Zapisovač letových údajů je zařízení určené pro zaznamenávání letových dat z různých senzorů v letadlech. Analýza letových údajů hraje důležitou roli ve vývoji a testování avioniky. Testování a hodnocení charakteristik letadla se často provádí pomocí testovacích manévrů. Naměřená data z jednoho letu jsou uložena v jednom letovém záznamu, který může obsahovat několik testovacích manévrů. Cílem této práce je identi kovat základní testovací manévry s pomocí naměřených letových dat. Teoretická část popisuje letové manévry a formát měřených letových dat. Analytická část popisuje výzkum v oblasti klasi kace založené na statistice a teorii pravděpodobnosti potřebnou pro pochopení složitých Gaussovských směšovacích modelů. Práce uvádí implementaci, kde jsou Gaussovy směšovací modely použité pro klasifi kaci testovacích manévrů. Navržené řešení bylo testováno pro data získána z letového simulátoru a ze skutečného letadla. Ukázalo se, že Gaussovy směšovací modely poskytují vhodné řešení pro tento úkol. Další možný vývoj práce je popsán v závěrečné kapitole.

<p>У оквиру докторске дисертације<br />представљен је нови алгоритам<br />хијерархијског кластеровања модела<br />Гаусових смеша, заснован на операцији<br />поделе и спајања. Алгоритам тежи<br />побољшању локално оптималног<br />решења одређеног иницијалном<br />констелацијом. Иницијализује се<br />локално оптималним параметрима,<br />добијеним коришћењем референтног<br />приступа сличног k‐means‐у и тежи ка<br />приближавању глобалном оптимуму<br />функције циља, итеративном<br />применом операција поделе и спајања<br />над кластерима Гаусових компоненти,<br />одређеним применом референтног<br />алгоритма.<br />Додатно побољшање алгоритма<br />осварено је увођењем принципа<br />селекције модела у сврху одређивања<br />најповољнијег односа тачности и<br />рачунске сложености, у задатку&nbsp;<span style="font-size: 12px;">селекције гаусијана унутар реалног</span></p><p>система за препознавање. Предложени<br />метод тестиран је како над вештачки<br />генерисаним подацима, тако и у<br />оквиру алгоритма селекције гаусијана,<br />на примеру система за континуално<br />препознавање говора. У оба случаја<br />забележено је побољшање резултата у<br />односу на резултате остварене<br />применом референтног алгоритма.<br />Побољшања алгоритма селекције<br />гаусијана избором оптималног скупа<br />системских параметара такође су<br />размотрена.</p> / <p>U okviru doktorske disertacije<br />predstavljen je novi algoritam<br />hijerarhijskog klasterovanja modela<br />Gausovih smeša, zasnovan na operaciji<br />podele i spajanja. Algoritam teži<br />poboljšanju lokalno optimalnog<br />rešenja određenog inicijalnom<br />konstelacijom. Inicijalizuje se<br />lokalno optimalnim parametrima,<br />dobijenim korišćenjem referentnog<br />pristupa sličnog k‐means‐u i teži ka<br />približavanju globalnom optimumu<br />funkcije cilja, iterativnom<br />primenom operacija podele i spajanja<br />nad klasterima Gausovih komponenti,<br />određenim primenom referentnog<br />algoritma.<br />Dodatno poboljšanje algoritma<br />osvareno je uvođenjem principa<br />selekcije modela u svrhu određivanja<br />najpovoljnijeg odnosa tačnosti i<br />računske složenosti, u zadatku&nbsp;<span style="font-size: 12px;">selekcije gausijana unutar realnog</span></p><p>sistema za prepoznavanje. Predloženi<br />metod testiran je kako nad veštački<br />generisanim podacima, tako i u<br />okviru algoritma selekcije gausijana,<br />na primeru sistema za kontinualno<br />prepoznavanje govora. U oba slučaja<br />zabeleženo je poboljšanje rezultata u<br />odnosu na rezultate ostvarene<br />primenom referentnog algoritma.<br />Poboljšanja algoritma selekcije<br />gausijana izborom optimalnog skupa<br />sistemskih parametara takođe su<br />razmotrena.</p> / <p>The dissertation presents a novel splitand‐<br />merge algorithm for hierarchical<br />clustering of Gaussian mixture models.<br />The algorithm tends to improve on the<br />local optimal solution determined by the<br />initial constellation. It is initialized by<br />local optimal parameters obtained by<br />using a baseline approach similar to kmeans,<br />and it tends to approach more<br />closely to the global optimum of the<br />target clustering function, by iteratively<br />splitting and merging the clusters of<br />Gaussian components obtained as the<br />output of the baseline algorithm.<br />The algorithm is further improved by<br />introducing model selection in order to<br />obtain the best possible trade‐off<br />between recognition accuracy and<br />computational load in a Gaussian<br />selection task applied within an actual<br />recognition system. The proposed<br />method is tested both on artificial data<br />and in the framework of Gaussian<br />selection performed within a real&nbsp;<span style="font-size: 12px;">continuous speech recognition system. In</span></p><p>both cases an improvement over the<br />baseline method has been observed.<br />Additional improvements of Gaussian<br />selection algorithm by using the optimal<br />set of system parameters are also<br />discussed.</p>

