Showing papers by "Helsinki University of Technology published in 2016"

Ladder Variational Autoencoders

Abstract: Variational Autoencoders are powerful models for unsupervised learning. However deep models with several layers of dependent stochastic variables are difficult to train which limits the improvements obtained using these highly expressive models. We propose a new inference model, the Ladder Variational Autoencoder, that recursively corrects the generative distribution by a data dependent approximate likelihood in a process resembling the recently proposed Ladder Network. We show that this model provides state of the art predictive log-likelihood and tighter log-likelihood lower bound compared to the purely bottom-up inference in layered Variational Autoencoders and other generative models. We provide a detailed analysis of the learned hierarchical latent representation and show that our new inference model is qualitatively different and utilizes a deeper more distributed hierarchy of latent variables. Finally, we observe that batch normalization and deterministic warm-up (gradually turning on the KL-term) are crucial for training variational models with many stochastic layers.

Giant Magnetostrictive Materials

Abstract: The magnetostrictive materials exhibit a strain caused by the orientation of the magnetic moment when exposed to a magnetic field. A particular class of magnetostrictive materials is called magnetic shape memory (MSM) alloys or ferromagnetic shape memory alloy (FSMA) materials, which can change their shape remarkably when subjected to magnetic field. In the following, the MSM materials are introduced and their extraordinary structure, properties, and performance are described.

The Water-Energy-Food Nexus and the Transboundary Context: Insights from Large Asian Rivers

Abstract: The water-energy-food nexus is a topical subject for research and practice, reflecting the importance of these sectors for humankind and the complexity and magnitude of the challenges they are facing. While the nexus as a concept is not yet mature or fully tested in practice, it has already encouraged a range of approaches in a variety of contexts. This article provides a set of definitions recognizing three perspectives that see the nexus as an analytical tool, governance framework and as an emerging discourse. It discusses the implications that an international transboundary context brings to the nexus and vice versa. Based on a comparative analysis of three Asian regions—Central Asia, South Asia and the Mekong Region—and their related transboundary river basins, we propose that the transboundary context has three major implications: diversity of scales and perspectives, importance of state actors and importance of politics. Similarly, introducing the nexus as an approach in a transboundary context has a potential to provide new resources and approaches, alter existing actor dynamics and portray a richer picture of relationships. Overall, the significance of water-energy-food linkages and their direct impacts on water allocation mean that the nexus has the potential to complement existing approaches also in the transboundary river basins.

Likelihood-free inference by ratio estimation

Abstract: We consider the problem of parametric statistical inference when likelihood computations are prohibitively expensive but sampling from the model is possible Several so-called likelihood-free methods have been developed to perform inference in the absence of a likelihood function The popular synthetic likelihood approach infers the parameters by modelling summary statistics of the data by a Gaussian probability distribution In another popular approach called approximate Bayesian computation, the inference is performed by identifying parameter values for which the summary statistics of the simulated data are close to those of the observed data Synthetic likelihood is easier to use as no measure of `closeness' is required but the Gaussianity assumption is often limiting Moreover, both approaches require judiciously chosen summary statistics We here present an alternative inference approach that is as easy to use as synthetic likelihood but not as restricted in its assumptions, and that, in a natural way, enables automatic selection of relevant summary statistic from a large set of candidates The basic idea is to frame the problem of estimating the posterior as a problem of estimating the ratio between the data generating distribution and the marginal distribution This problem can be solved by logistic regression, and including regularising penalty terms enables automatic selection of the summary statistics relevant to the inference task We illustrate the general theory on canonical examples and employ it to perform inference for challenging stochastic nonlinear dynamical systems and high-dimensional summary statistics

Scatter Matrices and Independent Component Analysis

Abstract: In the independent component analysis (ICA) it is assumed that the components of the multivariate independent and identically distributed observations are linear transformations of latent independent components The problem then is to find the (linear) transformation which transforms the observations back to independent components In the paper the ICA is discussed and it is shown that, under some mild assumptions, two scatter matrices may be used together to find the independent components The scatter matrices must then have the so called independence property The theory is illustrated by examples

Robust independent component analysis based on two scatter matrices

Abstract: Oja, Sirkia, and Eriksson (2006) and Ollila, Oja, and Koivunen (2007) showed that, under general assumptions, any two scatter matrices with the so called independent components property can be used to estimate the unmixing matrix for the independent component analysis (ICA). The method is a generalization of Cardoso’s (Cardoso, 1989) FOBI estimate which uses the regular covariance matrix and a scatter matrix based on fourth moments. Different choices of the two scatter matrices are compared in a simulation study. Based on the study, we recommend always the use of two robust scatter matrices. For possible asymmetric independent components, symmetrized versions of the scatter matrix estimates should be used.

