# [kand] Perustieteiden korkeakoulu / SCI

## Permanent URI for this collection

## Browse

### Recent Submissions

Now showing 1 - 20 of 3939

Item Infant Sleep Staging with Wearable Sensors and Deep Learning: Evaluating and Controlling Age-Related Effects(2024-10-30) Häggman, Matias; Airaksinen, Manu; Perustieteiden korkeakoulu; Korte, RiikkaEarly childhood sleep problems have been associated with a variety of health and developmental issues. Sleep staging is used in clinical studies to diagnose sleep-related health issues. Standardized sleep staging typically requires a medical sleep study, usually polysomnography (PSG), which includes an electroencephalogram (EEG) analysis performed by a medical professional. Conventional sleep staging is not a practical solution for studying sleep in home environments, and alternative methods are needed to enable long-term sleep monitoring at home for children. Ranta et al. (2021) developed a wearable device for infants called NApping PAnts (NAPPA), which uses an accelerometer and gyroscope to record abdominal movements during sleep. We developed a deep learning classifier based on a bidirectional gated recurrent unit (GRU) for automatic infant sleep classification using the NAPPA device, leveraging an annotated dataset of 36 sleep recordings. The classifier achieved a median accuracy of 77% in detecting stages of deep sleep, light sleep, and wakefulness using five movement and respiratory features derived from sensor signals. Infant sleep and its indicators, such as respiratory rate, are dynamic and change within the first two years after birth. This variability may pose challenges for deep learning-based classifiers, as the training data typically consists of infants of varying ages. In this study, we performed a data analysis focusing on the effects of age variability on the features recorded by the NAPPA system and then sought methods to improve the classifier’s performance by accounting for the dynamic nature of infant sleep and the influence of age. We found a clear relationship between infant age and the recorded features, as well as between age and classifier performance. However, the proposed methods ultimately did not improve classifier performance. Although performance did not improve, our findings emphasize the importance of age in the automatic staging of infant sleep when using wearable sensors. The relationship between age, sleep habits, and recorded features suggests that age-specific adjustments could potentially improve the accuracy of sleep classifiers.Item Stokastinen malli Markovin ketjulla: tuotemuutosten vaikutus markkinaosuuksiin tuotekategoriassa(2024-10-01) Herola, Veikko; Kohonen, Jukka; Perustieteiden korkeakoulu; Ilmonen, PauliinaTässä työssä esitellään malli, joka mallintaa päivittäistavaramarkkinan brändien tuotemuutosten vaikutusta markkinaosuuteen. Mallin pohjana toimii stokastinen prosessi, jota kuvataan Markovin ketjuna. Mallin käyttämä data on kuluttajakysely-dataa. Malli pystyy ilman tietoa todellisista markkinaosuuksista ennustamaan tarkasti eri tuoteryhmien brändien markkinaosuuksia. Tämän pohjalta Markovin ketjuun tehdyillä tuotemuutoksen mukaisilla muutoksilla pystytään antamaan arvio markkinaosuuden noususta. Lisäksi malliin pystytään lisäämään markkinoinnin vaikutusta kuvaava funktio, jonka avulla voidaan arvioida markkinaosuuden kasvun nopeutta. Tuloksien pätevyyttä arvioidaan permutointimenetelmällä. Tulokset kertovat myös käytettävän datan tarkkuudesta. Työ on tehty yhteistyössä Ravogen Oy:n kanssa.Item Optimizing the RF Readout Circuit for the Resistive Bolometers(2024-09-25) Hossain, Nafi; Chang, Yu-Cheng; Perustieteiden korkeakoulu; Raasakka, MattiEver since the detection of thermal radiation was discovered more than a century ago, it has found widespread applications in various fields of physics. One of these devices is called a bolometer, which measures thermal radiation via temperature-dependent resistance. Using these bolometers, microwave radiation can be detected, and it has applications in astrophysics and quantum computing. Given the wide variety of applications for microwave radiation detection, pushing the sensitivity limits of these detection techniques is important. In this thesis, the optimization of an RF readout circuit for resistive bolometers is carried out. The readout signal produced by external microwave radiation is measured as the change in the transmitted RF current through a readout resonator circuit. In Matlab, a simulation was created to study how the lumped circuit elements of the readout circuit affected the bolometer's sensitivity. The simulation found that with a superconducting resonator, the sensitivity of the bolometer is only relevant to the resonant