Sirvi Märksõna "machine learning" järgi

Nüüd näidatakse 1 - 20 84

listelement.badge.access-status Avatud juurdepääs ,
Agent Behavior Modeling in Roundabout Traffic
(Tartu Ülikool, 2022) Korp, Heidi; Yar Muhammad, Naveed Muhammad, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Autonomous vehicle (AV) industry has grown immensely in the last few years. Different aspects of assisted and autonomous driving, including perception, state estimation, motion planning etc. have received a lot of attention from the research and industrial community. Achievements in hardware industry have enabled to make real-time analysis about the situation in traffic. One of the major challenges that the AV industry faces today is understanding and predicting the behavior and future states of road users. Modeling such behaviors is not a trivial task and depends on multiple factors including traffic rules, the geometrical shape of the road, number of traffic participants etc. In this paper we propose two methods for predicting the future action of a vehicle that is about to enter the roundabout. The first method is based on the Recurrent Neural Network (RNN) architecture and aims to predict the destination of a vehicle. The second method uses the information about Surrounding Vehicles (SV) in addition to the Target Vehicle’s (TV) data to predict the course of action in terms of velocity. The results indicate that a correct assumption about the vehicle’s destination can be achieved in less than 0.4 seconds and that taking the SVs’ data into consideration is very helpful in modeling the vehicle’s future behavior.
listelement.badge.access-status Avatud juurdepääs ,
AI-Powered MMSE: Enhancing Cognitive Assessment
(Tartu Ülikool, 2024) Tenman, Konstantin; Gharib, Mohamad, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
The increasing incidence of dementia among the elderly highlights the critical demand for cognitive assessment tools that are both efficient and widely accessible. Traditional methods, such as the Mini-Mental State Examination (MMSE), are typically conducted in clinical settings using paper-based formats, limiting access due to resource constraints and needing trained professionals. This thesis addresses these challenges by converting the MMSE into an AI-powered, web-based application, allowing assessments to be completed at home with minimal non-professional assistance. The digital implementation leverages sophisticated artificial intelligence (AI) models, particularly the Llama 3.1:70B, for automating the administration of the MMSE. This makes it more consistent and sensitive to small changes in cognitive function. By leveraging Machine Learning (ML) and Natural Language Processing (NLP), the system improves the consistency, accuracy, and accessibility of cognitive assessments through web-based administration. Adopting the Design Science Research (DSR) framework, this study incorporates contemporary web technologies alongside a hybrid AI strategy, enhancing performance while safeguarding data privacy. In trials, the AI-powered MMSE achieved a 92.9% success rate in confirming response correctness compared to traditional methods and slightly higher user satisfaction despite longer administration times. While this work significantly improves cognitive assessment accessibility and sensitivity, further studies are needed to validate its effectiveness across diverse clinical settings. Future research should optimize response times, expand language support, and address ethical considerations in AI-driven cognitive assessments.
listelement.badge.access-status Avatud juurdepääs ,
Analyzing the relationships between crime and socio-economic and spatial factors using random forest: a case study of Tallinn
(Tartu Ülikool, 2024) Yu, Cheng-Wei; Uuemaa, Evelyn, juhendaja; Kalm, Kadi, juhendaja; Kmoch, Alexander, juhendaja; Zalite, Janis, juhendaja; Tartu Ülikool. Geograafia osakond; Tartu Ülikool. Loodus- ja täppisteaduste valdkond
The spatial factors of crime and its socioeconomic background are important topics in crime research. This study uses a grid framework to represent various spatial, environmental, and socioeconomic factors across Tallinn in 500-meter grids. The study aims to predict the number of crimes in each grid cell through a random forest machine learning model and identify the main contributing factors. Machine learning models do not explain causal relationships between variables but highlight possible correlations, so crime factors need to be discussed within Tallinn's context. Among various types of crime, the factor of commercial locations shows the strongest relationship with the number of crimes. These reflect the concentration of economic activities, assets, and the gathering of people, which are important conditions for crime motivations. Secondly, factors such as the number of renters and the population with low socioeconomic status are associated with the number of crimes against public order.