Collaborative supply chain configurations: The implications for supplier performance in production and inventory control

Abstract: Changing how and when a supplier delivers a product can transform a business model (Hoover et al., 2001). Vendor managed inventory (VMI) is one such mechanism that has been popular in recent literature (Holmstrom, 1998; Sabath et al., 2001). Unfortunately, however, practical examples of how VMI and other collaborative supply chain configurations can be precisely used to improve production planning and inventory control in supplier firms are difficult to find in industry. For example, the scope of standard solutions for VMI in commercial enterprise resource planning applications does not include recommendations for linking the replenishment collaboration to production and inventory control.

Topological Superfluids

Abstract: There are many topological faces of the superfluid phases of $^3$He. These superfluids contain various topological defects and textures. The momentum space topology of these superfluids is also nontrivial, as well as the topology in the combined (${\bf p},{\bf r})$ phase space, giving rise to topologically protected Dirac, Weyl and Majorana fermions living in bulk, on the surface and within the topological objects. The nontrivial topology lead to different types of anomalies, which extended in many different directions the Landau-Khalatninkov theory of superfluidity.

Random Fourier Features for Operator-Valued Kernels

Abstract: Devoted to multi-task learning and structured output learning, operator-valued kernels provide a flexible tool to build vector-valued functions in the context of Reproducing Kernel Hilbert Spaces To scale up these methods, we extend the celebrated Random Fourier Feature methodology to get an approximation of operator-valued kernels We propose a general principle for Operator-valued Random Fourier Feature construction relying on a generalization of Bochner's theorem for translation-invariant operator-valued Mercer kernels We prove the uniform convergence of the kernel approximation for bounded and unbounded operator random Fourier features using appropriate Bernstein matrix concentration inequality An experimental proof-of-concept shows the quality of the approximation and the efficiency of the corresponding linear models on example datasets

Modeling the effect of composition and temperature on the conductivity of synthetic copper electrorefining electrolyte

Abstract: The physico-chemical properties of the copper electrolyte significantly affect the energy consumption of the electrorefining process and the quality of the cathode product. Favorable conditions for electrorefining processes are typically achieved by keeping both the electrolyte conductivity and diffusion coefficient of Cu(II) high, while ensuring low electrolyte viscosity. In this work the conductivity of the copper electrorefining electrolyte was investigated as a function of temperature (50–70 °C) and concentrations of copper (Cu(II), 40–60 g/L), nickel (Ni(II), 0–20 g/L), arsenic (As(III), 0–30 g/L) and sulfuric acid (160–220 g/L). In total 165 different combinations of these factors were studied. The results were treated using factorial analysis, and as a result, four electrolyte conductivity models were built up. Models were constructed both with and without arsenic as the presence of As(III) appeared to cause non-linearity in some factor effects and thus impacted the conductivity in more complex ways than previously detailed in literature. In all models the combined effect of factors was shown to be minor when compared to the effect of single factors. Conductivity was shown to increase when copper, nickel and arsenic concentrations were decreased and increase with increased temperature and acidity. Moreover, the arsenic concentration was shown to decrease the level of conductivity more than previously suggested in the literature.

Well-Rounded Lattices: Towards Optimal Coset Codes for Gaussian and Fading Wiretap Channels

Abstract: The design of lattice coset codes for wiretap channels is considered. Bounds on the eavesdropper's correct decoding probability and information leakage are first revisited. From these bounds, it is explicit that both the information leakage and error probability are controlled by the average flatness factor of the eavesdropper's lattice, which we further interpret geometrically. It is concluded that the minimization of the (average) flatness factor of the eavesdropper's lattice leads to the study of well-rounded lattices, which are shown to be among the optimal in order to achieve these minima. Constructions of some well-rounded lattices are also provided.