frequency. However, for a resonator with a lossy inductor, the sensitivity vanishes when this stray loss dominates the system loss. The loss of the inductor is inversely proportional to $L^2$. Somehow, it corresponds to a larger characteristic impedance $Z_0$, reducing the Q factor of the resonator since $Q\propto\frac{1}{Z_0}$. In addition, tuning the coupling capacitance of the readout system to the transmission line showed that a higher capacitance produced more losses outside the resonator, which also reduced the total Q factor. In an attempt to characterize the noise profile of the readout signal, the results gave the instrument noise of the VNA, where transmission measurements were conducted, leaving the noise profile's dependence on the circuits' configuration inconclusive. For a complete picture of the optimization problem, further research is suggested characterizing the noise profile of the readout signal.Item Knowledge graphs for drug repurposing(2024-09-20) Tuominen, Olivia; Nikitin, Alexander; Perustieteiden korkeakoulu; Savioja, LauriThis bachelor’s thesis is a literature review that explores the use of knowledge graphs in drug repurposing. Drug repurposing is the process of identifying new applications for existing drugs. The development of a new drug is estimated to cost hundreds of millions or even billions of dollars, and the entire process typically takes 10 to 17 years. By utilizing existing drugs and the clinical trials already conducted on them, the drug development process is accelerated, significantly reducing costs. This thesis examines how knowledge graphs, which are network-based data structures, unify biomedical information into a comprehensive structure. These knowledge graphs are used as foundations for machine learning applications that predict new targets for drugs. The purpose of this thesis is to analyze and investigate how knowledge graphs enhance drug repurposing. This thesis assesses how knowledge graph-related machine learning methods are applied, while also discussing the challenges and opportunities of them. The research material comprises several scientific articles and reports that explore the structure of knowledge graphs, their application in drug repurposing, and the machine learning methods associated with them. Some of the material focuses on theoretical perspectives, while others provide practical examples of projects or case studies. This thesis demonstrates that knowledge graphs can unify biomedical information and integrate it efficiently. This enables the use of knowledge graphs in machine learning-based prediction algorithms. Baricitinib’s repurposing for the treatment of COVID-19, presented in this thesis, illustrates the significance of knowledge graphs in identifying new possibilities for existing drugs. This is exceptionally important in situations like pandemics, where the traditional drug development process is too slow to meet urgent needs. As a conclusion, the combination of knowledge graphs and machine learning holds significant potential for improving the cost-effectiveness of drug development in the future.Item Generalized poset sorting and its subproblems(2024-09-18) Paraoanu, Antti; Chalermsook, Parinya; Perustieteiden korkeakoulu; Savioja, LauriSorting algorithms solve the well known problem of finding the total order of a set by making comparisons. The problem of sorting can be generalized in two independent ways. The first generalization is to relax the requirement that the set must be ordered by a total order; instead only a partial order is required, yielding the poset sorting problem. The second is to restrict the set of comparisons the algorithm can make, yielding the generalized sorting problem. When both generalizations are combined one gets the generalized poset sorting problem. Additionally, there are various intermediary problems. This thesis surveys the current best algorithms for each variation of the aforementioned sorting problems. Additionally, the thesis presents an algorithm that solves generalized poset sorting in O(nlog(n)+nk+j²) queries where j=n(n-1)-m denotes the number of forbidden queries. This is an improvement over the naïve O(nlog(n)+nk+nj) query complexity.Item CNT synthesis and characterization utilizing bimetallic Fe-Cu catalysts for transparent conducting films(2024-10-25) Suominen, Lauri; Karakasidis, Anastasios; Perustieteiden korkeakoulu; Kauppinen, EskoSince their discovery in 1991, carbon nanotubes have been the subject of extensive research due to their promising properties. CNTs are conductive, highly durable, and have good optical properties, making them ideal for potential applications in multiple fields. One of these applications is the manufacturing of transparent