Browsing by Title

The geological Discrete Fracture Network (DFN) model is a statistical model for stochastically simulating rock fractures and minor faults (Fox et al. 2007). Unlike the continuum model approaches, DFN model geometries explicitly represent populations of individual or equivalent fractures (Wilson et al. 2011). Model construction typically involves stochastic approaches that create multiple deterministic realizations of the fracture network (Gringarten 1998). This study was made as a part of a broader Salpausselkä project to gain deeper understanding of the brittle structures in the study area.This thesis can be broken down to three steps: literaturereview of the DFN methodology, parameterization of the model variables, and DFN modeling itself. For the purposes of the DFN modeling one-dimensionalfracture intensities measured in the field (P10) had to be converted into their volumetric counterpart (P32). Wang’s (2005) C13 conversion factor was decided tobe the most appropriatemethod. Calculation of the angles between the scanlines and fracture normals (α), conversion factor C13, and P32 were done in Python by applying the methods presented by Wang (2005) and Fox et al. (2007, 2012). Fracture setswere weighted by their P10 intensities to get clearer picture of the dominant fracturing orientations. For better and automated classification clustering of the fracture poles into desired number of mean vectorswas conducted byusing kmeansfunctionof Python module MPLstereonet. The function finds centers of multi-modal clusters of data by using a numpy einsummodified for spherical measurements. Fracture set poles were divided into populationsby finding the mean vector with the smallest angular distance from each pole. C13 calculation was done by integrating over the probability distribution function(PDF)of each population.C13 values produced by the script fall within the expected range quoted by the reference literature(Wang 2005, Fox et al. 2007, 2012). In the final modeling phase the clustered groups were modeled in MOVE as finite surfaces and the resulting DFN model was compared to the Local anisotropy interpolator (LAI) model created by Ruuska (2018).Fracture populations were modeled on an outcrop level as well as interpolated over the whole study area, producing two different interpretations of the most dominant fracturing orientations.Based on the results, fracture set pole clustering with open source methods (MPLStereonet K-means) is a feasible approach. K-means clustering algorithm was superior to the expert approach on every level, though more studies are needed to ascertain the soundness of the methodology. Statements made at this point are merely tentative due to the quality and amount of the available data. Taking into account the results of the parallel MSc thesis (Ruuska, 2018) thesis, the DFN and clustered fracture populations constructed using aforementioned methods can be used as a tentative approximation of the preferred fracturing orientations within the boundaries of the study area. Outcrop level model shows the true, measured values and could be used as ground truth in future modeling efforts. Efficient production of large-scale brittle models could be possible with the added flexibility of the implicit modeling methods and automated clustering.

The presence of dislocations in metal crystals accounts for the plasticity of metals. These dislocations do not nucleate spontaneously, but require favorable conditions. These conditions include, but are not limited to, a high temperature, external stress, and an interface such as a grain boundary or a surface. The slip of dislocations leads to steps forming on the surface, as atomic planes are displaced along a line. If a void is placed very near a surface, the possibility of forming a dislocation platelet exists. The skip of the dislocation platelet would displace the surface atoms within a closed line. Repeating such a process may form a small protrusion on the surface. In this thesis, the mechanism with which a dislocations displace the surface atoms within a closed loop is studied by using molecular dynamics (MD) simulations of copper. A spherical void is placed within the lattice, and the lattice is then subjected to an external stress. The dislocation reactions which lead to the formation of the dislocation platelet after the initial dislocation nucleation on the void is studied by running MD simulations of a void with the radius of 3 nm under tensile stress. Since the dislocations are thermally activated, the simulation proceeded differently for each run. We describe the different ways the dislocations nucleate, and the dislocation reactions that occur when they intersect to form the platelet. The activation energy of this process was studied by simulating half of a much larger void, with a radius of 8 nm, in order to obtain a more realistic nucleation environment. Formulas connecting the observable and controllable simulation variables with the energies of the nucleation are derived. The activation energies are then calculated and compared with values from literature.

