments ...
Long-term combined heat and power production and trade planning
Havel, Petr; Dvořák, Michal
2016 - anglický
In this thesis a comprehensive framework
for solving long-term combined
heat and power (CHP) operations
planning problems is developed. The
framework has two main parts - the
first is a modelling framework which
allows for modelling arbitrary CHP
plants and is aimed at the formulation
of an optimization problem
for CHP production and trade planning.
The second is a solution algorithm
which exploits the knowledge
of the problem structure so that the
problem is solved more efficiently.
There exist very powerful stateof-
the-art general-purpose solvers
for mixed-integer linear programming
(MILP) problems, such as
Gurobi. However, even these solvers
fail to find a feasible solution within
reasonable time for production planning
problems of large dimensions.
An idea followed in this thesis is
to achieve reasonable computation
times of large problems by employing
the knowledge of the special problem
structure.
For this purpose, a customized
branch and bound (B&B)
algorithm is proposed. The algorithm
exploits the knowledge of
the block-diagonal problem substructure,
to obtain tight bounds.
The bounds are much tighter than
bounds produced by solving a linear
relaxation of the solved MILP
problem, which is the way of bound
computation commonly used within
general-purpose implementations
of B&B. Besides an enhanced horizon
cutting algorithm is developed,
with the purpose of providing
high-quality feasible solutions for
the customized B&B algorithm.
Efficiency of the proposed algorithm
was evaluated based on 64 test
cases using real-world data of three
existing CHP plants. The performance
of the proposed algorithm was
compared to plain Gurobi usage. In
most cases the proposed algorithm
finds a certificate of near-optimality
sooner than plain Gurobi does. More
importantly, the proposed algorithm
was able to find good feasible solutions
for problems, for which Gurobi
fails to find any feasible solution
within the specified time limit.
Klíčová slova:
MILP; CHP; optimization; operations planning; Lagrangian relaxation; branch-and-bound; heuristics
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Long-term combined heat and power production and trade planning
In this thesis a comprehensive framework for solving long-term combined heat and power (CHP) operations planning problems is developed. The framework has two main parts - the first is a modelling ...
Rare Earth Doped Fiber Lasers for Spectral Region about 2 Micrometers
Zvánovec, Stanislav; Peterka, Pavel; Písařík, Michael
2016 - anglický
Previous fiber laser research was strictly targeted for high efficiency, economical and exceedingly
thin‐disk lasers within beam quality parameters. Ytterbium doped silica fiber was the preferred fiber
for most industrial applications. However, rest of rare earth doped active fibers were only
demonstrated in basic principles via laboratory experiments, suggesting that many interesting
solutions have yet to be studied. The purpose of this thesis is to develop methodology of key fiber
laser components for wavelength range of 1.7 μm to 2.2 μm and to evaluate applications of these
components to basic fiber lasers concepts further to explore and exploit their potential uses.
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Rare Earth Doped Fiber Lasers for Spectral Region about 2 Micrometers
Previous fiber laser research was strictly targeted for high efficiency, economical and exceedingly thin‐disk lasers within beam quality parameters. Ytterbium doped silica fiber was the preferred ...
Prosody Utilization in Continuous Speech Recognition
Hanžl, Václav; Bartošek, Jan
2016 - anglický
This doctoral thesis covers the theme of prosody utilization in automatic recognition
of continuous speech. Even though automatic speech recognition (ASR) systems have
imoproved immensely over the last several decades, they still lack making use of one of
the most important aspect of information using speech, which is a prosody. There have
already been proofs from other languages about the favourableness of prosody usage in
ASR and doctoral thesis tries to investigate the potential of Czech regarding prosody
usage.
The research activities can be divided into three main areas: a) pitch detection algorithms
(PDA) as needed prerequisite for prosodic feature extraction, b) Czech lexical
stress system as potential clue from acoustic signal for word boundary detection (and its
usage in ASR) and c) classi cation of sentence/phrase modality in Czech based purely on
an acoustic signal.
Firstly, the eld of pitch detection algorithms, a framework for their evaluation and
comparison is presented. Several new evaluation criteria are proposed as an extension to
existing ones together with metrics evaluation over four speech pitch reference databases.
Besides pure comparison, few modi cations of existing PDA methods are presented.
Namely a transition probability function in PDA post-processing is investigated in terms
of candidate distance measure and new temporal-forgetting principle for speech is brought
in as extension of method by time domain.
Czech as a xed-stress language with lexical stress on the rst syllable is known to have
a weak lexical stress acoustic correlation. Nevertheless, methods of how stressed syllables
or stress-group boundaries can be detected from speech signal were investigated. A system
with sophisticated feature extraction followed by statistical machine learning methods
to model those phenomenon in Czech is presented. Detected stress-group boundaries
can be (in most of cases) mapped to word boundaries which can be used for prosodic
evaluation of ASR hypothesis. A metric for such prosodic score, which can be directly
used in prosodic N-best evaluation or ASR error detection, is proposed. Also, ASR lattice
rescoring algorithm for Czech is presented.
Czech phrase modality detection from acoustic signal is covered and together with
existing phrase boundary detector can such system serve as an punctuation module for
Czech dictation ASR system or in Czech dialogue system to support its natural language
processing (NLP) part.
Klíčová slova:
Prosody; speech technology; ASR; F0; pitch; lexical stress; stress group; modality; melodeme; prosodic hypothesis scoring
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Prosody Utilization in Continuous Speech Recognition
This doctoral thesis covers the theme of prosody utilization in automatic recognition of continuous speech. Even though automatic speech recognition (ASR) systems have imoproved immensely over the ...
Numerical Modeling of Nonlocal Energy Transport in Laser-Heated Plasmas
Liska, Richard; Weber, Stefan; Holec, Milan
2016 - anglický
Modeling of the nonlocal energy transport in laser-heated plasmas is a challenging task.