Well-Rounded Lattices for Reliability and Security in Rayleigh Fading SISO Channels

Abstract: For many wiretap channel models asymptotically optimal coding schemes are known, but less effort has been put into actual realizations of wiretap codes for practical parameters. Bounds on the mutual information and error probability when using coset coding on a Rayleigh fading channel were recently established by Oggier and Belfiore, and the results in this paper build on their work. However, instead of using their ultimate inverse norm sum approximation, a more precise expression for the eavesdropper's probability of correct decision is used in order to determine a general class of good coset codes. The code constructions are based on well-rounded lattices arising from simple geometric criteria. In addition to new coset codes and simulation results, novel number-theoretic results on well-rounded ideal lattices are presented.

Uncovering the ultimate performance of single-walled carbon nanotube films as transparent conductors

Abstract: The ultimate performance - ratio of electrical conductivity to optical absorbance - of single- walled carbon nanotube (SWCNTs) transparent conductive films (TCFs) is an issue of considerable application relevance. Here, we present direct experimental evidence that SWCNT bundling is detrimental for their performance. We combine floating catalyst synthesis of non-bundled, high-quality SWCNTs with an aggregation chamber, in which bundles with mean diameters ranging from 1.38 to 2.90 nm are formed from identical 3 microns long SWCNTs. The as-deposited TCFs from 1.38 nm bundles showed sheet resistances of 310 Ohms/sq. at 90% transparency, while those from larger bundles of 1.80 and 2.90 nm only reached values of 475 and 670 Ohms/sq., respectively. Based on these observations, we elucidate how networks formed by smaller bundles perform better due to their greater interconnectivity at a given optical density. Finally, we present a semi-empirical model for TCF performance as a function of SWCNT mean length and bundle diameter. This gives an estimate for the ultimate performance of non-doped, random network mixed-metallicity SWCNT TCFs at ~80 Ohms/sq. and 90% transparency.

Editorial corner – A personal view 3D-printed polymers for biomedical applications

Abstract: cant technology leaps take place simultaneously in several complementary areas. This is the case in polymeric biomaterials’ technology. Amazing advances have been made in the additive manufacturing technologies, i.e. 3D-printing. It is a family of different layer by layer manufacturing methods like fused deposition modeling, powder laser sintering, ink-jet printing, stereo lithography, two or multi photon polymerization and hydrogel 3D-printing. All of them applicable for processing of polymeric materials. Essential developments have taken place also in biomedical imaging and 3D-design to create the necessary digital models needed for the 3D-printing. At the same time great progress has been achieved in biopolymers’ development. They can be tailored to be biostable or bioresorbable in a controlled way, or in some cases even bioactive. Bioactivity has been achieved through addition of bioactive compounds like bioactive glass or through functionalization of the polymeric structure (www.kunststoffe-international.com/1332070, DOI: 10.1016/j.actbio.2011.06. 039, DOI: 10.1016/j.compscitech.2012.10.014, DOI: 10.1002/jbm.b.32863, DOI: 10.1002/pola.27400, DOI: 10.1039/C5TB01468A). The parallel advance which has taken place in the biomedical sciences has significantly improved the options in regenerative medicine and may even enable tissue engineering. For polymer research this important and fast developing area offers great opportunities of high societal impact. However, the demands for polymer properties are strict for the novel biomaterials. They should meet the various physical criteria set by the needs of the application, but in addition the demand of non-toxicity is a self-evident must in biomedical applications. Successful 3D-additive manufacturing sets requirements to master the viscosities and liquid to solid transitions in the printing process. In stereo lithography the oligomeric low viscosity prepolymers need to be fast photo curing; with photo curing chemistry that is non-toxic in the biological environment and can get approved by the authorities. Especially for resorbable polymers there is a need to find alternative curing chemistries in addition to acrylate or methacrylate based systems. (www.kunststoffe-international.com/1325854) Complementary research interplay is essential in this field of science. An excellent example of such an effort has been the ArtiVasc 3D-project funded by the 7 Framework Program of the European Union to prepare polymeric scaffolds for vascularized, cell tissue cultivated skin like tissue. Outcome of the project is presented in the video: ArtiVasc 3D on the Way to Bio-Artificial Tissue. Novel technologies, breakthroughs in biomedical sciences together with innovative novel polymers enable jointly significant enhancements for life!