When couples with children split or divorce, they are often unable to come to a mutual agreement concerning their child's place of residency, custody, the child's meetings with the other parent and the frequency of these meetings, or financial aid one parent is obliged to pay the other parent for the child. In many countries, these disagreements quite often lead to long disputes in court. A lot of research has been made (both in Finland and internationally) concerning the court's consideration of disputes about children. This thesis studies the disputes on custody and residency of a child in the district courts of Finland. The objective is to find out which factors play the biggest role in solving these disputes in court. Nine district courts of Finland have kindly provided the documents of the disputes concerning custody and residency of children from the period of 2004 - 2015. Only the cases where a dispute was solely between the parents of a child (no other relatives) and where the final decision was made by court (no agreement between the parties) are taken into analysis. Disputes are divided into two types - the ones where residency of a child was involved in a dispute (residency disputes) and the ones where it was not involved (custody disputes). The winner of a dispute is a dependent variable. A logistic regression model is applied for the custody disputes, and a cumulative logistic regression model is applied for the residency disputes. Due to results of the analysis, mothers win more disputes than fathers, but the difference is statistically significant only for the residency disputes. When only father is of a foreign background, it lowers father's winning chances in a custody dispute, but neither father's nor mother's foreign backgrounds are statistically significant for the residency disputes. A substantiated violence of father towards mother again acts negatively for fathers in custody disputes, and so does a non-substantiated accusation regarding alcohol or drug abuse by father. For the residency disputes, the main factors decreasing fathers' probability to win are mother hiring a legal assistant and father receiving legal aid (which takes place when father is not financially capable of hiring a legal assistant). Established conditions of a child at one of the parents increase the winning chances of that parent, but the effect is higher for fathers. All the accusations (both substantiated and non-substantiated in court) act in favor of fathers; these are substantiated mother's mental disorder, non-substantiated alcohol or drug abuse by mother and non-substantiated accusation regarding father's violence towards mother. At the same time, no variables regarding genders of children disputed about, genders of a judge or of legal assistants are statistically significant in the models. The same concerns the parents' demands in court, as well as the ages of parents (and their difference) and of children involved in disputes. This investigation can be extended by adding the disputes from other years and from other district courts into the analysis.

The vast amount of data created in the world today requires an unprecedented amount of processing power to be turned into valuable information. Importantly, more and more of this data is created on the edges of the Internet, where small computers, capable of sensing and controlling their environments, are producing it. Traditionally these so-called Internet of Things (IoT) devices have been utilized as sources of data or as control devices, and their rising computing capabilities have not yet been harnessed for data processing. Also, the middleware systems that are created to manage these IoT resources have heterogeneous APIs, and thus cannot communicate with each other in a standardized way. To address these issues, the IoT Hub framework was created. It provides a RESTful API for standardized communication, and includes an execution engine for distributed task processing on the IoT resources. A thorough experimental evaluation shows that the IoT Hub platform can considerably lower the execution time of a task in a distributed IoT environment with resource constrained devices. When compared to theoretical benchmark values, the platform scales well and can effectively utilize dozens of IoT resources for parallel processing.

With the increasing growth of data traffic in mobile networks there is an ever growing demand from the operators for a more scalable and cost efficient network core. Recent successes in deploying Software-Defined Networking (SDN) in data centers and large network backbones has given it credibility as a viable solution for meeting the requirements of even the large core networks. Software-Defined Networking is a novel new paradigm where the control logic of the network is separated from the network elements into logically centralized controllers. This separation of concerns offers more flexibility in network control and makes writing of new management applications, such as routing protocols, easier, faster and more manageable. This thesis is an empirical experiment in designing and implementing a scalable and fault- tolerant distributed SDN controller and management application for managing the GPRS Tunneling Protocol flows that carry the user data traffic within the Evolved Packet Core. The experimental implementation is built using modern open-source distributed system tools such as the Apache Zookeeper distributed coordination service and Basho's Riak distributed key-value database. In addition to the design, a prototype implementation is presented and its performance is evaluated.

