# [diss] Perustieteiden korkeakoulu / SCI

## Permanent URI for this collection

## Browse

### Browsing [diss] Perustieteiden korkeakoulu / SCI by Department "Department of Information and Computer Science"

Now showing 1 - 20 of 59

###### Results Per Page

###### Sort Options

Item Advances in Analysis and Exploration in Medical Imaging(Aalto University, 2014) Gonçalves, Nicolau; Vigário, Ricardo, Doc., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Neuroinformatics Research Group; Neuroinformatiikka; Perustieteiden korkeakoulu; School of Science; Oja, Erkki, Aalto Distinguished Prof., Aalto University, Department of Information and Computer Science, FinlandWith an ever increasing life expectancy, we see a concomitant increase in diseases capable of disrupting normal cognitive processes. Their diagnoses are difficult, and occur usually after daily living activities have already been compromised. This dissertation proposes machine learning methods for the study of the neurological implications of brain lesions. It addresses the analysis and exploration of medical imaging data, with particular emphasis to (f)MRI. Two main research directions are proposed. In the first, a brain tissue segmentation approach is detailed. In the second, a document mining framework, applied to reports of neuroscientific studies, is described. Both directions are based on retrieving consistent information from multi-modal data. A contribution in this dissertation is the application of a semi-supervised method, discriminative clustering, to identify different brain tissues and their partial volume information. The proposed method relies on variations of tissue distributions in multi-spectral MRI, and reduces the need for a priori information. This methodology was successfully applied to the study of multiple sclerosis and age related white matter diseases. It was also showed that early-stage changes of normal-appearing brain tissue can already predict decline in certain cognitive processes. Another contribution in this dissertation is in neuroscience meta-research. One limitation in neuroimage processing relates to data availability. Through document mining of neuroscientific reports, using images as source of information, one can harvest research results dealing with brain lesions. The context of such results can be extracted from textual information, allowing for an intelligent categorisation of images. This dissertation proposes new principles, and a combination of several techniques to the study of published fMRI reports. These principles are based on a number of distance measures, to compare various brain activity sites. Application to studies of the default mode network validated the proposed approach. The aforementioned methodologies rely on clustering approaches. When dealing with such strategies, most results depend on the choice of initialisation and parameter settings. By defining distance measures that search for clusters of consistent elements, one can estimate a degree of reliability for each data grouping. In this dissertation, it is shown that such principles can be applied to multiple runs of various clustering algorithms, allowing for a more robust estimation of data agglomeration.Item Advances in Extreme Learning Machines(Aalto University, 2015) van Heeswijk, Mark; Miche, Yoan, Dr., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Environmental and Industrial Machine Learning Group; Ympäristön ja teollisuuden alojen koneoppiminen; Perustieteiden korkeakoulu; School of Science; Oja, Erkki, Aalto Distinguished Prof., Aalto University, Department of Information and Computer Science, FinlandNowadays, due to advances in technology, data is generated at an incredible pace, resulting in large data sets of ever-increasing size and dimensionality. Therefore, it is important to have efficient computational methods and machine learning algorithms that can handle such large data sets, such that they may be analyzed in reasonable time. One particular approach that has gained popularity in recent years is the Extreme Learning Machine (ELM), which is the name given to neural networks that employ randomization in their hidden layer, and that can be trained efficiently. This dissertation introduces several machine learning methods based on Extreme Learning Machines (ELMs) aimed at dealing with the challenges that modern data sets pose. The contributions follow three main directions. Firstly, ensemble approaches based on ELM are developed, which adapt to context and can scale to large data. Due to their stochastic nature, different ELMs tend to make different mistakes when modeling data. This independence of their errors makes them good candidates for combining them in an ensemble model, which averages out these errors and results in a more accurate model. Adaptivity to a changing environment is introduced by adapting the linear combination of the models based on accuracy of the individual models over time. Scalability is achieved by exploiting the modularity of the ensemble model, and evaluating the models in parallel on multiple processor cores and graphics processor units. Secondly, the dissertation develops variable selection approaches based on ELM and Delta Test, that result in more accurate and efficient models. Scalability of variable selection using Delta Test is again achieved by accelerating it on GPU. Furthermore, a new variable selection method based on ELM is introduced, and shown to be a competitive alternative to other variable selection methods. Besides explicit variable selection methods, also a new weight scheme based on binary/ternary weights is developed for ELM. This weight scheme is shown to perform implicit variable selection, and results in increased robustness and accuracy at no increase in computational cost. Finally, the dissertation develops training algorithms for ELM that allow for a flexible trade-off between accuracy and computational time. The Compressive ELM is introduced, which allows for training the ELM in a reduced feature space. By selecting the dimension of the feature space, the practitioner can trade off accuracy for speed as required. Overall, the resulting collection of proposed methods provides an efficient, accurate and flexible framework for solving large-scale supervised learning problems. The proposed methods are not limited to the particular types of ELMs and contexts in which they have been tested, and can easily be