2025-03-22
00:00 |
DNS User Fingerprint
SAYED, Karim
In today's digital world, maintaining ones privacy has become a more difficult task. This paper explores whether advanced machine learning models can be utilized to fingerprint users can identify users based solely on data collected from a DNS server and to evalu- ate the aspects of a users behavior that can be used to identify them, compromising their privacy. By focusing on behavioral patterns and using feature engineering techniques along side time-series modeling of the users data, this work examines the privacy risks that can arise with DNS-based identification. This work will employ methods like mutual information, variance inflation factor (VIF), and Pearson correlation to select key features, testing them across models both sequential and non-sequential in an attempt to to identify the users. This research makes use of a publicly available dataset detailed in the paper "A user DNS fingerprint dataset", ensuring that the work can be compared with future studies and serve as a foundation for further research. Through time-series modeling and feature dimensionality reduction, an LSTM model was produced that is able to make predictions at 94% accuracy using only a subset of the pro- vided features. Additionally, non-temporal models were used to identify users at 79% accuracy and provide insights to what features impact the predictions the most.
Úplný záznam
|
2025-03-22
00:00 |
Enhancing Electric Vehicle Battery Lifetime - Ageing Influences and Degradation Forecasts
CHANDRABHANU NAMBIAR, Gokul
The broadening shift towards Electric Vehicle (EV) is essential for reducing global carbon emissions, yet the longevity and reliability of High Voltage Battery (HVB) that drives the vehicle, remain pivotal challenges hindering wider EV adoption. This thesis works on a deep dive analysis focused on identifying and quantifying the impact of various stress factors on the State of Health (SoH) of EV batteries. Starting off this research is the deployment of machine learning models, specifically Gradient Boosting Regressor (GBR) and Artifcial Neural Network (ANN), to forecast the degradation in SoH based on a rich dataset encapsulating real-world operational parameters of EV batteries. The collected features which encompasses battery pack temperature, voltage, vehicle mileage, along with detailed charging and discharging information, form the foundation of the analysis. Utilizing SHapley Additive exPlanations (SHAP) for interpretability, the study elaborates on the significance and influence of each of the features on battery SoH, offering a detailed understanding of extent of battery degradation. The findings not only highlight the paramount importance of some key charging and operational behavior parameters, but also pave the way for formulating actionable recommendations to reduce battery wear and possibly extend lifespan. By providing empirical insights into battery stress factors and predictive modeling of battery SoH degradation, this thesis contributes valuable knowledge towards the enhancement of Battery Management Systems (BMS) and promotes more sustainable EV usage patterns.
Úplný záznam
|
2025-03-22
00:00 |
Generating Semantic Networks from Natural Language using BERT
GRIFFO DUARTE, Raphael
This master thesis proposes SemNet, a new method for constructing semantic networks from natural language text. It compares SemNet to the benchmark model Netts, demonstrating comparable performance while highlighting key differences attributed to specific modules within Netts. SemNet excels in efficiency and scalability, particularly when leveraging GPU acceleration, making it a promising foundation for subsequent natural language processing tasks. SemNet effectively utilizes a SRL BERT model however, limitations in capturing complex relationships beyond verb-based interactions were identified. A comprehensive literature review helped define the model's architecture, leading to the selection of BERT as the optimal foundation given the balance of size and scalability compared to large language models and supervised learning approaches.
Úplný záznam
|
2025-03-22
00:00 |
Aspekty sociální distance ve vztahu sociální pracovník-klient
DONÉEOVÁ, Hana
Disertační práce se zaměřuje na aspekty sociální distance ve vztahu mezi sociálními pracovníky a jejich klienty v domovech pro seniory a v azylových domech. Cílem je prezentovat příčiny a podmínky fenoménu sblížení v jejich vztahu a analyzovat jeho dopady na kvalitu poskytovaných služeb. V rámci teoretických východisek je představen význam sociální práce jako profese a zároveň je zde zdůrazněna důležitost empatie, respektu a stanovení profesních hranic ve vztahu sociální pracovník-klient. Na teoretický rámec práce navazují tři výzkumné studie, kdy první z nich odhaluje za využití kvantitativní metody sběru dat, že sblížení je běžné a ovlivňuje kvalitu služeb. Výsledky z této výzkumné studie posloužily jako rámec pro prohlubující výzkumné studie, jež se zabývaly konkrétními zjištěními příčin a podmínek vzniku tohoto sblížení mezi sociálním pracovníkem a klientem. Na základě výsledků bylo zjištěno, že sociální pracovníci často používají sblížení jako strategii pro zvládání emocionálně náročných situací a že rodinné zázemí a profesní zkušenosti mají významný vliv na tento proces. V souvislosti s těmito studiemi bylo rovněž upozorněno na potřebu udržování profesních hranic a etických principů, aby nedošlo k tzv. syndromu vyhoření u osoby sociálního pracovníka. Zároveň byla v práci zdůrazněna důležitost otevřené diskuse a vzdělávání v oblasti příčin a podmínek vedoucích ke sblížení, aby se minimalizovala potenciální rizika pro sociální pracovníky i klienty.