conducting films, which are used in touch screens, photovoltaics and OLEDs. Despite the recent advances in the field, significant challenges still remain related to the manufacturing consistency and scalability, and controlling the properties of individual nanotubes. This thesis focuses on the synthesis of CNTs with the floating catalyst chemical vapor deposition method, with the aim of creating thin films with high transmittance, and low sheet resistance. One of the main components in FCCVD synthesis of CNTs is the catalyst, which acts as a seed for the growth of a nanotube. Along with an often used iron catalyst, a bimetallic iron-copper catalyst will be used in the experiments. This will be achieved by adding copper acetylacetonate into the solution containing ferrocene, ethanol and thiophene, which is then injected to the reactor. The synthesized SWCNT TCFs were characterized with numerous methods: UV-Vis-NIR and Raman spectroscopy, transmission electron microscopy and resistance measurements. It was discovered that the bimetallic catalyst had a noticeable impact on the characteristics of the thin film and CNTs: optoelectronic performance was improved slightly over the base ferrocene solution. In addition, the CNT diameter and metallicity were observed to change as copper was added as a catalyst. Overall, a sheet resistance at 90% transmittance of 80 ohms/square was achieved after gold chloride doping.Item The effect of carbon disulfide as a sulfur source in carbon nanotube synthesis and characteristics(2024-10-24) Salmela, Otto; Yasin, Ghulam; Perustieteiden korkeakoulu; Kauppinen, EskoSingle-walled carbon nanotubes (SWCNTs) are a promising nanostructure with a variety of possible applications thanks to their unique properties such as mechanical, electrochemical and conductive properties. Despite decades of rigorous research since their discovery in 1991, many challenges remain today in synthesizing carbon nanotubes (CNTs), particularly selective growth of CNTs, scalability and reproducibility. Floating catalyst chemical vapor deposition (FCCVD) offers a promising synthesis method, which tackles multiple challenges through, for example, the continuous output of CNTs. The inclusion of a growth promoter in the synthesis of CNTs has been studied extensively, and various elements have been used, of which sulfur has shown the most promise. The exact effect of sulfur in the synthesis remains unknown, and thus further studies to advance our understanding are needed. This paper explores particularly the effect of carbon disulfide (CS2) as a sulfur source in the synthesis and characteristics of transparent conducting films (TFCs) of CNTs. The CNTs were found to minimize in diameter to about 1.3 nm with an optimum content of sulfur according to ultravioletvisible-near infrared (UV-Vis-NIR) spectroscopy characterizations. Additionally, a lowest sheet resistance of 77 Ω/□ at 90 % transmittance was attained for the synthesized TCFs. Diverse characterizations with Raman spectroscopy and also transmission electron microscopy (TEM) were additionally utilized, along with the ones mentioned, to further the understanding of CS2 in the resulting CNTs.Item Effect of agarose gel elasticity on the rooting of plant seedlings(2024-10-27) Saartoala, Seela; Datar, Anagha; Perustieteiden korkeakoulu; Martikainen, Jani-PetriMechanical root-soil interaction, influenced by the properties of both the soil and the root system, is one of the deciding factors on the survival and success of plants. To anchor the growing plant to the ground and access water and nutrient resources, the roots are required to be able to penetrate into the soil. This root penetration is limited by the mechanical resistance of the soil. In this thesis, agarose gel, a standard and well-studied research culture medium, was used as a soil model, and the root penetration ability of young Arabidopsis thaliana seedlings was studied on different gel stiffnesses. A micropipette force sensor (MFS) indentation method was implemented to measure the Young's moduli of gels with varying concentrations of agarose and establish the relationship between agarose concentration and gel elasticity. Root response to different culture medium stiffnesses was studied by observing the ability of the emerging primary root to penetrate the agarose gel surface on different concentrations. The results for the gel elastic moduli followed a power law relation with respect to the agarose concentration and placed agreeably in the spectrum of literature values, affirming the suitability of the MFS indentation method. No consistent effect of storage time on Young's modulus was observed, suggesting that the gels could maintain a fairly constant elasticity