incorporated in new contexts and models.Item Advances in Mining Binary Data: Itemsets as Summaries(Helsinki University of Technolocy, Department of Information and Computer Science, 2008) Tatti, Nikolaj; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of ScienceMining frequent itemsets is one of the most popular topics in data mining. Itemsets are local patterns, representing frequently cooccurring sets of variables. This thesis studies the use of itemsets to give information about the whole dataset. We show how to use itemsets for answering queries, that is, finding out the number of transactions satisfying some given formula. While this is a simple procedure given the original data, the task transforms into a computationally infeasible problem if we seek the solution using the itemsets. By making some assumptions of the structure of the itemsets and applying techniques from the theory of Markov Random Fields we are able to reduce the computational burden of query answering. We can also use the known itemsets to predict the unknown itemsets. The difference between the prediction and the actual value can be used for ranking itemsets. In fact, this method can be seen as generalisation for ranking itemsets based on their deviation from the independence model, an approach commonly used in the data mining literature. The next contribution is to use itemsets to define a distance between the datasets. We achieve this by computing the difference between the frequencies of the itemsets. We take into account the fact that the itemset frequencies may be correlated and by removing the correlation we show that our distance transforms into Euclidean distance between the frequencies of parity formulae. The last contribution concerns calculating the effective dimension of binary data. We apply fractal dimension, a known concept that works well with realvalued data. Applying fractal dimension dimension directly is problematic because of the unique nature of binary data. We propose a solution to this problem by introducing a new concept called normalised correlation dimension. We study our approach theoretically and empirically by comparing it against other methods.Item Advances in Nonnegative Matrix Decomposition with Application to Cluster Analysis(Aalto University, 2014) Zhang, He; Yang, Zhirong, Dr., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Oja, Erkki, Aalto Distinguished Prof., Aalto University, Department of Information and Computer Science, FinlandNonnegative Matrix Factorization (NMF) has found a wide variety of applications in machine learning and data mining. NMF seeks to approximate a nonnegative data matrix by a product of several low-rank factorizing matrices, some of which are constrained to be nonnegative. Such additive nature often results in parts-based representation of the data, which is a desired property especially for cluster analysis. This thesis presents advances in NMF with application in cluster analysis. It reviews a class of higher-order NMF methods called Quadratic Nonnegative Matrix Factorization (QNMF). QNMF differs from most existing NMF methods in that some of its factorizing matrices occur twice in the approximation. The thesis also reviews a structural matrix decomposition method based on Data-Cluster-Data (DCD) random walk. DCD goes beyond matrix factorization and has a solid probabilistic interpretation by forming the approximation with cluster assigning probabilities only. Besides, the Kullback-Leibler divergence adopted by DCD is advantageous in handling sparse similarities for cluster analysis. Multiplicative update algorithms have been commonly used for optimizing NMF objectives, since they naturally maintain the nonnegativity constraint of the factorizing matrix and require no user-specified parameters. In this work, an adaptive multiplicative update algorithm is proposed to increase the convergence speed of QNMF objectives. Initialization conditions play a key role in cluster analysis. In this thesis, a comprehensive initialization strategy is proposed to improve the clustering performance by combining a set of base clustering methods. The proposed method can better accommodate clustering methods that need a careful initialization such as the DCD. The proposed methods have been tested on various real-world datasets, such as text documents, face images, protein, etc. In particular, the proposed approach has been applied to the cluster analysis of emotional data.Item Advances in the theory of nearest neighbor distributions(Aalto-yliopiston teknillinen korkeakoulu, 2010) Liitiäinen, Elia; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Aalto-yliopiston teknillinen korkeakouluA large part of non-parametric statistical techniques are in one way or another related to the geometric properties of random point sets. This connection is present both in the design of estimators and theoretical convergence studies. One such relation between geometry and probability occurs in the application of non-parametric techniques for computing information theoretic entropies: it has been shown that the moments of the nearest neighbor distance distributions for a set of independent identically distributed random variables are asymptotically characterized by the Rényi entropies of the underlying probability density. As entropy estimation is a problem of major importance, this connection motivates an extensive study of nearest neighbor distances and distributions. In this thesis, new results in the theory of nearest neighbor distributions are derived using both geometric and probabilistic proof techniques. The emphasis is on results that are useful for finite samples and not only in the asymptotic limit of an infinite sample. Previously, in the literature it has been shown that after imposing sufficient regularity assumptions, the moments of the nearest neighbor distances can be approximated by invoking a Taylor series argument providing the connection to the Rényi entropies. However, the theoretical results provide limited understanding to the nature of the error in the approximation. As a central result of the thesis, it is shown that if the random points take values in a compact set (e.g. according to the uniform distribution), then under sufficient regularity, a higher order moment expansion is possible. Asymptotically, the result completely characterizes the error for the original low order approximation. Instead of striving for exact