Fast Subset Convolution.

Abstract: A basic strategy to solve hard problems by dynamic programming is to express the partial solutions using a recurrence over the 2 subsets of an n-element set U . Our interest here is in recurrences that have the following structure: For each subset S U , in order to obtain the partial solution at S , we consider all possible ways to partition S into two disjoint parts, T and S n T , with T S . Fast subset convolution [1] is a technique to speed up the evaluation of such recurrences, assuming the recurrence can be reduced to a suitable algebraic form. In more precise terms, let R be an algebraic ring, such as the integers equipped with the usual arithmetic operations (addition, negation, multiplication). We seek a fast solution to:

Spirits and Ceremonies: How to Construct Religious Experience in Documentary Film

Abstract: IntroductionThere is a lot of research about religion and film concerning fiction films, but not very much dealing with documentary film, and practically none at all from the point of view of practice led or artistic research.I am a documentary film-maker but also a researcher, specialized in practice led research. This is also my point of view in this article. I have directed several films where spirituality or spiritual beliefs have been one of the topics. This time I am analysing two of my films: Kusum (2000) and Leap (2012) and reflecting the film-making process of these films.Although these films are meant for wider audiences and they are not strictly scientific, there is a strong scientific background in both films and they are based on serious research. Via these films and the experience of making them, I ponder the questions of presenting religious or spiritual experience in documentary film. How can the invisible be made visible in a visual medium? This question also has been asked in the literature (e.g. Bird, 1982: 4; Soderberg Widding, 2005: 77- 95). The problem is even more interesting in the field of documentary film, because traditionally the definition of documentary film has been based on the photographic, indexical relationship between the visible object and its representation. When one is constructing spiritual, abstract, conceptual, religious or theological phenomena, how can it be done? What are the means to do this visually, aurally and cinematically? Finally I try to define what kind of films Kusum and Leap are.Two cases: Kusum and LeapKusum is a documentary film about an Indian family fighting evil spirits and disease. The main character, Kusum, is a 14-year-old Indian girl who falls ill. She stops eating, isolates herself and suffers raving fits. Kusum's family initially tries Western medicine, but then opts for traditional Indian spiritual healing. Bhagat is a well-known religious healer throughout the region, and people travel hundreds of miles to see him. Bhagat's methods include conversation, rituals and herbal treatments. Joint trance sessions in which spirits talk constitute the core of his methodology. According to Bhagat evil spirits have attacked the family. He orders the family to travel together with him to the holy town of Balaji Mehndipur. It is believed that Balaji has special healing powers and that's why there are sick people everywhere. Kusum's family participates in Bhagat's sessions and various rituals. In the crucial session, the spirits surrender, beg for mercy and swear, on Bhagat's command, to leave the family alone.Leap is a film about the Hare Krishna movement told through two main characters. The film follows the spiritual paths of Keshava Madhava Das, a Finnish disciple and his guru Radhanath Swami, one of the movement's foremost spiritual leaders. Keshava faces the dilemmas of reconciling his personal life with a calling of a demanding religion. The film starts with Keshava's daily life as a seasonal tram driver in Helsinki. In pursuit of enlightenment he travels to Vrindavan in India, and together with thousands of other devotees, takes part in a pilgrimage led by Radhanath Swami. Keshava studies to become an altar priest. He is looking for love and finally gets married to an Indian disciple of Radhanath Swami. Together with his wife, Keshava settles into a family home in Mayapur, the "Vatican" of the movement. Both Keshava and Radhanath Swami must engage in self-sacrifice, asceticism and rigorous ritual discipline to follow in the footsteps of Krishna. The film follows their struggles over two years as they try to achieve their goals.Why these films?Spiritual and religious themes have always interested me, although I am not a religious person. It might be, that God does or does not exist, but religion, godhead and spirituality certainly exist. In my Taiga Nomads films (1992) shamanism was part of the life of the Evenki people living on the taiga areas of Siberia. …