<div>Multi-target tracking (MTT) is the problem to simultaneously estimate the number of targets and their states or trajectories. Numerous techniques have been developed for over 50 years, with a multitude of applications in many fields of study; however, there are two most widely used approaches to MTT: i) data association-based traditional algorithms; and ii) finite set statistics (FISST)-based data association free Bayesian multi-target filtering algorithms. Most data association-based traditional filters mainly use a statistical or simple model of the feature without explicitly considering the correlation between the target behavior</div><div>and feature characteristics. The inaccurate model of the feature can lead to divergence of the estimation error or the loss of a target in heavily cluttered and/or low signal-to-noise ratio environments. Furthermore, the FISST-based data association free Bayesian multi-target filters can lose estimates of targets frequently in harsh environments mainly</div><div>attributed to insufficient consideration of uncertainties not only measurement origin but also target's maneuvers.</div><div>To address these problems, three main approaches are proposed in this research work: i) new feature models (e.g., target dimensions) dependent on the target behavior</div><div>(i.e., distance between the sensor and the target, and aspect-angle between the longitudinal axis of the target and the axis of sensor line of sight); ii) new Gaussian mixture probability hypothesis density (GM-PHD) filter which explicitly considers the uncertainty in the measurement origin; and iii) new GM-PHD filter and tracker with jump Markov system models. The effectiveness of the analytical findings is demonstrated and validated with illustrative target tracking examples and real data collected from the surveillance radar.</div>

Práce pojednává o využití support vector machines v detekci řečové aktivity. V první části jsou zkoumány různé druhy příznaků, jejich extrakce a zpracování a je nalezena jejich optimální kombinace, která podává nejlepší výsledky. Druhá část představuje samotný systém pro detekci řečové aktivity a ladění jeho parametrů. Nakonec jsou výsledky porovnány s dvěma dalšími systémy, založenými na odlišných principech. Pro testování a ladění byla použita ERT broadcast news databáze. Porovnání mezi systémy bylo pak provedeno na databázi z NIST06 Rich Test Evaluations.

This thesis inquires the existing methods on the field of image recognition with regards to gesture recognition. Some methods have been chosen for deeper study and these are to be discussed later on. The second part goes in for the concenpt of an algorithm that would be able of robust gesture recognition based on data acquired within the AMI and M4 projects. A new ways to achieve precise information on participants position are suggested along with dynamic data processing approaches toward recognition. As an alternative, recognition using Gaussian Mixture Models and periodicity analysis are brought in. The gesture class in focus are speech supporting gestures. The last part demonstrates the results and discusses future work.

This master's thesis is dedicated to theme of knowledge discovery in Multimedia Databases, especially basic methods of classification and prediction used for data mining. The other part described about extraction of low level features from video data and images and summarizes information about content-based search in multimedia content and indexing this type of data. Final part is dedicated to implementation Gaussian mixtures model for classification and compare the final result with other method SVM.