over a time period of several weeks. A steep, steplike decrease of Arabidopsis primary root penetration ability with increasing agarose concentration was witnessed, indicating an upper threshold value of medium elasticity for root penetration. Future validation of the MFS indentation method should include the examination of the effect of micropipette spring constant and capillary wall thickness on measured elasticity values as well as the consideration of agarose gel viscoelastic properties. A direct experimental comparison between root penetration ability and gel Young's modulus is needed to find the critical gel stiffness and quantify the limit of root penetration more precisely.Item Teräspalkin pinnankarheudesta johtuvan alkusärön sekä jännitysintensiteettikertoimen analyysi väsytyskokeiden perusteella(2024-10-10) Rantala, Aino; Gustafsson, Tom; Perustieteiden korkeakoulu; Martikainen, Jani-PetriTässä kandidaatintyössä perehdytään kiinteän aineen mekaniikasta johdettuun murtumismekaniikkaan, jota tutkitaan teräksisten palkkien väsytyskokeiden tulosten pohjalta. Aineistoa sovelletaan palkkien pinnankarheudesta johtuvien alkusäröjen sekä jännitysintensiteettikertoimien määrityksiin. Tutkielmassa pyritään saamaan vastaus kahteen tutkimuskysymykseen. Näistä ensimmäisen tavoite on selvittää väsytyskokeen läpikäyneen aineiston avulla, onko kaksiosaisella Parisin lailla selvitetyt alkusärönpituudet luotettavat, kun tiedetään palkkiin kohdistetut voimat sekä muoto ja halkeamiseen tarvittava syklimäärä. Kun analyyttisia tuloksia verrataan tunnettuihin karheusjakaumiin, voidaan todeta kaksiosaisen Parisin lain antavan oikeansuuruisia, mutta tuskin täsmällisiä arvoja. Toinen tutkimuskysymys pohjautuu alkusäröjen määrityksen tuloksiin. Sen tavoitteena on selvittää, miten analyyttisesti ja elementtimenetelmällä lasketut jännitysintensiteettikertoimet ensimmäisen syklin aikana vertautuvat toisiinsa ja mistä erot johtuvat. Kun eri tavoin määritettyjen jännitysintensiteettikertoimien arvoja verrataan toisiinsa, huomataan niiden olevan samaa suuruusluokkaa. Kuitenkin elementtimenetelmä antaa hieman suurempia arvoja kuin analyyttinen tapa, mikä johtunee siitä, että elemettimenetelmä ottaa huomioon kappaleen todellisen muodon. Vaikkei tuloksissa ole suurta vaihtelua, ovat elementtimenetelmän ratkaisut joka tapauksessa konservatiivisempia.Item Molecular dynamics simulations of spike protein interaction with viral membrane(2024-10-11) Gyursánszky, Cristina; Ghasemitarei, Maryam; Perustieteiden korkeakoulu; Ala-Nissilä, TapioMolecular dynamics (MD) is a computational technique used to study the properties and predict the motion of classical molecular systems. MD simulations have various applications in biophysics, bioinformatics and chemistry, including drug design, protein allostery research, and structure prediction. MD simulations are based on calculating the forces acting on each atom, which allows for determining the time evolution of the system. SARS-CoV-2, the causal agent of the coronavirus disease 2019 (COVID-19), caused a global pandemic in 2020. SARS-CoV-2 consists of four structural proteins, from which the spike (S) protein is considered to be the most crucial. At the cellular level, the S protein binds to the ACE2 receptor and enables the virus’s entry into the host cell. It undergoes a process of S-palmitoylation, where a palmitic acid attaches to a cysteine residue. Previous research has indicated that palmitoylation increases protein hydrophobicity and contributes significantly to protein stability, trafficking, and interaction with other proteins. In this thesis, we studied the spike protein of SARS-CoV-2 in viral membrane using MD simulations. In addition, we investigated whether cysteine palmitoylation affects the stability of the spike protein in the viral membrane. This was examined by performing constant-velocity steered MD (SMD) simulations, where we pulled the tail of the spike protein out of the membrane. Furthermore, we evaluated the stability of the spike protein by calculating free energy profiles using the Jarzynski equality. We also determined the appropriate parameters for the SMD simulations. Root mean square deviation (RMSD) values calculated in this work showed that the protein achieved a stable structure during the production of particle dynamics. However, the results obtained from the SMD simulations were partly contrary to the previous research, which showed that palmitoylation should increase protein stability. These conflicting results are most likely due to choosing unsuitable parameter values for applying the Jarzynski equality. In addition, more samples of the SMD simulations should be collected to obtain more reliable results.Item The Nakayama