lu.se

Datavetenskap

Lunds Tekniska Högskola

Kalendarium

CS MSc Presentation November 1!

Föredrag

Tid: 2018-11-01 10:15 till: 12:00
Plats: E-huset E:2405 (Glasburen)
Kontakt: birger.swahn@cs.lth.se


Three Computer Science MSc theses to be presented on November 1, 2018

November 1 there will be three master thesis presentations in Computer Science at Lund University, Faculty of Engineering.

The presentation will take place in E-huset, room E:2405 Glasburen.

Note to potential opponents: Register as an opponent to the presentation of your choice by sending an email to the examiner for that presentation (firstname [dot] lastname [at] cs [dot] lth [dot] se). Do not forget to specify the presentation you register for! Note that the number of opponents may be limited (often to two), so you might be forced to choose another presentation if you register too late. Registrations are individual, just as the oppositions are! More instructions are found on this page.


09:15-10.00 (E:2405 Glasburen)

Presenter: Erik Jonasson
Title: Migration & Evaluation of Automatic Query Hint Generation Method in Persistent Systems
Examiner: Flavius Gruian
Supervisor: Per Andersson (LTH)

The use of so-called Object-relational-mapping tools is widely used in enterprises. It serves as an abstraction layer between the business logic and the underlying data layer. But it comes at a price, namely with a performance loss in the shape of the N + 1 problem if not configured correctly, a problem that is both time-consuming and error-prone to solve. In this report a method to automate prefetching created by Ibrahim & Cook (2006) has been evaluated for a web application. The obtained results from this study shows that the method can simplify the prefetching phase but in this case with not as optimal results compared to a hand-tuned version. In the tests carried out, the automated prefetching ended up eliminating 9% of the queries of a non tuned version. With the simplicity of use, the tool can be an efficient way to ensure increased performance without spending the time that manual tuning would require. However, the hand-tuned version still performs better in terms of executed queries than the version with automated prefetching. In the state of the current tool, there are still unsolved issues regarding the implementation, meaning that the performance of the tool potentially could be improved with continued development.


10:15-11.00 (E:2405 Glasburen)

Presenter: Fredric Berg
Title: Filtered Path Tracing using Halide
Examiner: Flavius Gruian
Supervisor: Michael Doggett (LTH)

Halide is an embedded language in C++ used for writing high-performance image and array proccessing code. This thesis will explore the possibilites and limitations of Halide by using it for an anisotropic reconstruction of images rendered with path tracing. By taking a Monte Carlo rendered image as an input and using the information from the path tracing process and surrounding pixels a new value can be calculated for the current pixel. The algorithm is first implemented in C++ and later using Halide. The results will be compared through runtime and similarity to the ground truth.


11:15-12.00 (E:2405 Glasburen)

Presenter: Max Söderman
Title: A study of hierarchical attention networks for text classification with an emphasis on biased news
Examiner: Jacek Malec
Supervisor: Marcus Klang (LTH)

The last few years the world has seen an increase in fake and biased news in the media and on the internet. Due to the pure amount of articles, the fake news are basically impossible to manually sort out from the rest. Automatic classification algorithms based on machine learning have therefore been suggested. This study examines different hierarchical attention networks for classifying texts. The main focus has been on biased news but the models have also been tested on other data. The best classifier for fake news was a 3-level hierarchical attention network with FastText embeddings with a f1-score of 0.96, recall of 0.96 and precision of 0.96.