In radiation therapy treatment planning, recent works have used machine learning based on historically delivered plans to automate the process of producing clinically acceptable plans. Compared to traditional approaches such as repeated weighted-sum optimization or multicriteria optimization (MCO), automated planning methods have, in general, the benefits of low computational times and minimal user interaction, but on the other hand lack the flexibility associated with general-purpose frameworks such as MCO. Machine learning approaches can be especially sensitive to deviations in their dose prediction due to certain properties of the optimization functions usually used for dose mimicking and, moreover, suffer from the fact that there exists no general causality between prediction accuracy and optimized plan quality.In this thesis, we present a means of unifying ideas from machine learning planning methods with the well-established MCO framework. More precisely, given prior knowledge in the form of either a previously optimized plan or a set of historically delivered clinical plans, we are able to automatically generate Pareto optimal plans spanning a dose region corresponding to plans which are achievable as well as clinically acceptable. For the former case, this is achieved by introducing dose--volume constraints; for the latter case, this is achieved by fitting a weighted-data Gaussian mixture model on pre-defined dose statistics using the expectation--maximization algorithm, modifying it with exponential tilting and using specially developed optimization functions to take into account prediction uncertainties.Numerical results for conceptual demonstration are obtained for a prostate cancer case with treatment delivered by a volumetric-modulated arc therapy technique, where it is shown that the methods developed in the thesis are successful in automatically generating Pareto optimal plans of satisfactory quality and diversity, while excluding clinically irrelevant dose regions. For the case of using historical plans as prior knowledge, the computational times are significantly shorter than those typical of conventional MCO. / Inom strålterapiplanering har den senaste forskningen använt maskininlärning baserat på historiskt levererade planer för att automatisera den process i vilken kliniskt acceptabla planer produceras. Jämfört med traditionella angreppssätt, såsom upprepad optimering av en viktad målfunktion eller flermålsoptimering (MCO), har automatiska planeringsmetoder generellt sett fördelarna av lägre beräkningstider och minimal användarinteraktion, men saknar däremot flexibiliteten hos allmänna ramverk som exempelvis MCO. Maskininlärningsmetoder kan vara speciellt känsliga för avvikelser i dosprediktionssteget på grund av särskilda egenskaper hos de optimeringsfunktioner som vanligtvis används för att återskapa dosfördelningar, och lider dessutom av problemet att det inte finns något allmängiltigt orsakssamband mellan prediktionsnoggrannhet och kvalitet hos optimerad plan. I detta arbete presenterar vi ett sätt att förena idéer från maskininlärningsbaserade planeringsmetoder med det väletablerade MCO-ramverket. Mer precist kan vi, givet förkunskaper i form av antingen en tidigare optimerad plan eller en uppsättning av historiskt levererade kliniska planer, automatiskt generera Paretooptimala planer som täcker en dosregion motsvarande uppnåeliga såväl som kliniskt acceptabla planer. I det förra fallet görs detta genom att introducera dos--volym-bivillkor; i det senare fallet görs detta genom att anpassa en gaussisk blandningsmodell med viktade data med förväntning--maximering-algoritmen, modifiera den med exponentiell lutning och sedan använda speciellt utvecklade optimeringsfunktioner för att ta hänsyn till prediktionsosäkerheter.Numeriska resultat för konceptuell demonstration erhålls för ett fall av prostatacancer varvid behandlingen levererades med volymetriskt modulerad bågterapi, där det visas att metoderna utvecklade i detta arbete är framgångsrika i att automatiskt generera Paretooptimala planer med tillfredsställande kvalitet och variation medan kliniskt irrelevanta dosregioner utesluts. I fallet då historiska planer används som förkunskap är beräkningstiderna markant kortare än för konventionell MCO.