computation of the moments through a Taylor series expansion, in some cases inequalities are more useful. In the thesis, it is shown that concrete upper and lower bounds can be established under general assumptions. In fact, the upper bounds rely only on a geometric analysis. The thesis also contains applications to two problems in nonparametric statistics, residual variance and Rényi entropy estimation. A well-established nearest neighbor entropy estimator is analyzed and it is shown that by taking the boundary effect into account, estimation bias can be significantly reduced. Secondly, the convergence properties of a recent residual variance estimator are analyzed.Item Advances in Wireless Damage Detection for Structural Health Monitoring(Aalto University, 2014) Toivola, Janne; Hollmén, Jaakko, Dr., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Parsimonious Modelling; Perustieteiden korkeakoulu; School of Science; Rousu, Juho, Prof., Aalto University, Department of Information and Computer Science, FinlandOne of the fundamental tasks in structural health monitoring is to extract relevant information about a structure, such as a bridge or a crane, and reach statistical decisions about the existence of damages in the structure. Recent advances in wireless sensor network technology has offered new possibilities for acquiring and processing structural health monitoring data automatically. The purpose of this dissertation is to explore various data processing methods for detecting previously unobserved deviation in measurements from accelerometer sensors, based on natural vibration of structures. Part of the processing is projected to be performed on resource constrained wireless sensors to ultimately reduce the cost of measurements. Data processing in the proposed detection systems is divided into following general stages: feature extraction, dimensionality reduction, novelty detection, and performance assessment. Several methods in each of the stages are proposed and benchmarked in offline experiments with multiple accelerometer data sets. The methods include, for example, the Goertzel algorithm, random projection, tensor decomposition, collaborative filtering, nearest neighbor classification, and evaluating detection accuracy in terms of receiver operating characteristic curves. Significant reductions are achieved in the amount of data transmitted from sensors and input to statistical classifiers, while maintaining some of the classification accuracy. However, the sensitivity and specificity in detection are worse than those of centralized methods operating on raw sensor data. The work proposed and evaluated several combinations of data processing stages for wireless damage detection. While better than random detection accuracy can be achieved with very small amount of data per accelerometer sensor, challenges remain in reaching specificity required in practical applications.Item Bayesian Latent Gaussian Spatio-Temporal Models(Aalto University, 2015) Luttinen, Jaakko; Ilin, Alexander, Dr., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Karhunen, Juha, Prof., Aalto University, Department of Information and Computer Science, FinlandThis work develops efficient Bayesian methods for modelling large spatio-temporal datasets. The main challenge is in constructing methods that can both capture complex structure and scale to large datasets. To achieve this, the developed methods use model structures which are flexible but enable efficient learning algorithms. The contributed methods are based on widely used linear latent variable models and Gaussian processes. The contributions of the thesis can be summarized as follows: Efficient linear latent variable models are extended to handle complex temporal dynamics and spatial structure. The high computational cost of Gaussian processes is significantly reduced by particular model formulations which allow computations to be performed separately in the spatial and temporal domains. Missing values have been taken into account and the modelling of outliers has been studied, because real-world datasets are often of poor quality. The thesis also develops procedures to speed up the standard approximate Bayesian inference algorithms significantly. The discussed methods are applicable to spatio-temporal datasets which are a collection of measurements obtained from a set of sensors at a set of time instances. The thesis focuses on datasets from climate processes by using the proposed methods to reconstruct historical sea surface and air temperature datasets, to denoise a badly corrupted weather dataset and to learn the dynamics of physical processes. The methods are also applicable to many other physical processes, brain imaging and typical multi-task problems. The goal of the modelling can be to predict measurements, interpolate spatial fields, reconstruct missing values, extract interesting features or remove noise.Item Bayesian latent variable models for learning dependencies between multiple data sources(Aalto University, 2014) Virtanen, Seppo; Klami, Arto, Dr., Helsinki Institute for Information Technology, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Statistical Machine Learning and Bioinformatics Group; Tilastollinen koneoppiminen ja bioinformatiikka; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Prof., Aalto University, Department of Information and Computer Science, FinlandMachine learning focuses on automated large-scale data analysis extracting useful information from data collections. The data are frequently high-dimensional and may correspond, for example, to images, text documents, or measurements of neural responses. In many applications data can be collected from multiple data sources, that is, views. This thesis presents novel machine learning methods for analyzing multiple data sources, especially for understanding relationships between them. The analysis provides a comprehensive summary of the data generating process, which may be used for exploring the relationships and for predicting observations of one or more sources. The methods are based on two assumptions: each view provides complementary information of the data generating process, and each view is corrupted by noise. The methods aim to utilize all available information (views), accumulating partly overlapping information and reducing view-specific noise. In particular, this thesis presents several Bayesian latent variable models that