En ontologi representation för vaccinationstäckning, säkerhet och biverkningar, och vaccineffektivitet för realtidsutvärderingar
Elise Lång
Avancerad (30hp)
kl 08:30, John Von Neumann (På svenska)
[Abstract]Data used when studying vaccination programs and vaccine evaluations are collected from multiple, and constantly changing data sources. The multiple data sources can collectively result in a silo formation and cause hindrance when researchers and public health organizations are conducting these studies. A solution for this hindrance are well-designed ontologies. Ontologies are becoming increasingly relevant with the massive increase of data online. Equally relevant is handling ontologies according to the FAIR principles, since it is considered the best practices of handling scientific data.
This thesis presents a design process for modeling an ontology in order to represent vaccine coverage, safety and adverse events, and vaccine effectiveness for real-time evaluation when existing ontologies are taken into consideration. The design process was inspired by the METHONTOLOGY method and the \acrshort{xd} method. The scope was specified via three stories each representing one aspect (vaccine coverage, safety and adverse events, and vaccine effectiveness), and relevant terms were gathered via knowledge acquisition. They were both validated by a domain expert before integration and implementation of each story, forming a module, were performed. In the integration and implementation, noticeable impact of the FAIR principles was seen since FAIRsFAIR's 17 recommendations were followed. The impact of the guidelines suggested foundational ontologies and reuse of ontologies which also follow the FAIR principles should be used to the extent possible. The modeled ontology was then evaluated using three procedures: internal evaluation, external evaluation, and FAIRness. Both the internal and external evaluation regarded the quality criterion accuracy and coverage. The internal evaluation included verification of requirements, also the coverage, and predicting the accuracy by semantic variance. The external evaluation was conducted by auditing, where four domain experts rated pairings and records where the acceptance of the pairings were significantly higher compared to the records. The external evaluation was also performed in two manners. In the first manner, the pairings and records were presented text-based. In the second manner, the pairings and the records were instead presented by reading them aloud and displaying them on a visualization created with OWLGrED. Where there is evidence suggesting the visual auditing receiving higher scores compared to the text-based auditing. Lastly, the FAIRness of the modeled ontology was evaluated when two ontology practitioners rated on a Likert scale to what degree the modeled ontology follows FAIRsFAIR's 17 recommendations. |
A Declarative Approach to GraphQL Schema Wrapping
Markus Gustafsson, Pontus Jarnemyr
Avancerad (30hp)
kl 16:00, Muhammad al-Khwarizmi (In English)
[Abstract]In recent years, GraphQL has become a popular Web API language. Due to GraphQL’s popularity, third-party libraries have also grown steadily alongside it. One such library is GraphQL Tools, which provides tools such as query delegation and schema wrapping. One core element of GraphQL is its schema. In part, the schema describes the relations between data in a data source. Every GraphQL server needs a schema, which in turn utilizes resolver functions that retrieve the data from the data source.
Schema wrapping allows developers to create GraphQL schemas that delegate queries to another remote server. This allows the user to modify the behavior of the remote schema, without modifying its contents. However, some JavaScript knowledge is required to implement the schema wrapping. This thesis proposes a declarative approach to wrapping schemas with the help of custom GraphQL directives. The main objective of the approach is to enable less knowledgeable developers to wrap schemas with less effort.