Robust state estimation solutions must deal with faulty measurements, called outliers, and unknown data associations, which lead to multiple feasible hypotheses. Take, for instance, the scenario of tracking two indistinguishable targets based on position measurements, where each measurement could refer to either of the targets or even be a faulty reading. Common estimation methods model the state as having a unimodal distribution, so they are called unimodal methods. Likewise, multimodal methods model the state as a multimodal distribution. Difficult problems, such as autonomous underwater vehicle (AUV) navigation relying on acoustic localization, frequently involve recurring outliers. In these situations, the correct hypothesis only emerges as the most likely one when a substantial number of measurements are considered. Robust solutions for these problems need to consider multiple hypotheses simultaneously, which, in turn, calls for the representation of multimodal distributions. In this work, a novel approximate inference method is presented, called the Gaussian mixture sum-product algorithm (GM-SPA), as it implements the sum-product algorithm (SPA) for Gaussian mixtures. The GM-SPA can exactly represent under-constrained linear measurements and approximate important non-linear models, such as range measurements and 2D pose kinematics. The outlier robustness of the GM-SPA is tested and compared against the particle filter (PF) and multimodal incremental smoothing and mapping (MMiSAM), both of which are non-parametric methods. Robustness, accuracy, and run-time are improved in simulation tests. The test problems include 1D localization with unknown data association, 3D linear target tracking with correlated outliers, and 2D range-only pose estimation with Gaussian mixture noise. / Robusta lösningar för tillståndsuppskattning måste kunna hantera felaktiga mätningar, så kallade outliers, och okända dataassociationer, vilket leder till flera möjliga hypoteser. Ta till exempel scenariot att spåra två likadana mål baserat på positionsmätningar, där varje mätning kan tillhöra något av målen eller till och med vara en felaktig avläsning. Vanliga skattningsmetoder modellerar tillståndet som en unimodal fördelning, och kallas därför unimodala metoder. På samma sätt modellerar multimodala metoder tillståndet som en multimodal fördelning. Svåra problem, som navigering av autonoma undervattensfarkoster (AUV) med hjälp av akustisk lokalisering, involverar ofta upprepade outliers. I dessa situationer framstår den korrekta hypotesen som den mest sannolika först när ett stort antal mätningar beaktas. Robusta lösningar för dessa problem måste ta hänsyn till flera hypoteser samtidigt, vilket i sin tur kräver representation av multimodala fördelningar. I detta arbete presenteras en ny approximativ inferensmetod, kallad Gaussian mixture sum-product algorithm (GM-SPA), eftersom den implementerar sum-product algorithm (SPA) för gaussiska blandningar. GM-SPA kan representera underbegränsade linjära mätningar exakt och approximera viktiga icke-linjära modeller, till exempel avståndsmätningar eller 2D-posekinematik. GM-SPA:s robusthet mot outliers testas och jämförs med partikelfiltret (PF) och multimodal incremental smoothing and mapping (MM-iSAM), som båda är icke-parametriska metoder. Robusthet, noggrannhet och körtid förbättras i simuleringstester. Simulerade tester inkluderar 1D-lokalisering med okänd dataassociation, 3D linjär målföljning med korrelerade outliers och 2D-ställningsuppskattning av endast räckvidd med Gaussiskt blandningsljud. / Soluções robustas para estimação de estado devem lidar com medidas defeituosas, chamadas de outliers, e com associações de dados desconhecidas, que levam a múltiplas hipóteses possíveis. Considere-se, por exemplo, o cenário de rastreamento de dois alvos indistinguíveis com base em medidas de posição, em que cada medida pode-se referir a qualquer um dos alvos ou até mesmo ser uma leitura defeituosa. Métodos de estimação comuns modelam o estado como tendo uma distribuição unimodal, sendo assim chamados de métodos unimodais. Da mesma forma, métodos multimodais modelam o estado como uma distribuição multimodal. Problemas difíceis, como a navegação de veículos subaquáticos autónomos (AUVs) baseada em localização acústica, frequentemente envolvem outliers recorrentes. Nestas situações, a hipótese correta apenas surge como a mais provável quando um número substancial de medidas é considerado. Soluções robustas para estes problemas precisam de considerar múltiplas hipóteses simultaneamente, o que, por sua vez, exige a representação de distribuições multimodais. Neste trabalho, é apresentado um novo método de inferência aproximada, chamado Gaussian mixture sum-product algorithm (GM-SPA), pois implementa o sum-product algorithm (SPA) para misturas Gaussianas. O GM-SPA pode representar exatamente medidas lineares sub-determinadas e aproximar modelos não lineares importantes, como medidas de distância e cinemática de pose 2D. A robustez a outliers do GM-SPA é testada e comparada com o filtro de partículas (PF) e com multimodal incremental smoothing and mapping (MM- -iSAM), ambos métodos não-paramétricos. A robustez, a exatidão e o tempo de execução em testes de simulação são melhorados. Os problemas de teste incluem localização 1D com associação de dados desconhecida, rastreamento linear de alvos em 3D com outliers correlacionados e estimação de pose 2D com base em medidas de distância com ruído de mistura Gaussiana.

This thesis offers a probabilistic solution to robust estimation using a novel adaptive estimator. Reliable state estimation is a mandatory prerequisite for autonomous systems interacting with the real world. The presence of outliers challenges the Gaussian assumption of numerous estimation algorithms, resulting in a potentially skewed estimate that compromises reliability. Many approaches attempt to mitigate erroneous measurements by using a robust loss function – which often comes with a trade-off between robustness and numerical stability. The proposed approach is purely probabilistic and enables adaptive large-scale estimation with non-Gaussian error models. The introduced Adaptive Mixture algorithm combines a nonlinear least squares backend with Gaussian mixtures as the measurement error model. Factor graphs as graphical representations allow an efficient and flexible application to real-world problems, such as simultaneous localization and mapping or satellite navigation. The proposed algorithms are constructed using an approximate expectation-maximization approach, which justifies their design probabilistically. This expectation-maximization is further generalized to enable adaptive estimation with arbitrary probabilistic models. Evaluating the proposed Adaptive Mixture algorithm in simulated and real-world scenarios demonstrates its versatility and robustness. A synthetic range-based localization shows that it provides reliable estimation results, even under extreme outlier ratios. Real-world satellite navigation experiments prove its robustness in harsh urban environments. The evaluation on indoor simultaneous localization and mapping datasets extends these results to typical robotic use cases. The proposed adaptive estimator provides robust and reliable estimation under various instances of non-Gaussian measurement errors.