learn a decomposition of latent variables; some of the variables capture information shared by multiple sources, whereas the remaining variables explain noise in each view. The latent variables may be efficiently inferred based on the observed data by using sparsity assumptions and Bayesian inference. The models are applied for analyzing neural responses to natural stimulation as well as for jointly modeling images and text documents.Item Bayesian Multi-Way Models for Data Translation in Computational Biology(Aalto University, 2014) Suvitaival, Tommi; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Prof., Aalto University, Department of Information and Computer Science, FinlandThe inference of differences between samples is a fundamental problem in computational biology and many other sciences. Hypothesis about a complex system can be studied via a controlled experiment. The design of the controlled experiment sets the conditions, or covariates, for the system in such a way that their effect on the system can be studied through independent measurements. When the number of measured variables is high and the variables are correlated, the assumptions of standard statistical methods are no longer valid. In this thesis, computational methods are presented to this problem and its follow-up problems. A similar experiment done on different systems, such as multiple biological species, leads to multiple "views" of the experiment outcome, observed in different data spaces or domains. However, cross-domain experimentation brings uncertainty about the similarity of the systems and their outcomes. Thus, a new question emerges: which of the covariate effects generalize across the domains? In this thesis, novel computational methods are presented for the integration of data views, in order to detect weaker covariate effects and to generalize covariate effects to views with unobserved data. Five main contributions to the inference of covariate effects are presented: (1) When the data are high-dimensional and collinear, the problem of false discovery is curbed by assuming a cluster structure on the observed variables and by handling the uncertainty with Bayesian methods. (2) Prior information about the measurement process can be used to further improve the inference of covariate effects for metabolomic experiments by modeling the multiple layers of uncertainty in the mass spectral data. (3-4) When the data come from multiple measurement sources on the same subjects - that is, from data views with co-occurring samples - it is unknown, whether the covariate effects generalize across the views and whether the outcome of a new intervention can be generalized to a view with no observed data on that intervention. These problems are shown to be possible to solve by assuming a shared generative process for the multiple data views. (5) When the data come from different domains with no co-occurring samples, the inference of between-domain dependencies is not possible in the same way as with co-occurring samples. It is shown that even in this situation, it is possible to identify covariate effects that generalize across the domains, when the experimental design at least weakly binds the domains together. Then, effects that generalize are identified by assuming a shared generative process for the covariate effects.Item Computational Methods for Analysis of Dynamic Transcriptome and Its Regulation Through Chromatin Remodeling and Intracellular Signaling(Aalto University, 2014) Äijö, Tarmo; Lähdesmäki, Harri, Prof., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Lähdesmäki, Harri, Prof., Aalto University, Department of Information and Computer Science, FinlandTranscription is the first step in gene expression in which genetic information is transferred from DNA to RNA. Gene expression is highly controlled through transcriptional regulation at many steps. Transcriptional regulation in eukaryotes occurs, e.g., through binding of transcription factors and chromatin remodeling via various epigenetic pathways. Additionally, dysregulated transcription has been reported in various diseases. Thus, transcription and transcriptional regulation are of great interest for research. In this work, we study the transcriptome and its regulation using bioinformatic and computational biology approaches. We propose computational methods, LIGAP and DyNB, for analysis of temporal gene expression profiles measured using microarrays and RNA-seq, respectively. LIGAP is a methodology based on Gaussian processes for simultaneous differential expression analysis between an arbitratory number of time series microarray data sets. DyNB, is an extension of the Gaussian-Cox process in which the Poisson distribution is replaced by the negative binomial distribution. Additionally, DyNB enables the study of systematic differences, such as differential differentiation efficiencies, between conditions. Sorad, is a modeling framework based on differential equations and Gaussian processes for analysis of intracellular signaling transduction through phosphoprotein activities. We also propose and demonstrate how the in silico models inferred using Sorad can be used in estimating modulation strategies to obtain desired signaling response. Finally, we study the determinants of nucleosome positioning and subsequent effects on gene expression. All the proposed methods are benchmarked against existing methods and, in addition, they are applied to real-life problems. The comparison studies validate the applicability of the presented methods and demonstrate their improved performance relative to existing methods. Our transcriptome studies led to increased knowledge on the early differentiation of human T cells, and provided a valuable resource of candidate genes for future functional studies of the differentiation process. Our nucleosome study revealed that within loci important for T cell differentiation only 6% of the nucleosomes are differentially remodelled between T helper 1 and 2 cells and cytotoxic T lymphocytes. The remodelled nucleosomes correlated with the known differentiation program, chromatin accessibility, transcription factor binding, and gene expression. Finally, our data supports the hypothesis that transcription factors and nucleosomes compete for DNA occupancy.Item Computational methods for comparison and exploration of event sequences(Aalto University, 