The declarative approach was tested by implementing a prototype. The effects on performance was tested with a benchmarking tool called Linköping GraphQL Benchmark (LinGBM). The prototype delegated queries to a remote server also created by LinGBM. The benchmarking tool was run against both the wrapper server and the remote server. The results indicated that the overhead of the wrapper server was mostly due to internal functions of GraphQL Tools. Therefore, automatically generating a server with the proposed declarative approach did not seem to impose significant performance losses. Considering the decreased programming effort by using the declarative approach, the slight performance loss can be seen as manageable. |
Towards better usability and design in APIs for distributed user interfaces
Johanna Palmkvist
Avancerad (30hp)
kl 10:15, John Von Neumann (På svenska)
[Abstract]In this paper, design decisions to improve usability in APIs for implementations of DUI applications has been investigated. Namely what abstractions an API like that should provide to the programmer, and how those abstractions should be presented syntax-wise to the programmer in the API. The exploration was done with a literature study, development of a game prototype with a DUI API, and three user sessions where an interview, and a sketch task was done with the participating developers. |
Automatic Testing and Validation of Level of Detail Reductions Through Supervised Learning
Olivia Shamon
Avancerad (30hp)
kl 15:15, IDA Donald Knuth (SaS) (In English)
[Abstract]The gaming industry is a multibillion-dollar industry, and by the end of 2022, there will
be 3.2 billion gamers worldwide. At the same time, as development time and costs must
be kept to a minimum, players demand every new game to be bigger and better than the
one before it. As demand for better and more visually appealing video games increases,
so does the complexity and size of these video games. To create a simple scene in a video
game, several thousands of assets are usually used, with each asset being extremely detailed,
consisting of millions of polygons, which means that the number of polygons that
must be drawn in a single frame of a video game can be in the order of several billions.
However, the number of polygons that can be displayed is limited by the available amount
of computer resources. LOD (Level of Detail) refers to the degree of complexity of a 3D
model representation. An asset’s LOD within the world of 3D video games is related to an
asset’s polygon count. In order to increase performance, one can create several versions of
the same asset with reduced polygon counts, which is referred to as LOD reduction. Usually,
a higher LOD is replaced by a lower LOD as the player moves away from the asset.
This is done since fewer details are needed for the player to perceive the asset for what it
is and not perceive the asset as poorly rendered, thus maintaining the perceived quality.
However, if the LOD transitions are not seamless and draw the player’s attention, it can
lead to deterioration in the perceived visual quality. Although these LOD reductions and
transitions are usually set automatically, the validation is done manually or not at all since
it is time-consuming, and mundane, and therefore prone to errors.
The purpose of this Master thesis project is to investigate and create a proof-of-concept
for automatic LOD reduction testing and validation using a supervised learning approach
with deep convolutional neural networks. There are two major problems identified: one
is a binary classification problem related to the identification of LOD transitions, and the
other is a multiclass classification problem related to the estimation of the quality of a LOD
transition. The results show that both methods can be used to automate the process of LOD
reduction testing and validation. Furthermore, we can see that the dataset used to train the
model does not have to be identical to the data at hand in order for the model to generalize
to other types of unseen asset types. Additionally, we give recommendations, among other
things, on what models to use, input format such as what concatenation method is the best,
and what data generation and training data should be used. |
Determining confidence in test quality assessment in large-scale software development
Cecilia Malmrud
Avancerad (30hp)
kl 13:15, IDA John von Neumann (In English)
[Abstract]Software testing can be cumbersome and complicated. The complexity of the tests increases
when the software itself becomes larger and more complex. When continuous integration
is applied to software development feedback to tests can be obtained regularly. These
tests are performed in stages and each loop provides test results. A theoretical model for
assigning confidence to different testing stages is presented in this thesis to aid in the un-
derstanding of the test quality. The input to the model was based on information given in
interviews in a case study performed at Ericsson AB. The model is based on the ISO/IEC
25010 standard for software product quality. The theoretical model presented is focused
on the early stages of integration and evaluated qualitatively. Its input is delivery test re-
sults, trouble reports both from customers and internal testers, and continuous integration
flow trends. It was concluded that the theoretical model can be easily automated as each
input source can be automatically collected. For developers working in the early stages of
integration the model could be of help to give insight into what confidence they can assign
their to test’s quality. For testers working later in the flow the suitability of the model re-
quires alterations that cannot be deduced from this thesis alone. For other stakeholders the
usefulness of the model depends on how involved their work is in the development chain. |
Swedish web site