listelement.badge.access-status Avatud juurdepääs ,
Andmenihke tuvastamine ja leevendamine kõnekeskuse andmete näitel
(Tartu Ülikool, 2025) Himuškin, Desiree; Aan, Janika, juhendaja; Aljanaki, Anna, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
The aim of this master thesis is to analyse how call center worktime change affects the performance of created machine learning models, see if this change creates a data drift and how to mitigate its effect. First, literature, existing solutions on how to predict service level and data drift algorithms are analysed. Secondly, the call center dataset is described and analysed. Thirdly is given an overview of used models and their features, which is followed by practical work which includes describing the effects of the change on the data, testing two data drift algorithms on the dataset, trying to alleviate the effects of the change on prediction accuracy and finally comparing the results. In the final part, conclusions on the effect of the change are made based on the results of the algorithms and the usability of deploying these models in other call centers is discussed.
listelement.badge.access-status Avatud juurdepääs ,
Application of Machine Learning Techniques to Ensure Safer Work Environments in Estonia
(Tartu Ülikool, 2023) Käära, Mario; Chakraborty, Roshni, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Occupational accidents are a major global concern which results in significant human and economic losses. In Estonia, over 4, 000 work-related accidents are recorded annually, and 428 fatalities were reported between 2001 and 2021. For example, work-related accidents led to a loss of 141, 000 workdays and approximately e5.3 million in 2021. Several studies across different countries have recently proposed automated data analytic tools and machine learning based models to understand occupational hazards and predict the likelihood and severity of accidents. These applications can identify high-risk workers and ensure robust safety management systems across various industries, such as construction and manufacturing. However, these proposed models are not directly applicable to Estonia, and no specific tools can handle the local settings. Through this Thesis, we aim to develop automated models based on machine learning techniques to predict the severity of occupational accidents in Estonia. We also identify critical factors for different industries contributing to these accidents. Our dataset consists of 82, 641 work-related accidents, featuring 37 variables, and spans the period from 2002 to 2022. The Thesis demonstrates that the best-performing models, including Support Vector Machine and Logistic Regression, can predict accident severity and identify crucial factors for targeted prevention strategies. The primary outcomes include critical insights into the important factors and the development of tailored machine learning models for occupations in specific economic sectors. Therefore, we propose accurate and efficient automated tools that can handle the inherent data challenges and ensure the significance of targeted modelling in accident prevention. The Thesis illustrates the potential of understanding the data patterns, developing specific data analytic tools and machine learning algorithms to improve decision-making in workplace safety and developing cost-effective prevention strategies.
listelement.badge.access-status Avatud juurdepääs ,
Automatic Speech-based Emotion Recognition
(Tartu Ülikool, 2018) Hook, Joosep; Anbarjafari, Gholamreza, supervisor; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Tehnoloogiainstituut
The main objectives of affective computing is the study and creation of computer systems which can detect human affects. For speech-based emotion recognition, universal features offering the best performance for all languages have not yet been found. In this thesis, a speech-based emotion recognition system using a novel set of features is created. Support vector machines are used as classifiers in the offline system on Surrey Audio-Visual Expressed Emotion database, Berlin Database of Emotional Speech, Polish Emotional Speech database and Serbian emotional speech database. Average emotion recognition rates of 80.21%, 88.6%, 75.42% and 93.41% are achieved, respectively, with a total number of 87 features. The online system, which uses Random Forests as it’s classifier, consists of two models trained on reduced versions of the first and second database, with the first model trained on only male samples and the second trained on both. The main purpose of the online system was to test the features’ usability in real-life scenarios and to explore the effects of gender in speech-based emotion recognition. To test the online system, two female and two male non-native English speakers recorded emotionally spoken sentences and used these as inputs to the trained model. Averaging over all emotions and speakers per model, it is seen that the features offer better performance than random guessing, achieving 28% emotion recognition in both models. The average recognition rate for female speakers was 19% in the first and 29% in the second model. For male speakers, the rates were 36% and 28%, respectively. These results show how having more samples for training for a particular gender affects emotion recognition rates in a trained model.
listelement.badge.access-status Avatud juurdepääs ,
Bakterite eristamine fluoromeetri spektrist masinõppe abil
(Tartu Ülikool, 2024) Rõõm, Rimmo; Rebane, Ott, juhendaja; Aljanaki, Anna, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
In this master thesis, the most suitable machine learning solution is found for the fluorometer device H2B-Spectral developed by LDI Innovation OÜ. The machine learning methods tested in this thesis aim to improve the differentiation of various microorganisms on selected solid surfaces. The device functions as a multi-channel fluorometer, exciting the measured sample surface with three different ultraviolet wavelengths and reading the emitted optical fluorescence signal on three different wavelength channels. Based on the obtained eight number data (one channel provides no information), the sensor's software must classify the measurement point into pre-learned classes. In this study, over thirteen classes of various microorganisms are measured, and different machine learning methods (including decision tree, random forest, KNN, support vector machine, ensemble voting) are compared for their classification performance. The most effective classification method identified in this study will be implemented in the standard machine learning system in the software for H2B-Spectral.