2013) Lijffijt, Jefrey; Mannila, Heikki, Prof., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Rousu, Juho, Prof., Aalto University, Department of Information and Computer Science, FinlandMany types of data, e.g., natural language texts, biological sequences, or time series of sensor data, contain sequential structure. Analysis of such sequential structure is interesting for various reasons, for example, to detect that data consists of several homogeneous parts, that data contains certain recurring patterns, or to find parts that are different or surprising compared to the rest of the data. The main question studied in this thesis is how to identify global and local patterns in event sequences. Within this broad topic, we study several subproblems. The first problem that we address is how to compare event frequencies across event sequences and databases of event sequences. Such comparisons are relevant, for example, to linguists who are interested in comparing word counts between two corpora to identify linguistic differences, e.g., between groups of speakers, or language change over time. The second problem that we address is how to find areas in an event sequence where an event has a surprisingly high or low frequency. More specifically, we study how to take into account the multiple testing problem when looking for local frequency deviations in event sequences. Many algorithms for finding local patterns in event sequences require that the person applying the algorithm chooses the level of granularity at which the algorithm operates, and it is often not clear how to choose that level. The third problem that we address is which granularities to use when looking for local patterns in an event sequence. The main contributions of this thesis are computational methods that can be used to compare and explore (databases of) event sequences with high computational efficiency, increased accuracy, and that offer new perspectives on the sequential structure of data. Furthermore, we illustrate how the proposed methods can be applied to solve practical data analysis tasks, and describe several experiments and case studies where the methods are applied on various types of data. The primary focus is on natural language texts, but we also study DNA sequences and sensor data. We find that the methods work well in practice and that they can efficiently uncover various types of interesting patterns in the data.Item Computational Modeling and Simulation of Language and Meaning: Similarity-Based Approaches(Aalto University, 2014) Lindh-Knuutila, Tiina; Honkela, Timo, Prof., Aalto University, Department of Information and Computer Science, Finland; Creutz, Mathias, Dr., Aalto University, Department of Information and Computer Science, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Oja, Erkki, Aalto Distinguished Professor, Aalto University, FinlandThis dissertation covers various similarity-based, data-driven approaches to model language and lexical semantics. The availability of large amounts of text data in electronic form allows the use of unsupervised, data-driven methodologies. Compared to linguistic models based on expert knowledge, which are often costly or unavailable, the data-driven analysis is faster and more flexible. The same methodologies can be often used regardless of the language. In addition, data-driven analysis may be exploratory and offer a new view on the data. The complexity of different European languages was analyzed at syntactic and morphological level using unsupervised methods based on compression and unsupervised morphology induction. The results showed that the unsupervised methods are able to produce useful analyses that correspond to linguistic models. The distributional word vector space models represent the meaning of words in a text context of co-occurring words, collected from a large corpus. The vector space models were evaluated with linguistic models and human semantic similarity judgment data. Two unsupervised methods, Independent Component Analysis and Latent Dirichlet Allocation, were able to find groups of semantically similar words, corresponding reasonably well to the evaluation sets. In addition to validating the results of the unsupervised methods with the evaluation data, the research was also exploratory. The unsupervised methods found semantic word sets not covered by the evaluation set, and the analysis of the categories of the evaluation sets showed quality differences between the categories. In the agent simulation models, the meaning of words was directly linked to the perceived context of the agent. Each agent had a subjective conceptual memory, in which the associations between words and perceptions were formed. In a population of simulated agents, the emergence of a shared vocabulary was studied through simulated language games. As a result of the simulations, a shared vocabulary emerges in the community.Item Covert timing channels, caching, and cryptography(Aalto University, 2011) Brumley, Billy Bob; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; Nyberg, Kaisa, Prof.Side-channel analysis is a cryptanalytic technique that targets not the formal description of a cryptographic primitive but the implementation of it. Examples of side-channels include power consumption or timing measurements. This is a young but very active field within applied cryptography. Modern processors are equipped with numerous mechanisms to improve the average performance of a program, including but not limited to caches. These mechanisms can often be used as side-channels to attack software implementations of cryptosystems. This area within side-channel analysis is called microarchitecture attacks, and those dealing with caching mechanisms cache-timing attacks. This dissertation presents a number of contributions to the field of side-channel analysis. The introductory portion consists of a review of common cache architectures, a literature survey of covert channels focusing mostly on covert timing channels, and a literature survey of cache-timing attacks, including selective related results that are more generally categorized as side-channel attacks such as traditional timing attacks. This dissertation includes eight publications relating to this field. They contain contributions in areas such as side-channel analysis, data cache-timing attacks, instruction cache-timing attacks, traditional timing attacks, and fault attacks. Fundamental themes also include attack mitigations and efficient