In the context of data science and machine learning, feature selection is a widely used technique that focuses on reducing the dimensionality of a dataset. It is commonly used to improve model accuracy by preventing data redundancy and over-fitting, but can also be beneficial in applications such as data compression. The majority of feature selection techniques rely on labelled data. In many real-world scenarios, however, data is only partially labelled and thus requires so-called semi-supervised techniques, which can utilise both labelled and unlabelled data. While unlabelled data is often obtainable in abundance, labelled datasets are smaller and potentially biased. This thesis presents a method called distribution matching, which offers a way to do feature selection in a semi-supervised setup. Distribution matching is a wrapper method, which trains models to select features that best affect model accuracy. It addresses the problem of biased labelled data directly by incorporating unlabelled data into a cost function which approximates expected loss on unseen data. In experiments, the method is shown to successfully minimise the expected loss transparently on a synthetic dataset. Additionally, a comparison with related methods is performed on a more complex EMNIST dataset.

Tourism is one of the main contributors in the fight against poverty, as it has become one of the strongest drivers of trade and prosperity in the global south. Protected area tourism is an especially quickly growing segment of the industry, having an important role in regional development on many rural areas of global south. However, territories labelled as protected areas represent a great variety of spaces. This research aims at unifying the holistic picture of protected area tourism governance by analysing, how protected areas with divergent landownership arrangements, management objectives and associated regulations influence tourism development and its local socio-economic impacts at the grass roots. This comparative case-study survey scrutinizes local-level tourism governance and territorial regulations on three neighbouring protected areas in Taita Taveta County, Kenya. The Tsavo National Parks are state-owned conservancies focusing on conserving biodiversity. LUMO community wildlife sanctuary is a nature tourism project owned and orchestrated by a local community, which aims to advance local socio-economic development via tourism while preserving the environment at the same time. The third area, Sarova, is a private-owned conservancy harnessed solely for nature tourism and profit-making. The areas are liable to same legislative framework and international phenomena have similar influence on them, which makes comparison on their divergent management objectives and local-level regulations expedient. By giving voice to local-level tourism stakeholders, it is possible to point out how the category (i.e. public, private or community) of the land owner and the areas’ respective management objectives influence tourism operations and impact the socio-economic outcomes from both conservation and tourism. The comparative analyses focus first on spatial socially constructed preconditions for tourism development and second, on its developmental outcomes that will primarily be analysed by reflecting the livelihood changes generated by protected area tourism and protection regulations in place. The data-set was gathered during field research in February–March 2016, and it is mainly based on semi-structured interviews with tourism employees, employers and regional experts. The principal method of interviewing is supplemented by observation and statistics, and the data is analysed by thematic and qualitative content analyses. The protected areas’ management objectives and associated regulations have drastic impacts on tourism development within their respective borders. The local administrations of the protected areas were identified as the primary institutions to explain the stark spatial differences in the case-study areas tourist numbers. Instead of the mere ”type” of the landowner, the areas’ respective management objectives and associated regulations determined whether protected area tourism generated livelihoods or other positive socio-economic outcomes. Altogether, similar preconditions for tourism development and similar socio-economic outcomes cannot be expected from all territories labelled as protected areas.