In the past years, Deep Convolutional Neural Networks have proven to be very useful for 2D Object Detection in many applications. These types of networks require large amounts of labeled data, which can be increasingly costly for companies deploying these detectors in practice if the data quality is lacking. Pool-based Active Learning is an iterative process of collecting subsets of data to be labeled by a human annotator and used for training to optimize performance per labeled image. The detectors used in Active Learning cycles are conventionally pre-trained with a small subset, approximately 2% of available data labeled uniformly at random. This is something I challenged in this thesis by using image metadata. With the motivation of many Machine Learning models being a "jack of all trades, master of none", thus it is hard to train models such that they generalize to all of the data domain, it can be interesting to develop a detector for a certain target metadata domain. A simple Monte Carlo method, Rejection Sampling, can be implemented to sample according to a metadata target domain. This would require a target and proposal metadata distribution. The proposal metadata distribution would be a parametric model in the form of a Gaussian Mixture Model learned from the training metadata. The parametric model for the target distribution could be learned in a similar manner, however from a target dataset. In this way, only the training images with metadata most similar to the target metadata distribution can be sampled. This sampling approach was employed and tested with a 2D Object Detector: Faster-RCNN with ResNet-50 backbone. The Rejection Sampling approach was tested against conventional random uniform sampling and a classical Active Learning baseline: Min Entropy Sampling. The performance was measured and compared on two different target metadata distributions that were inferred from a specific target dataset. With a labeling budget of 2% for each cycle, the max Mean Average Precision at 0.5 Intersection Over Union for the target set each cycle was calculated. My proposed approach has a 40 % relative performance advantage over random uniform sampling for the first cycle, and 10% after 9 cycles. Overall, my approach only required 37 % of the labeled data to beat the next best-tested sampler: the conventional uniform random sampling. / De senaste åren har Djupa Neurala Faltningsnätverk visat sig vara mycket användbara för 2D Objektdetektering i många applikationer. De här typen av nätverk behöver stora mängder av etiketterat data, något som kan innebära ökad kostnad för företag som distribuerar dem, om kvaliteten på etiketterna är bristfällig. Pool-baserad Aktiv Inlärning är en iterativ process som innebär insamling av delmängder data som ska etiketteras av en människa och användas för träning, för att optimera prestanda per etiketterat data. Detektorerna som används i Aktiv Inlärning är konventionellt sätt förtränade med en mindre delmängd data, ungefär 2% av all tillgänglig data, etiketterat enligt slumpen. Det här är något jag utmanade i det här arbetet genom att använda bild metadata. Med motiveringen att många Maskininlärningsmodeller presterar sämre på större datadomäner, eftersom det kan vara svårt att lära detektorer stora datadomäner, kan det vara intressant att utveckla en detektor för ett särskild metadata mål-domän. För att samla in data enligt en metadata måldomän, kan en enkel Monte Carlo metod, Rejection Sampling implementeras. Det skulle behövas en mål-metadata-distribution och en faktisk metadata distribution. den faktiska metadata distributionen skulle vara en parametrisk modell i formen av en Gaussisk blandningsmodell som är tränad på träningsdata. Den parametriska modellen för mål-metadata-distributionen skulle kunna vara tränad på liknande sätt, fast ifrån mål-datasetet. På detta sätt, skulle endast träningsbilder med metadata mest lik mål-datadistributionen kunna samlas in. Den här samplings-metoden utvecklades och testades med en 2D objektdetektor: Faster R-CNN med ResNet-50 bildegenskapextraktor. Rejection sampling metoden blev testad mot konventionell likformig slumpmässig sampling av data och en klassisk Aktiv Inlärnings metod: Minimum Entropi sampling. Prestandan mättes och jämfördes mellan två olika mål-metadatadistributioner som var framtagna från specifika mål-metadataset. Med en etiketteringsbudget på 2%för varje cykel, så beräknades medelvärdesprecisionen om 0.5 snitt över union för mål-datasetet. Min metod har 40%bättre prestanda än slumpmässig likformig insamling i första cykeln, och 10 % efter 9 cykler. Överlag behövde min metod endast 37 % av den etiketterade data för att slå den näst basta samplingsmetoden: slumpmässig likformig insamling.