yet secure software implementation of cryptosystems. Concrete results include, but are not limited to, four practical side-channel attacks against OpenSSL, each implemented and leading to full key recovery.Item Data-driven Analysis for Natural Studies in Functional Brain Imaging(Aalto University, 2013) Ylipaavalniemi, Jarkko; Vigário, Ricardo, Doc., Aalto University, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Oja, Erkki, Prof., Aalto University, FinlandIn neuroscience, functional magnetic resonance imaging (fMRI) has become a powerful tool in human brain mapping. Typically, fMRI is used with a rather simple stimulus sequence, aiming at improving signal-to-noise ratio for statistical hypothesis testing. When natural stimuli are used, the simple designs are no longer appropriate. The aim of this thesis is in developing data-driven approaches for reliable inference of brain correlates to natural stimuli. Since the beginning of the nineteenth century, neuroscience has focused on the idea that distinct regions of the brain support particular mental processes. However, modern research recognizes that many functions rely on distributed networks, and that a single brain region may participate in more than one function. These rapid paradigm changes in neuroscience raise important methodological challenges. Purely hypothesis-driven methods have been used extensively in functional imaging studies. As the focus in brain research is shifting away from functional specialization towards interaction-based functional networks, those approaches are no longer appropriate. In contrast to the classic statistical hypothesis testing approaches, modern machine learning methods allow for a purely data-driven way to describe the data. They do not use the stimuli, and make no assumptions about whether the brain processes are stimulus related or not. The recordings for each brain region may contain a complicated mixture of activity, which is produced by many spatially distributed processes, and artifacts. Each process can be described as a component having a separate time series and spatial extent, and producing simultaneous changes in the fMRI signals of many regions. The main contribution of the thesis is a reliable independent component analysis (ICA) approach, which is available in the Arabica toolbox. The usefulness of the approach was tested extensively with fMRI data, showing that the method is capable of providing insights into the data that would not be attainable otherwise. The new method was also theoretically analyzed and its asymptotic convergence was proven. The theory offers a thorough explanation of how the method works and justifies its use in practice. Then, the new method is further developed for analyzing networks of distributed brain activity, by combining it with canonical correlation analysis (CCA). The extension was shown to be particularly useful with fMRI studies that use natural stimuli. The approach is further extended to be applicable in cases where independent subspaces emerge, which often happens when using real measurement data that is not guaranteed to fit all the assumptions made in the development of the methods.Item Developing augmented reality solutions through user involvement(VTT Technical Research Centre of Finland, 2015) Siltanen, Sanni; Tossavainen, Timo, Dr., ZenRobotics Ltd, Helsinki, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Oja, Erkki, Emeritus Prof., Aalto University, Department of Computer Science, FinlandAugmented reality (AR) technology merges digital information into the real world. It is an effective visualization method; AR enhances user's spatial perception skills and helps to understand spatial dimensions and relationships. It is beneficial for many professional application areas such as assembly, maintenance and repair. AR visualization helps to concretize building and construction projects and interior design plans – also for non-technically oriented people, who might otherwise have difficulties in understanding what the plans actually mean in the real context. Due to its interactive and immersive nature AR is applied for games and advertising as well. Although AR is proven to be a valuable visualization method it is not yet commonly used in beneficial consumer level applications. This work first finds out reasons for this and then focuses on developing AR towards wider use. The work is threefold: it considers human factors affecting adoption of the technology, economic factors affecting the viability of AR technology, and development of applications and technical solutions that support these factors. In this thesis user centric and participatory methods are used to find out reasons that hinder the use of AR, especially in interior design. The outcomes of the studies are manifold: desired features for AR services, bottlenecks preventing the use, user experience (UX) issues and business viability factors. A successful AR solution needs to have a viable business ecosystem besides a reliable technical framework. The presented application development in assembly guidance and interior design visualization considers UX factors and demonstrates the use of AR in the field of question. A serious bottleneck for using AR in interior design arises from a typical use situation; a consumer wants to redesign a room. The space where the interior design plan is made is not empty and augmentation does not look realistic when the new furniture is rendered on top of the existing furniture. This problem can be solved by using diminished reality, which means that the old furniture is removed digitally from the view. This work presents a diminished reality solution for AR interior design. A complete pipeline implementing diminished reality functionality is described. Algorithms and methods are developed to achieve real time high quality diminished reality functionality. The presented practical solution has a great effect for the whole AR interior design field, and enhances it towards real use. The possibilities of using AR are huge. In order to make beneficial AR solutions, researchers should be able to reveal the users' needs – both existing and emerging ones – and develop technology to fulfil those needs. This thesis demonstrates that this can be achieved by developing augmented reality solutions through user involvement.Item Developing fast machine learning techniques with applications to steganalysis problems(Aalto University, 