argument (Nakayama's lemma)(2024-10-20) Kemppainen, Konsta; Orlich, Milo; Perustieteiden korkeakoulu; Kivinen, OscarA module is an algebraic structure similar to a vector space except the scalars need not be invertible and their multiplication need not be commutative. Modules are used extensively in commutative and noncommutative algebra. In the field of commutative algebra, there is a result commonly known as Nakayama's lemma, which states that for any finitely generated module over a commutative ring, there exists a certain nontrivial element of the ring that scales every element of the module to zero. The proof strategy of Nakayama's lemma can be generalized into a result that we call the Nakayama argument. We use the Nakayama argument to prove Nakayama's lemma along with a famous result in linear algebra known as the Cayley--Hamilton theorem, which states that a matrix satisfies its own characteristic equation. We also introduce the notion of the Jacobson radical as the intersection of all maximal left-ideals of a ring and prove some elementary properties assuming Krull's theorem, which we prove as a consequence of Zorn's lemma. We show that linear combinations of the elements of a module with coefficients from the Jacobson radical of the ring of scalars are never sufficient to express every element of a nonzero finitely generated module, a result known as the Jacobson--Azumaya theorem. We remark that this result is sometimes referred to as the noncommutative analog of Nakayama's lemma, since Nakayama's lemma easily implies the result of the Jacobson--Azumaya theorem for finitely generated modules over a commutative ring, even without the need to assume Zorn's lemma.Item Mapping the phase space of Portevin-Le Chatelier effect models(2024-10-18) Mehtätalo, Juho; Mäkinen, Tero; Perustieteiden korkeakoulu; Martikainen, Jani-PetriThe Portevin-Le Chatelier effect (PLC) is an instability phenomenon of plastic flow occuring in certain metal alloys under certain external conditions. The effect is characterized by serrations in the plastic regime of the stress-strain curve, and the localisation of strain into strain bands. The spatio-temporal behavior and the type of serrations in the stress-strain curve depend on the external conditions of the experiment (temperature and strain rate), and the bands can be roughly divided into three categories, type A, B and C bands. Here we focus on the type A bands that propagate along the sample. The spatio-temporal behaviour of type A PLC bands of the Al-3\%Mg-alloy has been previously mapped to the Alessandro-Beatrice-Bertotti-Montrosi (ABBM) model the statistical behavior of which is known. The ABBM model is a stochastic differential equation, which describes the one-dimensional movement of a quasi-particle in a disordered landscape. The quasi-particle exhibits avalanche-like behaviour, where the movement of the particle occurs in velocity bursts. Modelling the type A PLC bands with the ABBM model introduces modifications to the ABBM model. The distance travelled by the ABBM quasi-particle is not limited in the standard ABBM model, but in the PLC experiments, the size of the specimen limits the distance. Thus in when simulating the type A bands with the ABBM model, an artificial limit to the travel distance is introduced by assuming a uniform distribution of starting locations for the type A PLC bands, and discarding the simulations where the travel distance exceeds the length of the specimen. This introduces a linear cutoff to the size distribution of the ABBM model. In addition, the PLC parameters strain and strain rate are mapped linearly to the ABBM parameters. In this thesis, the behaviour of the size-restricted ABBM model is studied using computer simulations. First, a model for mapping the parameters of the non-restricted ABBM model to the average velocity profile of the size-restricted ABBM model is developed using effective parameters. Then the suitability of the effective parameters in reproducing the duration and velocity distributions is studied. It is observed that the shape of these distributions change fundamentally, and description based on effective parameters is not enough.Item Characterization of a graphene bolometer for ultrasensitive microwave detection(2024-10-10) Hertell, Aku; Ma, Jian; Perustieteiden korkeakoulu; Möttönen, MikkoIn circuit quantum electrodynamics (cQED), the states of qubits, representing the outcomes of quantum computations, are read by detecting microwave pulses emitted by the qubits. Existing methods for detecting these pulses include quantum-limited amplifiers, Josephson photomultipliers and absorption in resonators with a semi-irreversible state change. As these detection schemes have