Recent Submissions 

  1. Scheduling for MIG-Capable GPUs 

    Oberholzer, Pascal (2021)
    The development of heterogeneous CPU-GPU systems for modern data centers in recent years increased the computational power massively. Multi-Instance GPUs (MIGs) allow to split a single physical GPU into multiple independent instances. Thus, more than one user can get exclusive acces to the same physical GPU. However, there are many different configurations for a single MIG-capable GPU that need to be controlled by a central instance. ...
    Master Thesis
  2. Input Convex Neural Networks for Energy Optimization in an occupied Apartment 

    Schalbetter, Adrian (2020)
    The building sector in Switzerland accounts for approximately 50% of the final energy consumption, where the largest share is being consumed by space heating. Consequently, there is a lot of potential to save energy and reduce carbon dioxide emissions. Apart from expensive building renovations, efficient control strategies can lead to a reduction of energy consumption. Traditional control strategies are often based on On/Off controllers. ...
    Master Thesis
  3. On Factors Impacting the choice of Solar Desalination 

    Vetter, Marc Maurice (2015)
    Increasing global demand has already outstripped the natural supply of fresh water in many areas of the world and is driving the development of desalination technologies. Among renewable energy sources for Desalinatinon, Solar Desali-nation is the predominant one and formed for this reason the focus of this the-sis. The cyclical nature of insolation however confronts the system engineer with dilemma to choose between operating the ...
    Master Thesis
  4. Share to learn? The role of actor attitudes and information exchange for policy-oriented learning concerning carbon sinks in forests vs. timber 

    Burgdorf, Silvia (2021)
    Climate change and its effects require effective solutions to mitigate and avoid greenhouse gas emissions. The Kyoto Protocol and the Paris Agreement translate global to Swiss mitigation targets, which bring in forests and timber as potential carbon sinks. However, it remains unclear how the sinks are best combined in a local context, especially when forest stands simultaneously fulfill multiple functions. As a result, there is a potential ...
    Master Thesis
  5. Experimental study on the seismic and aseismic deformation during the failure of granitic rock 

    Niu, Zihua (2021)
    During the exploration of geothermal energy, fluids are often injected into the high temperature host rocks in the upper lithosphere for efficient heat or electricity production. It is known that large-scale injection activity will perturb the subsurface stress field, resulting in nucleation and propagation of fractures or reactivation of the existing faults. These mechanisms damage the rock and are accompanied by both static (aseismic) ...
    Master Thesis
  6. Coupling georadar field data with numerical modeling to increase the effectiveness in the exploration of crystallized mineral cavities 

    Nikolaidis, Prodromos (2021)
    Demand for mineral specimens which are naturally occurring in crystallized mineral cavities in rocks (or Mineral Specimen Cavities – MSCs) is currently on the rise. Predictions indicate that this trend will most probably be maintained at least for the forthcoming years. In order for the supply to satisfy the demand, the prospecting process, i.e., the activity of discovering and extracting these mineral specimens, needs to be accordingly ...
    Master Thesis
  7. Axial-Flux Machine Simulation and Optimisation for Wheel-Hub Applications 

    Messin-Roizard, Théophile (2021)
    Master Thesis
  8. Memory Deduplication on Serverless Systems 

    Qiu, Wei (2021)
    Master Thesis
  9. Building a Neural Network Explainer 

    Stalder, Steven (2021)
    Master Thesis
  10. The Bayesian Transformer 

    Cinquin, Tristan (2021)
    Master Thesis
  11. Development of graph-based management tool based on Graph Neural Networks 

    Graf, David (2021)
    Knowledge Graphs, a form of connected data, created a new research field to apply machine learning algorithms called Graph Neural Networks (GNN). We study several GNN models by integrating them into a newly proposed framework called GrafAE, introducing multiple novel GNN models. Furthermore, in this work, we research the effect of embedding dimensionality, a vector representation of the knowledge graphs’ features, and the impact of ...
    Master Thesis
  12. Consent Routing: Towards Bilaterally Trusted Communication Paths 

    Blarer, Mathias (2021)
    Master Thesis
  13. Event-triggered multi-hop communication for wireless sensor networks 

    Gatschet, Tobias (2020)
    Wireless Sensor Networks (WSNs) play an important role in today’s society in Internet of Things, automation, environmental monitoring etc. However, as these networks should be able to operate autonomously and for a long period of time without recharging, careful energy management is required. A central aspect of such networks is the energy budget for communication which has received much attention in recent years. To ensure network-wide ...
    Master Thesis
  14. Packing problems in groups, including mutually unbiased bases: symmetries and dual witnesses 

    Doppelbauer, Nikolaus (2021)
    We investigate the problem of upper bounding the size of independent sets in both finite and infinite graphs. We focus on bounds achieved by convex relaxations for highly symmetric graphs defined on groups. In particular, we study dual bounds which lower bound the achievable upper bounds. The first chapter is concerned with finite graphs and a semidefinite programming relaxation of maximum independent set. The group structure makes ...
    Master Thesis
  15. Leveraging Application Layer Path-Awareness with SCION 

    Gessner, Jonas (2021)
    Path-awareness in the next-generation Internet architecture SCION can be used to increase the reliability and performance of networking operations. It enables the unique opportunity to select inter-AS paths according to properties like latency or bandwidth, which is not possible with the current Internet Protocol. This thesis focuses on how path-awareness in SCION can be leveraged in consumer applications, in particular, video calls. ...
    Master Thesis
  16. Exploiting Structured Sparsity in Transformers through Data Centric Optimizations 

    de Fremond, Siméone (2021)
    As deep learning is becoming increasingly costly, both in terms of computational power and storage power, many efforts have been invested in finding ways to reduce that cost. Especially in the transformer architecture, where the attention matrix grows quadratically with the sequence length, and it is in fact a lower triangular matrix. The value of implementing a fast triangular matrix matrix operator is thus highlighted. Here we ...
    Master Thesis
  17. Automated Mining of User Account Access Graphs 

    Bartholomä, Marc-Philippe (2021)
    Most users have a multitude of different accounts making it hard to get an overview for security analysis. In particular, as there are frequently multiple ways to obtain access to an account, be it via recovery methods or through storage in a password manager. To model and analyze such relationships, prior work introduced a formalism called account access graphs, but to work with it, it is necessary to create the graph manually from ...
    Master Thesis

View more