2010) Miche, Yoan; Lendasse, Amaury; Bas, Patrick; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Simula, OlliIn the history of human communication, the concept and need for secrecy between the parties has always been present. One way of achieving it is to modify the message so that it is readable only by the receiver, as in cryptography for example. Hiding the message in an innocuous medium is another, called steganography. And the counterpart to steganography, that is, discovering whether a message is hidden in a specific medium, is called steganalysis. Other concerns also fall within the broad scope of the term steganalysis, such as estimating the message length for example (which is quantitative steganalysis). In this dissertation, the emphasis is put on classical steganalysis of images first — the mere detection of a modified image — for which a practical benchmark is proposed: the evaluation of a sufficient amount of samples to perform the steganalysis in a statistically significant manner, followed by feature selection for dimensionality reduction and interpretability. The fact that most of the features used in the classical steganalysis task have a physical meaning, regarding the image, lends itself to an introspection and analysis of the selected features for understanding the functioning and weaknesses of steganographic schemes. This approach is computationally demanding, both because of the feature selection and the size of the data in steganalysis problems. To address this issue, a fast and efficient machine learning model is proposed, the Optimally-Pruned Extreme Learning Machine (OP-ELM). It uses random projections in the framework of an Artificial Neural Network (precisely, a Single Layer Feedforward Network) along with a neuron selection strategy, to obtain robustness regarding irrelevant features, and achieves state of the art performances. The OP-ELM is also used in a novel approach at quantitative steganalysis (message length estimation). The re-embedding concept is proposed, which embeds a new known message in a suspicious image. By repeating this operation multiple times for varying sizes of the newly embedded message, it is possible to estimate the original message size used by the sender, along with a confidence interval on this value. An intrinsic property of the image, the inner difficulty, is also revealed thanks to the confidence interval width; this gives an important information about the reliability of the estimation on the original message size.Item Distributed optimization algorithms for multihop wireless networks(Aalto-yliopiston teknillinen korkeakoulu, 2010) Schumacher, André; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Aalto-yliopiston teknillinen korkeakoulu; Orponen, Pekka, Prof.Recent technological advances in low-cost computing and communication hardware design have led to the feasibility of large-scale deployments of wireless ad hoc and sensor networks. Due to their wireless and decentralized nature, multihop wireless networks are attractive for a variety of applications. However, these properties also pose significant challenges to their developers and therefore require new types of algorithms. In cases where traditional wired networks usually rely on some kind of centralized entity, in multihop wireless networks nodes have to cooperate in a distributed and self-organizing manner. Additional side constraints, such as energy consumption, have to be taken into account as well. This thesis addresses practical problems from the domain of multihop wireless networks and investigates the application of mathematically justified distributed algorithms for solving them. Algorithms that are based on a mathematical model of an underlying optimization problem support a clear understanding of the assumptions and restrictions that are necessary in order to apply the algorithm to the problem at hand. Yet, the algorithms proposed in this thesis are simple enough to be formulated as a set of rules for each node to cooperate with other nodes in the network in computing optimal or approximate solutions. Nodes communicate with their neighbors by sending messages via wireless transmissions. Neither the size nor the number of messages grows rapidly with the size of the network. The thesis represents a step towards a unified understanding of the application of distributed optimization algorithms to problems from the domain of multihop wireless networks. The problems considered serve as examples for related problems and demonstrate the design methodology of obtaining distributed algorithms from mathematical optimization methods.Item The Effects of Mobility on Mobile Input(Aalto University, 2014) Bergström-Lehtovirta, Joanna; Oulasvirta, Antti, Prof., Aalto University, Department of Communication and Networking, Finland; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; School of Science; Kaski, Samuel, Prof., Aalto University and University of Helsinki, Helsinki Institute for Information Technology HIIT, FinlandMobile interfaces are designed for interaction while the user is on the move across mobile contexts. Walking, handling a wallet, visually attending to the environment, and even simply carrying a mobile device, however, can have a negative effect on mobile human–computer interaction (HCI). Understanding the negative effects of mobility is important because potential exists for overcoming them via good interface design. Previous work has shown that mobility decreases input performance with a mobile interface. However, the causes of declines in performance often remain unclear and, with them, possible avenues for compensation. I argue that systematic variation of physical constraints emerging from mobile conditions in controlled experiments can reveal considerable effects of mobility on mobile input. Among these physical constraints are competing allocations of hand function, body movement, and various sensory modalities. The thesis contributes to mobile HCI research by examining the effects of four causes of physical constraints in mobility: 1) gripping of the device, 2) walking, 3) manipulation of external objects, and 4) sensory feedback. The research includes four studies, isolating one constraint each in controlled experiments. In the first two, the levels of grip position and walking speed are varied systematically, for modeling of their effects on mobile input. The other two vary the presence of external objects