In order to include such a transport into simulations of plasmas, we propose the nonlocal
transport hydrodynamic model, which provides a kinetic model and the classical
fluid description at the same time. It resides in direct solution of electron and
photon transport equations based on the BGK collision operator which gives an inherent
coupling of energy transport to the plasma fluid. Our high-order discontinuous
Galerkin scheme of the BGK transport equations and the fluid energy equations gives
solutions obeying any regime of transport, i.e. between the local diffusion asymptotic
and the collisionless transport asymptotic of free-streaming particles, which is demonstrated
in the case of exact steady transport and approximate multi-group diffusion
numerical tests. As an application of the nonlocal transport hydrodynamic model, we
present simulation results of the ultra-intense laser prepulse interaction with solid targets
of different atomic numbers, and results of the laser-driven shock in a plastic foam
which is related to study of warm-dense-matter state of carbon. The simulations are
calculated using our new Plasma Euler and Transport Equations nonlocal transport
hydrodynamic code PETE.
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Numerical Modeling of Nonlocal Energy Transport in Laser-Heated Plasmas
Modeling of the nonlocal energy transport in laser-heated plasmas is a challenging task. In order to include such a transport into simulations of plasmas, we propose the nonlocal transport ...
Hierarchical probabilistic model of language acquisition
Vavrečka, Michal; Lhotská, Lenka; Štěpánová, Karla
2016 - anglický
In this thesis, I propose an unsupervised
computational model of language acquisition
through visual grounding. I especially
focus on a case where the language
input is in a form of variable length sentences.
The state-of-the-art cognitive architectures
with the focus on grounding
language in vision are explored. I take
an advantage of probabilistic Bayesian
models which are besides neural networks
one of the main tools used in a computational
cognitive modeling. The probabilistic
(Bayesian) models have been used
in the tasks such as language processing,
decision making or causality learning. In
the first part of the thesis newly proposed
method for estimating a number of clusters
in data is described. In the second
part of the thesis I focus on the description
of the cognitive architecture itself. The
developed hierarchical cognitive architecture
processes separately visual (static)
and language (time-sequence) data and
combines them in a multimodal layer. The
important feature is a compositionality of
the system - ability to derive meaning
of previously unheard sentences and unseen
objects and its ability to learn all
features describing the object from sentences
of variable length. The proposed
architecture was implemented into the humanoid
robot iCub and tested on both
artificially generated data and on the realworld
data.
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Hierarchical probabilistic model of language acquisition
In this thesis, I propose an unsupervised computational model of language acquisition through visual grounding. I especially focus on a case where the language input is in a form of variable ...
Modeling of Spiral Polysilicon Divider in High Voltage MOSFET Transistor and Leakage
Dobeš, Josef; Paňko, Václav
2015 - anglický
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Modeling of Spiral Polysilicon Divider in High Voltage MOSFET Transistor and Leakage
Model Transformation Approach to Automated Model Driven Development
Richta, Karel; Viet Cuong Nguyen
2015 - anglický
One of the contemporary challenges of software evolution is to adapt a software system
to the changing of requirements and demands from users and environments. An ultimate
goal is to encapsulate these requirements into a high-level abstraction, giving the ability
to achieve large-scale adaptation of the underlying software implementation. Model-Driven
Engineering (MDE) is one of the enabling techniques that supports this objective. In MDE,
the e ective creation of models and their transformation are core activities to enable the
conversion of source models to target models in order to change model structures or translate
models to other software artifacts. The main goal is to provide automation and enable
the automated development of a system from its corresponding models. There are several
approaches on this matter from high level. However, there is still absence of clear methodology
and results on how to apply MDE for a speci c domain with speci c requirements such
as the web domain. This research brings contribution toward the solution to automated
model development by providing an overview of existing approaches and introducing a novel
approach in the emerging eld of web applications and services.
To cope with current trend in the growing of complexity of web services as programmatic
backbones of modern distributed and cloud architecture, we present an approach using
domain speci c language for modeling of web services as the solution to the challenge in
scalability of web service modeling and development. We analyze the current state of the
problem domain and implement a domain speci c language called Simple Web Service Modeling
to support automated model-driven development of such web services. This approach
is the solution to the problem in web service development of software-as-service systems that
require the support for tenant-speci c architecture.
In the domain of web application quality assurance, we build a modeling language for
model driven testing of web application that focuses on automation and regression testing.
Our techniques are based on building abstractions of web pages and modeling state-machinebased
test behavior using Web Testing Modeling Language - a domain speci c language
that we developed for web page modeling. This methodology and techniques aim at helping
software developers as well as testers to become more productive and reduce the time-tomarket,
while maintaining high standards of web application. The proposing techniques is
the answer to the lack of concrete methods and toolset in applying model driven development
to speci c areas such as web application testing and services. The results of this work can
be applied to practical purposes with the methodological support to integrate into existing
software development practices.
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Model Transformation Approach to Automated Model Driven Development
One of the contemporary challenges of software evolution is to adapt a software system to the changing of requirements and demands from users and environments. An ultimate goal is to encapsulate ...
Numerical Studies of Plasma Instabilities
Kulhánek, Petr; Karas, Vladinír; Horký, Miroslav
2015 - anglický
Plné texty jsou dostupné v digitálním repozitáři ČVUT.
Numerical Studies of Plasma Instabilities
NUŠL poskytuje centrální přístup k informacím o šedé literatuře vznikající v ČR v oblastech vědy, výzkumu a vzdělávání. Více informací o šedé literatuře a NUŠL najdete na webu služby.
Vaše náměty a připomínky posílejte na email nusl@techlib.cz
Provozovatel
Zahraniční báze