Multi-view Kernel Completion

Abstract: In this paper, we introduce the first method that (1) can complete kernel matrices with completely missing rows and columns as opposed to individual missing kernel values, (2) does not require any of the kernels to be complete a priori, and (3) can tackle non-linear kernels. These aspects are necessary in practical applications such as integrating legacy data sets, learning under sensor failures and learning when measurements are costly for some of the views. The proposed approach predicts missing rows by modelling both within-view and between-view relationships among kernel values. We show, both on simulated data and real world data, that the proposed method outperforms existing techniques in the restricted settings where they are available, and extends applicability to new settings.

Nano-indentation studies and macro-scale mechanical testing of calcium modified (Pb0.75Ba0.25)(Zr0.70Ti0.30)O3 ceramics

Abstract: The main motivation of present article is to clarify the role of calcium ions in the formation of the mechanical properties of (Pb0.75Ba0.25)(Zr0.70Ti0.30)O3 ceramics. For this purpose the base ceramics as well as ceramics contain 0.5; 1; 2; 3; 4 at.% of calcium were prepared using conventional mixed oxide method. In the experiment, the nanoindentation techniques were applied to evaluate the elastic modulus and hardness values as a function of indentation depth. The results were compared with these in macroscale obtained by ultrasound techniques and shown that not always expensive nanoscale characterisation methods reveal real materials properties in microscale.

PARTIAL TRACKING IN SINUSOIDAL MODELING - An Adaptive Prediction-based RLS Lattice Solution

Abstract: Partial tracking plays an important role in sinusoidal mode ling analysis, being the stage in which the model parameters are obtained. This is accomplished by coherentl y grouping the spectral peaks found in each frame into time-evolving tracks of varying frequency and amplitu de. The main difficulties faced by partial tracking algorithms are the analysis of polyphonic signals and the pu rsuit of tracks exhibiting strong modulations in frequency and amplitude. In these circumstances, linear pr ediction over the trajectory of a given track has been shown to improve partial tracking performance. This pa per proposes an adaptive RLS lattice filter for the purpose of prediction in partial tracking. A new heurist ic which certifies the filter convergence is also presented. Computer simulation results are shown to compar e the proposed implementation with that of other predictors. The performance of the proposed solution is sim ilar to that of competing methods, albeit with reduced computational complexity as well as improved numer ical stability.

Towards a conceptual framework for the conservation of urban heritage

Abstract: Men come together in the city, Aristole wrote, in order to live. They remain a group in order to lead a good life. The ordered system of meanings and symbols in terms of social interaction evolved by individuals as members of a group is called culture . Their dwelling in urban space, by encoding and decoding meanings according to the norms dictated by their cultural values, gives the city its raison d'etre, that in turn gives orientation and identity to societies in space and time. This evolved essence of the urban space, enhancing Man's knowledge of Man, accrues a heritage value by documenting a transformation of a culture, and its protection secures a cultural continuity. Ever since the concept of time was known, symbols of the past were appreciated for reinforcing a cultural continuity. This past proclaimed by mythology, ideology, nationalism, local pride, or romantic ideas, is the heritage . • The particular way of proclaiming a heritage by ascribing diverse values on it moulded each culture with a unique identity. Among the values ascribing on heritage moulded each culture with a unique identity. Among the values ascribed on heritage are documentary, historic,

Water stress in global transboundary river basins: significance of upstream water use on downstream stress