limitations in scalability, a bolometer, which detects photons by measuring a temperature change, has been suggested as an alternative detection method with a higher potential applicability. However, single photons have not been detected with state-of-the-art bolometers, because the devices have only recently reached sufficient theoretical sensitivity limits. This thesis examines the sensitivity of a graphene bolometer, evaluating its viability for single microwave photon detection. We found that the lowest noise equivalent power of the bolometer, representing the power of the weakest detectable pulse, is 900 zW·Hz-1/2. Furthermore, the time constant characterizing the detection timescale of the bolometer was below 200 ns. These results are especially promising considering the brevity of the obtained timescale, which is on par with the state-of-the-art values. Due to the optimization potential still present in the sample and the already low time constants, bolometers based on a similar graphene-based design have significant potential for single microwave photon detection.Item Free-swimming kinematics of Artemia(2024-10-10) Rislakki, Ensio; Ayala Lara, Rafael; Perustieteiden korkeakoulu; Backholm, MatildaA huge variety of organisms inhabit the oceans, lakes and rivers of the earth. These forms of life differ vastly in shape and size, from the largest mammal, blue whale, with body length of 22-24 m to single bacteria in the size scale of 1-2 mum. This amounts to a 10^14-fold range in Reynolds number; the ratio of inertial and viscous forces. This is remarkably large compared to the 10^5-fold range in aerial locomotion. Furthermore, in aquatic locomotion, the Reynolds number is found to have values less than or equal to 1 for the smallest organisms, which never occurs in aerial locomotion. One would certainly expect the physics of the locomotion to be completely different in the extremes of this scale, and indeed, this turns out to be the case. The two extremes, micro- and macroscale, are relatively well understood, since in these regimes one can ignore the inertial or the viscous force, respectively. However, a particular challenge is proposed by the scale in between: the mesoscale. In this regime, characterized by a Reynolds number close to one, modeling swimming becomes exceedingly difficult. Unlike the microscale, where the linear Stokes equation applies, or the macroscale, governed by the quasi-linear Euler equation, locomotion at the mesoscale is described by the full-form Navier-Stokes equation, which is nonlinear and often numerically unstable. This range in Reynolds number is inhabited by the species Artemia salina, which was the subject of the experiments for this thesis. In this thesis, the strategies for swimming in mesoscale are analysed in great detail and some remarks are made on the scaling of the kinematics as the Artemia grow. Swimming kinematics are measured using an inverted microscope and an image analysis software. Much attention is paid to ensure that the Artemia swim undisturbed, allowing the locomotion to be observed as naturally as possible. Additionally, the concept of symmetry breaking area is introduced to assess the significance of non-reciprocal motion in swimmers at intermediate Reynolds numbers.Item Näkökulmia lentoyhtiöiden strategioihin(2024-09-20) Pelkonen, Santtu; Holopainen, Mari; Perustieteiden korkeakoulu; Rajala, RistoTässä kandidaatintyössä tarkastellaan lentoyhtiöiden strategioita eri näkökulmista. Työ on kirjallisuuskatsaus, jonka tavoite on koota yhteen erilaisia näkökulmia ja työkaluja, joita kirjallisuudessa on käytetty lentoyhtiöiden strategioiden arvioimiseksi, ymmärtämiseksi ja luokittelemiseksi. Työssä tarkasteltavat näkökulmat on edelleen rajattu lentoyhtiöiden sisäiseen näkökulmaan. Ulkoisten tekijöiden vaikutusta strategiaan ei erikseen tutkita. Työssä arvioidaan myös näkökulmien hyödyllisyyttä. Perinteinen jaottelu halpalentoyhtiöiden (LCC) ja perinteisten lentoyhtiöiden (FSC) välillä selitetään ja analysoidaan, kuinka tämä jaottelu on muuttunut hybridimallien myötä. Tämän jälkeen tarkastellaan markkinafokusnäkökulmaa lentoyhtiöiden strategioihin. Lisäksi tarkastellaan eri tapoja, joilla kasvustrategioita ja -metodeja on luokiteltu. Lopuksi työssä tarkastellaan, miten Milesin ja Snown yritysstrategioiden typologiaa on hyödynnetty lentoyhtiöiden strategioiden selittämiseksi. Työn tulokset osoittavat, että lentoyhtiöiden strategioita voidaan tarkastella monipuolisesti ja että perinteinen LCC-FSC jaottelu ei yksin riitä selittämään kaikkia alan toimintamalleja. Näin ollen on tärkeää soveltaa useita eri näkökulmia lentoyhtiöiden strategioiden ymmärtämiseksi.Item Laajojen kielimallien parantamat ohjelmointivirheilmoitukset