and sensory feedback. The findings highlight the constraints' negative effect on manual input performance. However, all of the studies also reveal unaffected sensory or motor resources of the user. Across the four studies, the following findings were made. First, a model for the functional area of the thumb predicts the reachable interface elements on a mobile touchscreen as a function of grip, hand size, and screen size. Secondly, a function describing the tradeoff between walking speed and input performance demonstrates that while walking hampers input performance, users can adjust to an optimal walking speed for mobile interaction. Thirdly, interface design is shown to significantly affect input performance when the user simultaneously manipulates external objects or when sensory feedback is limited. This work calls for mobile HCI research to consider the operationalization of mobile conditions in controlled experiments that can extend knowledge of the effects of mobility on interaction. It also invites exploitation of the empirical results and the proposed methods and models in practice for interface evaluation and design.Item Efficient symbolic model checking of concurrent systems(Aalto University, 2011) Dubrovin, Jori; Junttila, Tommi, Dr.; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Perustieteiden korkeakoulu; Niemelä, Ilkka, Prof.Design errors in software systems consisting of concurrent components are potentially disastrous, yet notoriously difficult to find by testing. Therefore, more rigorous analysis methods are gaining popularity. Symbolic model checking techniques are based on modeling the behavior of the system as a formula and reducing the analysis problem to symbolic manipulation of formulas by computational tools. In this work, the aim is to make symbolic model checking, in particular bounded model checking, more efficient for verifying and falsifying safety properties of highly concurrent system models with high-level data features. The contributions of this thesis are divided to four topics. The first topic is symbolic model checking of UML state machine models. UML is a language widely used in the industry for modeling software-intensive systems. The contribution is an accurate semantics for a subset of the UML state machine language and an automatic translation to formulas, enabling symbolic UML model checking. The second topic is bounded model checking of systems with queues. Queues are frequently used to model, for example, message buffers in distributed systems. The contribution is a variety of ways to encode the behavior of queues in formulas that exploit the features of modern SMT solver tools. The third topic is symbolic partial order methods for accelerated model checking. By exploiting the inherent independence of the components of a concurrent system, the executions of the system are compressed by allowing several actions in different components to occur at the same time. Making the executions shorter increases the performance of bounded model checking. The contribution includes three alternative partial order semantics for compressing the executions, with analytic and experimental evaluation. The work also presents a new variant of bounded model checking that is based on a concurrent instead of sequential view of the events that constitute an execution. The fourth topic is efficient computation of predicate abstraction. Predicate abstraction is a key technique for scalable model checking, based on replacing the system model by a simpler abstract model that omits irrelevant details. In practice, constructing the abstract model can be computationally expensive. The contribution is a combination of techniques that exploit the structure of the underlying system to partition the problem into a sequence of cheaper abstraction problems, thus reducing the total complexity.Item Extending data mining techniques for frequent pattern discovery : trees, low-entropy sets, and crossmining(Aalto-yliopiston teknillinen korkeakoulu, 2010) Heikinheimo, Hannes; Mannila, Heikki, Prof.; Tietojenkäsittelytieteen laitos; Department of Information and Computer Science; Aalto-yliopiston teknillinen korkeakoulu; Mannila, Heikki, Prof.The idea of frequent pattern discovery is to find frequently occurring events in large databases. Such data mining techniques can be useful in various domains. For instance, in recommendation and e-commerce systems frequently occurring product purchase combinations are essential in user preference modeling. In the ecological domain, patterns of frequently occurring groups of species can be used to reveal insight into species interaction dynamics. Over the past few years, most frequent pattern mining research has concentrated on efficiency (speed) of mining algorithms. However, it has been argued within the community that while efficiency of the mining task is no longer a bottleneck, there is still an urgent need for methods that derive compact, yet high quality results with good application properties. The aim of this thesis is to address this need. The first part of the thesis discusses a new type of tree pattern class for expressing hierarchies of general and more specific attributes in unstructured binary data. The new pattern class is shown to have advantageous properties, and to discover relationships in data that cannot be expressed alone with the more traditional frequent itemset or association rule patterns. The second and third parts of the thesis discuss the use of entropy as a score measure for frequent pattern mining. A new pattern class is defined, low-entropy sets, which allow to express more general types of occurrence structure than with frequent itemsets. The concept can also be easily applied to tree types of pattern. Furthermore, by applying minimum description length in pattern selection for low-entropy sets it is shown experimentally that in most cases the collections of selected patterns are much smaller than by using frequent itemsets. The fourth part of the thesis examines the idea of crossmining itemsets, that is, relating itemsets to numerical variables in a database of mixed data types. The problem is formally defined and turns out to be NP-hard, although it is approximately solvable within a constant-factor of the optimum solution. Experiments show that the algorithm finds itemsets that convey structure in both the binary and the numerical part of the data.

- «
- 1 (current)
- 2
- 3
- »