listelement.badge.access-status Avatud juurdepääs ,
Benchmarking Energy and Performance of Standard Machine Learning Libraries: An Empirical Study
(Tartu Ülikool, 2025) Kont, Kadri-Ketter; Anwar, Hina, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Programmeerimiskeele ja teegi valiku mõju masinõppeülesannete energiakulule ei ole seni põhjalikult uuritud. Lõputöö eesmärk oli võrrelda kolme masinõppes levinud programmeerimiskeelt ja nende teeke, keskendudes energiatõhususele, käitusajale ja mudeli täpsusele. Klassifitseerimisülesande andmestiku põhjal rakendati standardteekide abil igas keeles viis masinõppealgoritmi. Uurimuses analüüsiti, kuidas keele ja teegi valik mõjutab energiatarbimist, käitusaega ja täpsust, ning uuriti nende näitajate vahelisi kompromisse. Tulemuste kinnitamiseks viidi läbi statistiline analüüs, mille abil võrreldi algoritmide energiakasutust ja jõudlust erinevate implementatsioonide puhul. Töö tulemusena tehti ülevaade, kuidas standardteekide valik mõjutab masinõppe algoritmide energiatõhusust.
listelement.badge.access-status Avatud juurdepääs ,
Breimani pügamisteoreemi üldistus
(Tartu Ülikool, 2025) Jesse, Mihkel; Lember, Jüri, juhendaja; Alasoo, Kaur, juhendaja; Riet, Ago-Erik, juhendaja; Tartu Ülikool. Matemaatika ja statistika instituut; Tartu Ülikool. Loodus- ja täppisteaduste valdkond
Klassifikatsiooni- ja regressioonipuud (CART-puud) on masinõppemeetod. Käesolevas bakalaureusetöös käsitletakse CART-puude kasvatamise, pügamise ja rakendamisega seonduvat teooriat ning üldistatakse Breimani pügamisteoreemi, lisades uudse karis tusliikme. Tavaliste CART-puude puhul kasutatakse riskiliiget tükeldamisotsuse hindamiseks, pakutud karistusliikme abil hoitakse tükeldamisel sarnaseid elemente koos. See lähenemine võimaldab CART-puude rakendamist ka klasterdusülesannetes.
listelement.badge.access-status Avatud juurdepääs ,
Calibration of Multi-Class Probabilistic Classifiers
(Tartu Ülikool, 2022) Valk, Kaspar; Kull, Meelis, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Classifiers, machine learning models that predict probability distributions over classes, are not guaranteed to produce realistic output. A classifier is considered calibrated if the produced output is in correspondence with the actual class distribution. Calibration is essential in safety-critical tasks where small deviations between the predicted probabilities and the actual class distribution can incur large costs. A common approach to improve the calibration of a classifier is to use a hold-out data set and a post-hoc calibration method to learn a correcting transformation for the classifier’s output. This thesis explores the field of post-hoc calibration methods for classification tasks with multiple output classes: several existing methods are visualized and compared, and three new non-parametric post-hoc calibration methods are proposed. The proposed methods are shown to work well with data sets with fewer classes, managing to improve the stateof- the-art in some cases. The basis of the three suggested algorithms is the assumption of similar calibration errors in close neighborhoods on the probability simplex, which has been previously used but never clearly stated in the calibration literature. Overall, the thesis offers additional insight into the field of multi-class calibration and allows for the construction of more trustworthy classifiers.