Differentiaaliliikkuvuusspektrometri (Differential Mobility Particle Sizer; DMPS) -laitteistoa voidaan käyttää ilmakehän aerosolihiukkasten lukumääräkokojakauman mittaamiseen. DMPS-laitteisto koostuu impaktorista, kuivaajasta, bipolaarisesta diffuusiovaraajasta, differentiaaliliikkuvuusanalysaattorista (Differential Mobility Analyzer; DMA) ja kondensaatio- hiukkaslaskurista (Condensation Particle Counter; CPC). Tässä työssä verrataan DMPS-laitteistossa rinnakkain mittaavan muokatun A20 CPC:n ja TSI 3776 CPC:n laskentastatistiikkaa. Pienimmillä aerosolihiukkasilla on vaikutus ympäristöön ja terveyteen, minkä takia on kasvava tarve mitata tarkasti myös pienimpien hiukkasten kokojakaumaa. Aerosolihiukkasten lukumääräkokojakaumaan ja siitä johdettavien suureiden epävarmuuksia ei kuitenkaan tunneta vielä täysin. Työssä pyritään parantamaan perinteisen CPC:n laskentastatistiikkaa ja tutkimaan lukumääräkokojakauman sekä siitä johdettavien suureiden, kuten muodostumisnopeuden (Formation Rate; J) ja kasvunopeuden (Growth Rate; GR), epävarmuuksia. Perinteinen, ilman suojavirtausta toimiva, CPC voidaan muokata havaitsemaan jopa alle 3 nm hiukkasia kasvattamalla lämpötilaeroa saturaattorin ja kondenserin välillä ja muuttamalla aerosolivirtausta. Tässä työssä A20 CPC:n aerosolivirtaus optiikan läpi kasvatettiin 2.5 litraan minuutissa diffuusiosta johtuvien häviöiden minimoimiseksi ja laskentastatistiikan parantamiseksi. Verrattuna TSI 3776 CPC:hen muokatulla A20 CPC:llä on 50 kertaa suurempi aerosolivirtaus, joten voimme olettaa, että muokattu A20 mittaa TSI 3776 UCPC:hen verrattuna enemmän hiukkasia pienemmällä epävarmuudella. Muokatulla A20 CPC:llä on parempi laskentastatistiikka, jonka ansiosta kokojakauman laskennasta johtuva suhteellinen virhe on pienempi. Muokatulla A20 CPC:llä on TSI 3776 CPC:hen verrattuna 50 kertaa suurempi aerosolivirtaus ja se laskee keskimäärin 50 kertaa enemmän hiukkasia koko DMPS-laitteiston mittaamalla kokoalueella (1-40 nm). Muokatulla A20 CPC:llä laskettu GR on noin 60% suurempi pienimmillä (3-6 nm) hiukkasilla ja noin 3% suurempi 6-11 nm hiukkasilla. Myös J on noin 30% suurempi muokatulla A20 CPC:llä laskettuna 3-6 nm hiukkasille. CPC:n laskennasta johtuva epävarmuus on syytä huomioitava määritettäessä DMPS-mittauksen kokonaisvirhettä. Laskentastatistiikalla on merkitystä paitsi lukumääräkokojakaumaan, myös sen johdannaissuureisiin.