Úplný záznam
|
2025-03-15
00:01 |
Ecological interactions of herbivory and predation in various terrestrial ecosystems
MARAIA, Heveakore
My thesis investigates ecological interactions, especially herbivory and predation related to herbivorous insect along gradients in various terrestrial ecosystems - including islands vs. mainland, along latitudinal gradient in tropical and temperate forests, and in their vertical gradients, and finally in fire-mammal dominated ecosystems in subtropical savanna. The thesis examines tri-trophic interactions, specifically within plants, arthropod herbivores, and insectivorous predator dynamics, to explore how both biotic (top-down as well as bottom-up) and abiotic factors influence herbivory patterns. Through manipulative experiments, multi-site and multi-trophic studies, the research provides novel and valuable insights into the roles of plant defences, clines in herbivory, herbivory rate and accumulation and seasonality affecting herbivory damage. The first chapter (Chapter I) evaluates the effects of vertebrate and invertebrate predators on herbivory across elevational gradients on island vs. mainland, where plant defences are expected to play different roles. Chapter I reveals that vertebrate predators are more effective in controlling herbivory damage on islands than on the mainland. The second chapter (Chapter II) examines how seasonality, microclimate, and leaf traits influence insect herbivory along a vertical gradient in a tropical rainforest of Papua New Guinea. The novelty of this chapter is in the focus on herbivory rate on marked leaves, revisited for nearly a year, along a complete vertical gradient in tropical forest. The third chapter (Chapter III) investigates insect herbivory in a South African savanna-forest mosaic, highlighting how herbivory caused by chewing and mining insects varies with plant traits, mammal herbivores' density, and fire frequency across rainfall gradients. Very importantly, this chapter represents one of very few studies focusing on insect herbivory in savanna, finding it to be an important yet neglected topic. The final chapter (Chapter IV) explores insect herbivory across a latitudinal gradient, emphasising the varying effects of seasonality in temperate and tropical forests. Chapter IV represents a robust study rejecting the hypothesis, that insect herbivory is stronger/higher in tropical than in temperate forests. In Chapter IV, I argue that difference needs to be made between studies surveying standing herbivory vs. herbivory rate. Overall, my thesis study enhances the understanding of complex ecological interactions between herbivorous insect and plants and deepens our knowledge of how environmental factors shape the interplay between them and between the predators of insect, and the overall role of herbivory of terrestrial ecosystems. Besides other contributions to science, the thesis provides measures of standing herbivory damage and herbivory rates for many woody plant species, from various parts of the globe, representing thus another valuable contribution to other researchers.
Úplný záznam
|
2025-03-15
00:01 |
Community ecology and genetics of bird - parasite interactions
DAMNJANOVIC, Dragomir
This thesis investigates the interplay between genetic diversity, population structure, and haemosporidian parasite diversity in avian host-parasite systems, providing novel insights into ecological and evolutionary dynamics. Chapter I focuses on the genetic diversity and population structure of red- and white-spotted bluethroat subspecies, with a particular emphasis on the endangered, red-spotted population in the Krkonoše Mountains. Using genomic tools, this chapter highlights the genetic challenges faced by small, isolated populations and their implications for conservation. Chapter II employs next-generation sequencing to reveal a previously underexplored diversity of haemosporidian parasites in Eurasian bluethroat subspecies. It highlights geographic variation in parasite prevalence and identifies novel lineages, offering a deeper understanding of host-parasite co-evolution and its implications for subspecies differentiation. Chapter III expands the focus to Southeast Asia, characterizing haemosporidian diversity in three sympatric passerines: the Little Spiderhunter, White-rumped Shama, and Blue-winged Pitta. This chapter uncovers unique parasite-host associations in tropical ecosystems and contributes to understanding malaria's biogeographic patterns. Chapter IV investigates the diversity and selection pressures on key immune genes (TLR3, TLR4, and MHC class I exon 3) in avian hosts, elucidating their role in host-parasite co-evolution. By linking immunogenetic variation with parasite infection, it demonstrates how genetic diversity influences host survival and resilience. Finally, Chapter V synthesizes the findings, providing insights into the ecological, evolutionary, and conservation implications of host-parasite interactions.
Úplný záznam
|
2025-03-08
00:01 |
Úplný záznam
|
2025-03-08
00:01 |
Community assembly and population genetic structure of moths along elevational gradient in a tropical rainforest
IBALIM, Sentiko
This Thesis examines the community assembly and population genetic structure of moth communities along Mt. Wilhelm elevational gradient (50-3700 m a.s.l) in Papua New Guinea. The Thesis utilizes an extensively sampled DNA molecular barcode data (mtDNA-CO1) obtained through barcoding and metabarcoding techniques. First, we retrieved phylogenetic structure of the moth family Geometridae and used this to assess how abiotic and biotic factors influence the community assembly. We then expanded this study to determine the diversity trends of all moth families using a metabarcoding approach, making this the first ever large-scale survey of Lepidoptera in the region. We recorded nine common superfamilies, 49 families, 893 genera and 2,322 species from 11.5 million metabarcode reads. We described their diversity trends at super familial and familial levels and assessed alpha and beta diversity trends for species with increasing elevation. Finally, we explored the population structure of four common and widespread moth species sequenced through metabarcoding. We used haplotype networks and the FST fixation indexes to investigate gene flow among nine surveyed moth communities and their historical population demographics through Bayesian Skyline Plots (BSP) and infer the implications for conservation.
Úplný záznam
|
2025-03-08
00:01 |
Úplný záznam
|
2025-03-08
00:01 |
Úplný záznam
|
|
|
|
host ::