listelement.badge.access-status Avatud juurdepääs ,
Causality Management and Analysis in Requirement Manuscript for Software Designs
(Tartu Ülikool, 2023) Oluyide, Olumide Olugbenga; Gambo, Ishaya Peni, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
For software design tasks involving natural language, the results of a causal investigation provide valuable and robust semantic information, especially for identifying key variables during product (software) design and product optimization. As the interest in analytical data science shifts from correlations to a better understanding of causality, there is an equal task focused on the accuracy of extracting causality from textual artifacts to aid requirement engineering (RE) based decisions. This thesis focuses on identifying, extracting, and classifying causal phrases using word and sentence labeling based on the Bi-directional Encoder Representations from Transformers (BERT) deep learning language model and five machine learning models. The aim is to understand the form and degree of causality based on their impact and prevalence in RE practice. Methodologically, our analysis is centered around RE practice, and we considered 12,438 sentences extracted from 50 requirement engineering manuscripts (REM) for training our machine models. Our research reports that causal expressions constitute about 32% of sentences from REM. We applied four evaluation metrics, namely recall, accuracy, precision, and F1, to assess our machine models’ performance and accuracy to ensure the results’ conformity with our study goal. Further, we computed the highest model accuracy to be 85%, attributed to Naive Bayes. Finally, we noted that the applicability and relevance of our causal analytic framework is relevant to practitioners for different functionalities, such as generating test cases for requirement engineers and software developers and product performance auditing for management stakeholders.
listelement.badge.access-status Avatud juurdepääs ,
Classification of human Y chromosome haplogroups based on dense and sparse genetic data using machine learning approaches
(Tartu Ülikool, 2022) Espinosa, Jose Rodrigo Flores; Roy, Kallol, juhendaja; Karmin, Monika, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
The genetic data of human Y chromosomes is classified into haplogroup categories based on the underlying phylogenetic tree, where a haplogroup represents a monophyletic clade on the tree. Current methods for the assignment of these categories work by representing a known human Y chromosome phylogeny as tree data structure. For an individual Y chromosome to be assigned a haplogroup using this representation, strategies based on breadth-first search (BFS) are often used. The tree is traversed in a manner that paths showing supporting evidence from mutations are further explored eventually leading to a leaf node and final classification. This strategy shows high efficiency when dense genotyping/sequencing data are available. However, in case of lower density genetic data such as genotyping arrays or ancient DNA data, BFS-based strategies often fail to reach a leaf node due to uncertainty and lack of information of where to go next. In this work we leverage the increasing availability of world-wide panels of Y chromosome data with available curated haplogroup categories. We present a novel method on the application of a K-nearest neighbors classifier to both low-density and high-density types of data. The main goal is to assess the extent to which this approach can be useful in the challenging cases where BSF-based methods fail to produce a tractable and meaningful result. To achieve this, we have employed different DNA sequence encodings together with dimensionality reduction techniques. We have also investigated a novel method of DNA representation using Word2vec contextual embeddings. The DNA snippets are represented as text words and the whole DNA sequence is a text sentence. Encoding the DNA sequences in this manner gives rich contextual information that helps in haplogroup classification and can be extended to other applications in genomics. The results show that classification accuracy is high (>98%) with next-generation sequencing (NGS) and genotyping arrays, high-density and lower-density data classes respectively. Performance however is low (<60% on average) when classifying ancient DNA data, which has the lowest level of resolution and higher levels of error. We observe that in many of the challenging cases KNN fails to correctly predict the label at its finest degree of resolution but does classifies correctly at the main category level which can be useful in practice.
listelement.badge.access-status Avatud juurdepääs ,
Comparison of category-level, item-level and general sales forecasting models
(Tartu Ülikool, 2020) Ruusmann, Laura; Dumas, Marlon, juhendaja; Muuli, Eerik, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Sales forecasting is the process of estimating future sales. In this thesis, multiple methods are tested out for achieving best forecasting accuracy with lowest computational requirements. Three families of methods are investigated: a traditional statistical forecasting approach (ARIMA), classical machine learning techniques (specifically ensemble methods) and a third one based on deep learning methods (specifically recurrent neural networks with LSTM architectures). The study uses real-world sales transaction data from a large retail company in a Baltic country and the aim of this thesis is to improve their current sales forecasting system. Here we show that improving on their current sales forecasting is possible and additionally analyse the influence of promotional sales to prediction accuracy. The results show that using a combination of multiple item-level decision tree-based ensemble models yields the best prediction accuracy with regard to training complexity. Additionally, when comparing accuracy of forecasts for promotional sales and non-promotional sales, a variant of ARIMA achieves the most accurate results when forecasting promotional sales.
