Dissertation topics

3D scene reconstruction from images is a fundamental problem of computer vision. Our goal will be to advance the state of the art, publish at CVPR, ICCV, ECCV, in IJCV and in PAMI. We will collaborate with ETH Zurich, JR Graz, NII Tokyo, Google, Microsoft and Leica. The topic is suitable for students with interest in algorithms, experimental work, and engineering of really working systems. ½-1 year stay abroad expected. people.ciirc.cvut.cz/~pajdla/

Evolutionary algorithms are, in the last 20 years, one of the most successful methods for solving non-traditional optimization problems, such as search for the most suitable documents containing required information, discovery of the most interesting knowledge in available data, or other kinds of optimization tasks in which the values of the objective function can be obtained only empirically. Because evolutionary algorithms employ only function values of the objective function, they approach its optimum much more slowly than optimization methods for smooth functions, which make use of information about the objective function gradients as well, possibly also about its second derivatives. This property of evolutionary algorithms is particularly disadvantageous in the context of costly and time-consuming empirical way of obtaining values of the objective function. However, evolutionary algorithms can be substantially speeded up if they employ the empirical objective function only sometimes when evaluating objective function values, whereas they mostly evaluate only a sufficiently accurate regression model of that function. Among most promising regression models are models based on Gaussian processes. Therefore, it is not surprising that exactly these models belonged among the first that started to be used for accelerating evolutionary optimization. Nevertheless, research into accelerating evolutionary algorithms by means of Gaussian processes is only at a beginning. It should be contributed also by the proposed thesis.

The focus on sustainability in building renovation and maintenance includes implementing sustainable materials based on recycled resources. The acoustic and thermal properties of these materials are crucial for comfort and energy efficiency. Some transient methods such as the Hot-Disk (THD) or the novel Hot-Bridge (THB) will be used to assess simultaneously both thermal conductivity and heat capacity. For acoustic parameters impedance tube methods will be used to evaluate sound absorption as well as sound transmission coefficients. The thermal properties will be studied mainly at the University of Brescia, while the acoustic parameters will be evaluated mainly at the CTU.

The topic is focused on evaluation and design of spaces for various purposes. Modeling and measurement of new room acoustic parameters will lead to recommendations for acoustic designers as well as for cataloguing of selected spaces. Further information can be obtained from supervisor. http://fyzika.feld.cvut.cz/index.php?_mli=jc&_uid=21

The topic is to deal with investigation of acoustic waves propagating through nonlocal dispersion zones using non-uniform profiles of the density and elastic properties of materials, which enables us to control reflection and transmission of the waves in the specified spectral range. As nonlocal dispersion zones, it is supposed to use spatially confined regions with functionally graded material properties. To control transmission and reflection coefficients it is further possible to employ locally periodic/quasi-periodic structures based on nonlocal dispersion zones because even a relatively small number of repeating nonlocal dispersion zones evinces acoustic energy frequency band gaps.

Within this topic students will focus on development of active system for control of radiation from vibrating bodies including design of appropriate transducers and corresponding signal processing. Further information can be obtained from supervisor. http://fyzika.feld.cvut.cz/index.php?_mli=jc&_uid=21

Flutter is a dynamic instability of an elastic structure - an aircraft - in a fluid flow, caused by positive feedback between the body's deflection and the force exerted by the fluid flow. In a linear system, 'flutter point' is the point at which the structure is undergoing simple harmonic motion - zero net damping - and so any further decrease in net damping will result in a self-oscillation and eventual failure. Prevention of aircraft flutter is done traditionally by mechanical modifications - by changing mechanical stiffness and/or mass distribution essentially. The weight penalty is essentially inevitable in this case however. For this reason, active control approaches have been considered in recent years, and in a very limited number of cases, they have got to final implementations. In all cases however, the solutions aimed at large transportation or military projects. The main goal of this project though is to develop solutions applicable for small transportation aircraft: either of the GA category, or "ultralights" (small sports aircraft) where the flutter issue is critical. The specifics for these categories are related to technical limitations in instrumentation and to legal and certification issues (often fully mechanical instrumentation is expected or requested). Flutter led to many documented catastrophic causalties of ultalight aircraft in the Czech Republic in the past. Expected technical outputs are simulation flight mechanics models for selected case studies, designed control laws, simulation validation and verification, realization of an experimental setup, and experimental tests in wind tunnel. Collaboration on experimental validation of results is planned with the Aerospace Department, FME CVUT, which also serves as an accreditted lab for flutter cerrtificates of ultralight aircraft.

The thesis should contain: i) a survey of active learning methods as an extension of semi-supervised learning, ii) survey of structured data representations with special focus on deep neural networks and their uncertainty, iii) a novel method of active learning for structured data such as text documents, JSON files, or graphs, iv) an analysis of suitability of uncertainty representations for these data, v) an algorithm using specific properties of these data for improved active learning, vi) detailed comparison of all proposed methods with existing state of the art.

The research is focused on the intersection of active learning and deep neural networks within the realm of natural language processing. The study explores and develops advanced uncertainty representation and measurement methodologies for deep neural networks. Contribution aims to enhance the efficiency and performance of active learning strategies across multiple natural language processing tasks.

Active learning is a well-known class of algorithms for supervised learning aiming to minimize the training effort of the human annotator of the cost of training. However, active learning strategies are almost nonexistent for weakly supervised learning that aims to relax the need for strict input-output relation of the trained model. Weak supervision is used, e.g., in interactive explainable learning or metric learning. The aim of this thesis is to introduce active strategies in this domain. The challenge is to design probabilistic models that are capable of expressing uncertainty over actions that are used in this domain, such as sets of unknown cardinality. We have recently proposed a novel class of tractable probabilistic models that can be used for this task.

Theoretical form of navigation solutions in terms of position, velocity, and attitude (PVA) estimates is already well known; however, providing PVA estimates with high accuracy under challenging environmental conditions still plays a key role of current scientific interests. A prime concern is paid to the robustness and accuracy of navigation solutions under real flying conditions where the accuracy can be strongly affected by the varying vibrating, maneuvering impacts as well as environmental impacts and turbulence. Therefore, the area of this PhD study project addresses such problems and will focus on R&D of enhanced integrated solution ensuring improved performance with extended consideration of a sensory assembly. The emphasis will be put onto integration of available navigation sources, e.g. magnetometers, pressure sensors, GNSS positioning, laser scanners, radars etc. to guarantee accuracy even under unfriendly environmental conditions. The navigation solution will be applied on unmanned aerial vehicles (UAVs) whose control systems rely on precise and accurate navigation with a minimal latency. The R&D of such a control system is also going to be a part of this project. This control should then handle situations such as landing of the UAV on a moving platform no matter the environmental conditions and availability of GNSS signal.

The topic deals with designing new adaptive algorithms of image synthesis that focus their computational effort on solving important components of direct and indirect illumination. An important aspect of the work is considering a specified range of computational power that the algorithm can use and thus working on a time budget basis. We assume exploiting scalable data structures and algorithms based on evaluating importance of simulated effects and their subsequent adaptive usage by the density estimation methods and Monte Carlo integration. Further information can be found at http://dcgi.felk.cvut.cz/people/bittner