Best Papers 2017

Posted by

The JAACS annually awards a prize for the best Bachelor, Master, and Doctoral thesis at the INF, IIUN and DIUF that was defended during the academic year. The winners are selected by the respective institutes and are presented the award at the institute’s end-of-year event.

 

 

   PhD
Dr. Michel MartiContributions to Intuitionistic Epistemic Logic
INF, BernAbstractMarti analyzed the combination of intuitionistic and epistemic logic. He proved the soundness and completeness of of formal systems for intuitionistic common and distributed knowledge by means of so-called canonical model constructions and introduced new models of intuitionistic justification logics.

 Laudatio

Marti has worked in the general field of non-classical logics with a focus on intuitionistic epistemic logics. He has been able to solve some interesting problems in the context of intuitionistic common knowledge and intuitionistic distributed logic and to make some promising first steps into the new field of intuitionistic justification logics.

http://www.ltg.unibe.ch/staff/mmarti

  PhD
Dr. Andrei LapinApproaches for Cloudification of Complex High Performance Simulation Systems
IIUN, NeuchâtelAbstractScientific computing is often associated with ever-increasing need for computer resources to conduct experiments, simulations and gain outcomes in a reasonable time frame. With continuously increasing need for higher computing power, one of the solutions could be to offload certain resource-intensive applications to a cloud environment with resources available on-demand. We address efficient migration of MPI-based scientific applications to clouds

Laudatio

The thesis tackles the cloudification  of complex high performance simulation systems by addressing various aspects: (a) overview of high performance and cloud computing domains, (b) analysis of  existing simulation problem types, (c) describing and analysing an example Monte Carlo simulator stemming from the hydrogeology domain, (d) presenting two cloudification methodologies, (e) applying the methodologies to the example simulator, and (f) evaluating the potential application of methodologies in a real case study. The thesis concentrates on devising solutions that facilitate the adaptation of scientific applications to ever emerging new technologies and methodologies. It was thus in particular necessary to understand the high performance computing challenges as well as the mathematical grounds of the solutions and the application domain usage.

  PhD
Dr. Leonardo AngeliniA Framework for Abstracting, Designing and Building Tangible Gesture Interactive Systems
DIUF, FribourgAbstractThis thesis discusses tangible gesture interaction, a novel paradigm for interacting with computer that blends concepts from the more popular fields of tangible interaction and gesture interaction. The thesis presents a framework for designing interactions and building interfaces for everyday objects, leveraging our innate skills to manipulate physical objects and communicate through gestures.

Laudatio

Dr Leonardo Angelini has done his PhD in co-supervision between the University of Fribourg and the University of Applied Sciences in Fribourg. This PhD is the result of 10 years of collaboration between Prof. D. Lalanne and Prof. E. Mugelini.

Dr Leonardo Angelini is truly one of the best PhD students that we had between the two institutions. He succeeded to combine the best competencies of Human-IST and Human-Tech institutes. Not only he has developed, and evaluated with users, several systems combining gestural with tangible interaction, he has developed a conceptual framework to support their design, development and evaluation. During his PhD, Dr Angelini has published about 20 scientific articles and organized several international workshops which is remarkable at this stage of the career. We wish him a fruitful and long career in the academic field.

  Master
Simon JenniAn Approach to Unsupervised Visual Representation Learning
INF, BernAbstract

This work introduces a novel self-supervised learning model based on deep neural networks. This model can learn by itself what objects are through learning to distinguish real images from realistic images that it hallucinates from a cartoon input.
The proposed method can be applied to training models with very few examples of input-output mappings and dramatically boost their performance.

Laudatio

This thesis presents very original work which produced a performance comparable and also superior to current state of the art methods by the leading groups at the international level. More precisely, this competes with the best results from Google Deep Mind, from the FAIR group at Facebook, and from the research groups of the University of Berkeley and Carnegie Mellon University. An evolution of this work has been submitted to one of the 3 top tier conferences in computer vision and machine learning.

  Master
Michele Alberti
Using Linear Discriminant Analysis for Deep Neural Network Initialization
DIUF, FribourgAbstract

This work tackles the long training time of Neural Networks with smart weights initialization conveyed through Principal Component Analysis (PCA) and Linear Discriminant Analysis (LDA) algorithms. We show that these two algorithms can be used to achieve faster and better performances compared to State of Art initialization methods.

Laudatio

The Master thesis of Michele Alberti is an outstanding contribution to deep neural networks for image analysis. Michele has extended a given framework with novel initialization algorithms and has made extensive experiments. He has obtained interesting results allowing him to question approaches usually used by deep learning practitioners. Finally, he came up with novel ideas, enabling new learning strategies with less data, while simultaneously providing higher recognition accuracy. The work has given raise to several publications.

  Bachelor
Joel NiklausMachine Learning for Indoor Positioning
INF, BernAbstractThis bachelor thesis explores the potentials of using advanced machine learning algorithms to perform indoor positioning tasks on smartphones using Wi-Fi received signal strength and on-board inertial sensors measurements. Intensive experiment results show that the system achieves highly accurate room-level positioning, which can be further used to enhance the indoor tracking.

 

Laudatio

Recently, machine learning has been popular in computer vision studies. This works applies advanced machine learning techniques in wireless networks and illustrates how it can be used to enhance the performance of indoor positioning.

  Bachelor
Simon StuderSimple Protocol for Encrypted Messaging (SPEM)
DIUF, FribourgAbstractSPEM (Simple Protocol for Encrypted Messaging) is a protocol for end-to-end encrypted instant messaging. It is designed to protect the privacy of its users without compromising usability. SPEM relies on decentralized servers for message delivery, allowing anyone to manage their own messaging servers.

 

Laudatio

In his BSc-thesis, Simon Studer developed a lean messaging protocol with end-to-end encryption, aiming at being an improvement over, for instance, WhatsApp that at the start of the thesis did not use end-to-end encryption. The protocol that Simon designed is very clever and takes security as well as efficiency aspects into account. Simon worked extremely independently, reducing the supervision effort to regular meetings with interesting discussions on new ideas Simon had. Technically as well as presentation-wise, “Simple Protocol for Encrypted Messaging (SPEM)” by Simon Studer is an absolutely excellent and prize-worthy BSc-thesis.

 

Download PDF