listelement.badge.access-status Avatud juurdepääs ,
Computer vision meets microbiology: deep learning algorithms for classifying cell treatments in microscopy images
(Tartu Ülikool, 2023) Zeynalli, Ali; Fishman, Dmytro, juhendaja
Cell classification is one of the most complex challenges in cellular research that has significant importance to personalised medicine, cancer diagnostics and disease prevention. The accurate classification of cells based on their unique characteristics provides valuable insights into a patient's health status and in guiding treatment decisions. Thanks to recent technological advancements, cellular research has experienced significant progress in the use of deep learning and has become a valuable tool for tackling complicated tasks such as cell classification. In this study, we explored the capability of state-of-the-art deep learning models such as ResNet, ViT and Swin Transformer to automatically classify brightfield and fluorescent microscopy images across single and multiple channels into four cell treatments: Palbociclib, MLN8237, AZD1152, and CYC116. The results have revealed that Swin Transformer surpasses the other models for cell treatment classification on multi-channel fluorescent and brightfield images, achieving the highest accuracy of 86% and 59%, correspondingly. However, the highest accuracy achieved on single-channel brightfield images was 61%, using the ResNet-50 model. The previous research has shown that combining multiple channels yields better performance which necessitates further investigation into the capacity of deep learning models for automating the cell treatment classification of single- and multi-channel brightfield microscopy images.
listelement.badge.access-status Avatud juurdepääs ,
Creating prediction models for cervical cancer forecasting
(2022) Jerina, Valerija; Kolde, Raivo, juhendaja; Tartu Ülikool. Matemaatika ja statistika instituut; Tartu Ülikool. Loodus- ja täppisteaduste valdkond
The aim of this bachelor’s thesis is to create prediction models for cervical cancer (ICD-10 C53) and pre-cancerous condition (ICD-10 R87.613) forecasting. The analysis is based on health data of 10% of Estonian population that was provided by STACC OÜ. The thesis gives an overview on cervical cancer, shows which prediction models were created using different machine learning algorithms, evaluates their performance, and gives an overview on factors that might affect risk of getting the diseases.
listelement.badge.access-status Avatud juurdepääs ,
Creation of Materials to Teach Data Science via Self-Driving
(Tartu Ülikool, 2024) Kreegipuu, Artur; Tampuu, Ardi, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Andmeteaduse projektides esineb tihti kitsaskohti, millega õpilased võivad olla teoreetiliselt tuttavad, kuid puuduvad praktiliselt kogemused. Sageli esinevaid probleeme saab demonst-reerida isejuhtivate mudelautode abil. Praktiliste ülesannete kaudu saavad õpilased kogeda, kuidas isejuhtivate närvivõrkude arendamise etapis tehtavad andmeteaduslikud vead mõju-tavad isejuhtiva mudeli sooritusvõimet. Lõputöö raames loodi isejuhtivate mudelautode abil praktilised õppematerjalid eesmärgiga panna õpilased mõistma, ära tundma ja ennetama andmeteaduses laialdaselt levinud probleeme. Selleks loodi probleeme demonstreerivad praktilised ülesanded, mille käigus tuleb õpilastel koguda andmeid, treenida isejuhtivaid närvivõrke ja katsetada loodud mudeleid rajal. Kõik ülesanded lahendati testimise eesmärgil erinevates valgusoludes ja iga ülesande eeldatud tulemist filmiti video. Loodud õppemater-jalidele koguti tagasisidet kahelt masinõppe eksperdilt.
listelement.badge.access-status Avatud juurdepääs ,
Crowdsourced Perception of Neighbourhood Safety: A Case Study Using Airbnb Reviews
(Tartu Ülikool, 2025) Rodríguez Sánchez, María Alejandra; Cabral Pinheiro, Victor Henrique, juhendaja; Hadachi, Amnir, juhendaja; Sowinska, Katarzyna, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
Käesolev magistritöö uuris rahvahankeandmete potentsiaali linnakeskkonna turvatunde tajumise kaardistamisel, keskendudes Airbnb arvustustele kui subjektiivse info allikale. Töö eesmärgiks oli hinnata, kas need arvustused sisaldavad tähenduslikku teavet naabruskonna kohta ja kas turvatunnetus varieerub linna eri osades. Uuring viidi läbi Stockholmi näitel ning analüüs jagunes kolme etappi: naabruskonnateemaliste arvustuste tuvastamine, meeleoluanalüüs ning ruumiline ja võrdlev analüüs. Esimeses etapis testiti erinevaid järelevalveta klasterdusmeetodeid, et eristada naabruskonna-teemalisi ja muid arvustusi. Klasterdustulemuste piirangute tõttu töötati välja heuristiline filtreerimisviis, kasutades TF-IDF vektoriseerimist, SBERT-i ja koosinussarnasust. Teises etapis rakendati kolme meeleoluklassifikaatorit (VADER, TextBlob ja SBERT-i põhine klasterdus), et määrata arvustuste sentiment positiivse, neutraalse või negatiivsena. Kõigis mudelites ilmnes valdavalt positiivne hoiak, kuid mudelite täpsus varieerus. Kolmandas etapis koondati tulemused kuusnurkse ruudustiku abil, mis asetati Stockholmi kaardile, et uurida ruumilisi mustreid, ning seejärel võrreldi neid politseiraportite andmetega Spearmani korrelatsioonanalüüsi abil. Tulemused näitasid nõrka ja statistiliselt ebaolulist seost tajutud turvalisuse ja registreeritud kuritegude vahel, viidates sellele, et Airbnb arvustused ja ametlikud andmed kajastavad linna erinevaid kogemuslikke tahke. Töö rõhutab kasutajapõhise sisu väärtust subjektiivsete linna tajude kaardistamisel, kuid toob esile ka selle piirangud võrreldes ametlike turvalisuse näitajatega.