Abstract: Growing population andwater demand have increased pressure onwater resources in various parts of the globe, includingmany transboundary river basins.While the impacts of upstreamwater use on downstreamwater availability have been analysed inmany of these international river basins, this has not been systematically done at the global scale using coherent and comparable datasets. In this study, we aim to assess the change in downstreamwater stress due to upstreamwater use in theworld’s transboundary river basins.Water stress wasfirst calculated considering only local water use of each sub-basin based on country-basinmesh, then comparedwith the situationwhen upstreamwater use was subtracted fromdownstreamwater availability.We found that water stress was generally already highwhen considering only local water use, affecting 0.95–1.44 billion people or 33%–51%of the population in transboundary river basins. After accounting for upstreamwater use, stress level increased by at least 1 percentage-point for 30–65 sub-basins, affecting 0.29–1.13 billion people. Altogether 288 out of 298middle-stream and downstream sub-basin areas experienced some change in stress level. Further, we assessedwhether there is a link between increasedwater stress due to upstreamwater use and the number of conflictive and cooperative events in the transboundary river basins, as captured by two prominent databases. No direct relationshipwas found. This supports the argument that conflicts and cooperation events originate from a combination of different drivers, amongwhich upstream-inducedwater stressmay play a role. Ourfindings contribute to better understanding of upstream–downstreamdynamics inwater stress to help address water allocation problems.

Using Stocks and Flows Diagrams to Understand Business Process Behavior

Abstract: Business Process Modeling has over the years focused in the activities and the logic how work gets done. That is reflected in the modeling notations like BPMN, which show the sequence of activities, their performers and the different paths each instance of the process can take. However, the performance of an organization and its business processes are measured aggregating the flow of the results of single process instances. Often the flows are not running smoothly but there are variations, delays, accumulations and other phenomena, which can be causes for the processes not to meet their expected performance levels. Therefore understanding the behavior of the business process over time is critical for any improvement initiative. In this paper we show how stocks and flows diagrams can be used to model business processes and how that model can be simulated to understand its behavior over time. Simulations can help in revealing the critical points to remove bottlenecks and improve the overall performance of the processes. First we a simple introduction to modeling business processes using stocks and flows diagrams. The we describe a real life case using stocks and flows models and simulation to reveal a problem and proposing a solution in health care environment.

Design and Analysis of Downlink Channel Estimation Based on Parametric Model for Massive MIMO in FDD Systems.

Abstract: This paper investigates downlink channel estimation in frequency-division duplex (FDD)-based massive multiple-input multiple-output (MIMO) systems. To reduce the overhead of downlink channel estimation and uplink feedback in FDD systems, cascaded precoding has been used in massive MIMO such that only a low-dimensional effective channel needs to be estimated and fed back. On the other hand, traditional channel estimations can hardly achieve the minimum mean-square-error (MMSE) performance due to lack of the a priori knowledge of the channels. In this paper, we design and analyze a strategy for downlink channel estimation based on the parametric model in massive MIMO with cascaded precoding. For a parametric model, channel frequency responses are expressed using the path delays and the associated complex amplitudes. The path delays of uplink channels are first estimated and quantized at the base station, then fed forward to the user equipment (UE) through a dedicated feedforward link. In this manner, the UE can obtain the a priori knowledge of the downlink channel in advance since it has been demonstrated that the downlink and the uplink channels can have identical path delays. Our analysis and simulation results show that the proposed approach can achieve near-MMSE performance.

A multiple instance learning approach for sequence data with across bag dependencies

Abstract: In Multiple Instance Learning (MIL) problem for sequence data, the learning data consist of a set of bags where each bag contains a set of instances/sequences. In many real world applications such as bioinformatics, web mining, and text mining, comparing a random couple of sequences makes no sense. In fact, each instance of each bag may have structural and/or temporal relation with other instances in other bags. Thus, the classification task should take into account the relation between semantically related instances across bags. In this paper, we present two novel MIL approaches for sequence data classification: (1) ABClass and (2) ABSim. In ABClass, each sequence is represented by one vector of attributes. For each sequence of the unknown bag, a discriminative classifier is applied in order to compute a partial classification result. Then, an aggregation method is applied to these partial results in order to generate the final result. In ABSim, we use a similarity measure between each sequence of the unknown bag and the corresponding sequences in the learning bags. An unknown bag is labeled with the bag that presents more similar sequences. We applied both approaches to the problem of bacterial Ionizing Radiation Resistance (IRR) prediction. We evaluated and discussed the proposed approaches on well known Ionizing Radiation Resistance Bacteria (IRRB) and Ionizing Radiation Sensitive Bacteria (IRSB) represented by primary structure of basal DNA repair proteins. The experimental results show that both ABClass and ABSim approaches are efficient.