Reunalaskennan tarkoituksena on siirtää tiedonkäsittelyä lähemmäs tiedon lähdettä, sillä keskitettyjen palvelinten laskentakyky ei riitä tulevaisuudessa kaiken tiedon samanaikaiseen analysointiin. Esineiden internet on yksi reunalaskennan käyttötapauksista. Reunalaskennan järjestelmät ovat melko monimutkaisia ja vaativat yhä enemmän ketterien DevOps-käytäntöjen soveltamista. Näiden käytäntöjen toteuttamiseen on löydettävä sopivia teknologioita. Ensimmäiseksi tutkimuskysymykseksi asetettiin: Millaisia teknisiä ratkaisuja reunalaskennan sovellusten toimittamiseen on sovellettu? Tähän vastattiin tarkastelemalla teollisuuden, eli pilvipalveluntarjoajien ratkaisuja. Teknisistä ratkaisuista paljastui, että reunalaskennan sovellusten toimittamisen välineenä käytetään joko kontteja tai pakattuja hakemistoja. Reunan ja palvelimen väliseen kommunikointiin hyödynnettiin kevyitä tietoliikenneprotokollia tai VPN-yhteyttä. Kirjallisuuskatsauksessa konttiklusterit todettiin mahdolliseksi hallinnoinnin välineeksi reunalaskennassa. Ensimmäisen tutkimuskysymyksen tuloksista johdettiin toinen tutkimuskysymys: Voiko Docker Swarmia hyödyntää reunalaskennan sovellusten operoinnissa? Kysymykseen vastattiin empiirisellä tapaustutkimuksella. Keskitetty reunalaskennan sovellusten toimittamisen prosessi rakennettiin Docker Swarm -konttiklusteriohjelmistoa, pilvipalvelimia ja Raspberry Pi -korttitietokoneita hyödyntäen. Toimittamisen lisäksi huomioitiin ohjelmistojen suorituksenaikainen valvonta, edellisen ohjelmistoversion palautus, klusterin laitteiden ryhmittäminen, fyysisten lisälaitteiden liittäminen ja erilaisten suoritinarkkitehtuurien mahdollisuus. Tulokset osoittivat, että Docker Swarmia voidaan hyödyntää sellaisenaan reunalaskennan ohjelmistojen hallinnointiin. Docker Swarm soveltuu toimittamiseen, valvontaan, edellisen version palauttamiseen ja ryhmittämiseen. Lisäksi sen avulla voi luoda samaa ohjelmistoa suorittavia klustereita, jotka koostuvat arkkitehtuuriltaan erilaisista suorittimista. Docker Swarm osoittautui kuitenkin sopimattomaksi reunalaitteeseen kytkettyjen lisälaitteiden ohjaamiseen. Teollisuuden tarjoamien reunalaskennan ratkaisujen runsas määrä osoitti laajaa kiinnostusta konttien käytännön soveltamiseen. Tämän tutkimuksen perusteella erityisesti konttiklusterit osoittautuivat lupaavaksi teknologiaksi reunalaskennan sovellusten hallinnointiin. Lisänäytön saamiseksi on tarpeen tehdä laajempia empiirisiä jatkotutkimuksia samankaltaisia puitteita käyttäen.

The rise of large language models (LLMs) has revolutionized natural language processing, par- ticularly through transfer learning and fine-tuning paradigms that enhance the understanding of complex textual data. This thesis builds upon the concept of fine-tuning to improve the under- standing of Finnish Wikipedia articles. Specifically, a BERT-based language model is fine-tuned to create high-quality document representations from Finnish texts. The learned representations are applied to downstream tasks, where the model’s performance is evaluated against baseline models. This thesis draws on the SPECTER paper, published in 2020, which introduced a training frame- work for fine-tuning a general-purpose document embedder. SPECTER was trained using a document-level training objective that leveraged document link information. Originally, SPECTER was designed for scientific articles, utilizing citations between articles. The training instances con- sisted of triplets of query, positive, and negative papers, with the aim of capturing the semantic similarity of the documents. This work extends the SPECTER framework to Finnish Wikipedia data. While scientific articles have citations, Wikipedia’s cross-references are used to build a document graph that captures the relatedness between articles. Additionally, Wikipedia data is publicly available as a full data dump, making it an attractive choice for the dataset in this thesis. One of the objectives is to demonstrate the flexibility of the SPECTER framework on a new dataset that has a similar networked structure to that of scientific articles. The fine-tuned model can be used as a general-purpose tool for various tasks and applications; however, in this thesis, its performance is measured in topic classification and cross-reference ranking. The Transformer-based language model produces fixed-length embeddings, which are used as features in the topic classification task and as vectors to measure the L2 distance of article vectors in the cross-reference prediction task. This thesis shows that the proposed model, WikiSpecter, optimized with a document-level objective, outperformed baseline models in both tasks. The performance indicates that Finnish Wikipedia provides relevant cross-references that help the model capture relationships across a range of topics.