listelement.badge.access-status Avatud juurdepääs ,
Deepfakes for Paper Vote Privacy Defence
(Tartu Ülikool, 2025) Habanen, Anette; Villemson, Jan, juhendaja; Laur, Sven, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
The recent rise of artificial intelligence (AI) solutions has also had a significant impact on electoral processes. Most notably, deepfakes created by generative AI applications can (and have been) used to spread misinformation during the campaigns, but they can also be used for cyberattack automation, biased social media bots, etc. This thesis instead presents a positive use case for generative AI in manipulating video material required as proof of voting by potential coercers. For this, I have created a pipeline that takes a video of a voting ballot and replaces its critical content (in our case, the digits on the ballot). To achieve this, a YOLO model is used to find the digits, a WavePaint image inpainting model is used to cover up the old digits, and a separate image of the new digits is used to place it into the video. Additionally, I have implemented the prototype application in the form of a webpage.
listelement.badge.access-status Avatud juurdepääs ,
Depressiooni ja ärevuse automaatne tuvastamine spontaansest kirjalikust keelest: andmete kogumise pilootuuring
(Tartu Ülikool, 2019) Sirts, Kairit; Akkermann, Kirsti, juhendaja; Tartu Ülikool. Sotsiaalteaduste valdkond; Tartu Ülikool. Psühholoogia instituut
Uurimistöö eesmärgiks oli välja töötada meetod tekstilise andmestiku kogumiseks, mille alusel saaks hiljem arendada masinõppel põhinevaid meetodeid depressiooni ja ärevuse riski automaatseks hindamiseks. Töö käigus koostati ankeet, mille abil koguti tekstilist materjali ligi 300-st vabatahtlikust koosnevalt mugavusvalimilt. Kogutud tekstid sisaldasid nii etteantud pildi kirjeldust kui ka vabalt valitud sündmuse või mälestuse kirjeldust. Valimis osalenute emotsionaalset seisundit mõõdeti EEK-2 skriiningtesti abil. Ligi 42% isikutest ületas depressiooni ning ligi 30% isikutest ärevuse alaskaala riskilävendi. Esialgsed eksperimendid masinõppe mudelitega, mis püüdsid ennustada, kas inimese EEK-2 skoor ületab depressiooni ja/või ärevuse riskilävendi, edukaid tulemusi ei andnud. Kokkuvõttes tundub, et etteantud pildi kirjeldamine ei ole sobivaim viis soovitud andmestiku kogumiseks ja pigem peaks kasutama selliseid kirjutamise ülesandeid, mis oleks inimese endaga rohkem seotud.
listelement.badge.access-status Avatud juurdepääs ,
Developing a Volleyball Game with an AI Opponent Using Reinforcement Learning
(Tartu Ülikool, 2021) Marran, Tanel; Matiisen, Tambet, juhendaja; Tartu Ülikool. Loodus- ja täppisteaduste valdkond; Tartu Ülikool. Arvutiteaduse instituut
This thesis integrates reinforcement learning into a game development project by creating a competitive volleyball game, where the user can play against an artificial intelligence (AI) trained using reinforcement learning techniques. The work elaborates on what reinforcement learning is, brings forth some of the challenges of adding machine learning to a game, describes the development environment Unity and its machine learning package ML-Agents as well as analyzes the finished game and its AI.