opetuksessa(2024-09-06) Komulainen, Onni; Sorva, Juha; Perustieteiden korkeakoulu; Savioja, LauriTämä kandidaatintyö on kirjallisuuskatsaus, joka käsittelee laajojen kielimallien käyttämistä ohjelmointivirheiden aiheuttamien virheilmoitusten parantamisessa. Työn tavoitteena on koota olemassa oleva tutkimustieto laajojen kielimallien käytöstä virheilmoitusten parantamisessa ja arvioida niiden käyttöönoton mahdollisuutta ohjelmoinnin opetuksessa. Työssä tarkastellaan tekijöitä, jotka vaikuttavat laajojen kielimallien tuottamien virheilmoitusten selitysten tarkkuuteen. Tärkein tekijä on käytetty kielimalli, mutta myös kielimallin lämpötila, annettu syöte sekä käytetty ohjelmointikieli vaikuttavat merkittävästi vastausten tarkkuuteen. Tulokset osoittavat, että laajoilla kielimalleilla parannetut käännösaikaiset virheilmoitukset ovat riittävän laadukkaita ja että opiskelijat hyötyvät niiden käytöstä. Kuitenkin laajempaan käyttöönottoon tarvitaan lisätutkimusta siitä, miten kielimallien tuottamat selitykset vaikuttavat opiskelijoiden oppimiseen sekä ajonaikaisten laajoilla kielimalleilla parannettujen virheilmoitusten laadusta.Item Explainable empirical risk minimization in high-stakes decision making(2024-09-18) Frank, Pauli; Jung, Alexander; Perustieteiden korkeakoulu; Savioja, LauriExplainability is a major aspect of gaining trust in machine learning (ML) models. Explainable machine learning (XML) helps the user of an ML model to understand why the model made a certain prediction. The need for explainability grows as the use of ML models becomes more common in high-stakes decision making. High-stakes decisions are made in high-stakes fields. High-stakes fields are fields such as medicine and criminal justice, where individual decisions might have significant effects on, for example, a person’s life or an organization’s operation. Like many other fields, high-stakes fields have also adopted ML as a tool. This bachelor’s thesis is a literature review that analyzes the usability of explainable empirical risk minimization (EERM) in high-stakes decision making. EERM offers a new approach to explaining ML models. EERM generates a subjective, interpretable hypothesis that approximates an ML model. Subjectivity is gained through user signals that categorize the users of the model. Based on the research of the use of other explainability methods in high-stakes fields, it could be possible to apply EERM as a decision support tool in high-stakes decision making. Three major, unsolved problems of EERM that are discussed in this paper are the choosing of user signals, different applications of EERM in high-stakes fields, and balancing between accuracy and explainability. However, these problems must be studied further to better analyze the application possibilities of EERM in high-stakes decision making. EERM would also require humans to make the final decisions and analyze the predictions based on the generated hypothesis. EERM could also be used together with other explainability methods.Item Integrating Design Thinking into Agile software development from the viewpoint of Requirements Engineering(2024-09-06) Nahum, Shavit; Eloranta, Tuomo; Perustieteiden korkeakoulu; Savioja, LauriThis bachelor's thesis explored the integration of Design Thinking (DT) into Agile software development (ASD) from the viewpoint of Requirements Engineering (RE). Through a comprehensive literature review, the study revealed that successful integration between DT and ASD requires comparing and analyzing their similarities and differences. In addition, while Agile methodologies improve project success, inadequate RE practices often lead to project failures. DT, with its user-centered approach and focus on innovation, has the potential to address these shortcomings and enhance current RE practices, leading to better alignment with user needs. However, the methods for implementing the integration itself remain unclear, making the topic of significant importance given the increase in project failure rates due to ambiguities in RE processes and the growing popularity of ASD. Therefore, this thesis first compared the similarities and differences between DT and ASD, and then explored the integration strategies of DT into RE in ASD further. There are three key similarities and differences between DT and ASD that the literature agrees upon. The similarities include encountering Wicked Problems, close interaction between team members and users, and the collection and utilization of user feedback. The key differences lie in the process of problem discovery, prototypes, and the team dynamics and thinking styles of team members. Three integration strategies were identified in this thesis: Upfront DT, Infused DT, and Continuous DT. Upfront DT involves conducting a separate workshop before development to gather user insights, whereas Infused DT integrates DT techniques into existing RE processes throughout the development lifecycle. Continuous DT combines both strategies, starting with an Upfront DT strategy and continuing with Infused DT during development. Continuous DT was the most common strategy, highlighting companies’ desires to change the entire organizational mindset towards innovative solutions. Upfront DT is also widely used, emphasizing the importance of understanding user requirements early to avoid wasted efforts. Infused DT appears to be underexplored in the literature, despite its potential to enhance existing RE activities at a lower cost. This thesis highlights that while the benefits of integrating DT into RE in ASD are clear, challenges remain regarding the implementation of these strategies, thus more research is needed, particularly with Infused DT. Limitations of this thesis include inconsistent terminology in the literature and gaps in the used research papers that examined software development as a whole instead of focusing solely on ASD, as well as lack of studies examining real-life examples of the integration strategies.Item Startup-yritysten onnistumisen ennustaminen(2024-08-02) Lahti, Tomi; Kohonen, Jukka; Perustieteiden korkeakoulu; Kohonen, JukkaNopeasti kehittyvässä startup-ekosysteemissä startup-menestyksen tarkka ennustaminen on ratkaisevan tärkeää yrittäjille, sijoittajille ja päättäjille. Tässä työssä tarkastellaan maantieteellisen sijainnin, startupin toimialan sekä rahoituksen keskeisiä rooleja startup-yrityksen menestymisen ennustamisessa. Käytämmä määriteltyinä yrityskaupan saavuttamisena. Tutkimuksessa käytetään "Startup data" -aineistoa, eli tietoa, joka on peräisin tunnetusta sähköisestä alustasta nimeltä Kaggle. Aineisto sisältää tietoja yhdysvaltalaisista startup-yrityksistä ja niiden ominaisuuksista. Tutkimme, miten eri muuttujat vaikuttavat startup-yrityksen onnistumuseen eli onko startup "acquired" eli onnistunut tai "closed" eli epäonnistunut. Näiden ulottuvuuksien tutkimiseen käytämme logistista regressioanalyysiä, sekä mallin ennustuskysyn testaamiseen, ristiinvalidointia. Tuloksemme paljastavat, että vaikka maantieteellinen sijainti ja toimiala vaikuttavat merkittävästi startupin menestykseen, rahoituksen vaikutus, mukaan lukien riskipääoma, enkelisijoitukset ja eteneminen rahoituskierrosten (A-D) kautta, on kannattavaa. Tutkimuksessa korostuu näiden tekijöiden monimutkainen vuorovaikutus startupin menestyksen kanssa. Tämä tutkimus edistää ymmärrystä siitä, mikä ajaa startupin menestystä, ja tarjoaa oivalluksia, jotka voivat kertoa strategioista tukevien startup-ekosysteemien edistämiseksi. Tunnistamalla menestykseen vaikuttavat avaintekijät, tämä tutkielma luo pohjaa tulevaisuuden tutkimiselle ennakoiviin malleihin ja menetelmiin, ehdottaen keinoja kehittyneiden analyysien ja koneoppimisen sisällyttämiseksi ennusteiden tarkentamiseen edelleen.Item Two-point model and OEDGE simulations of the scrape-off layer for attached JET-ILW L-mode plasmas(2024-09-13) Kuivasniemi, Henri; Groth, Mathias; Perustieteiden korkeakoulu; Groth, MathiasDue to the severe nature of plasma-solid interactions, the separation of the two states of matter is of interest in fusion research. In diverted tokamaks, the core plasma is magnetically separated from the reactor vessel, with plasma-solid interactions restricted to a thin and cold region of plasma known as the scrape-off layer (SOL). The scrape-off layer is inherently connected to the erosion of the vessel's surfaces and transport of impurities to the core plasma, and as such, understanding the phenomena of the SOL is vital for developing fusion energy. Two 1D fluid models, the analytic two-point model and the numerical code OEDGE, are used to predict and understand the SOL of the JET ITER-Like-Wall tokamak in attached low-recycling and high-recycling conditions in low-confinement mode plasmas with boundary conditions to the models given by measurements from the low-field side (LFS) target. The sensitivity of the models on the boundary conditions is examined by considering different ion temperature profiles at the target, as well as different shifts to the position of the LFS strike point. Electron temperature from both models is found to agree strongly with measurements in both conditions. The limitations of 1D modeling are demonstrated for the plasma parameter electron density, for which in low-recycling conditions an unphysical dip towards the separatrix is predicted. In high-recycling conditions, the density at the separatrix is overestimated by a factor 2 due to the absence of radial transport mechanisms.