Consulta ofertas de otros estudios y especialidades
Source to Source code transformations such as software randomisation at source code level is an effective solution to increase the safety and the security of safety critical systems. Previous prototypes at BSC have been implemented for C and CUDA in older compiler infrastructures. In this thesis, such methods will be ported to the modern compiler infrastructure of Clang and will be tested with large scale applications/industrial use cases.
Source to Source code transformations such as software randomisation at source code level (https://upcommons.upc.edu/handle/2117/96522) is an effective solution to increase the safety and the security of safety critical systems (https://www.youtube.com/watch?v=To_wmz8xIVU). Previous prototypes at BSC have been implemented for C and CUDA in older compiler infrastructures such as CIL (https://gitlab.bsc.es/lkosmidi/tasa_cil). In this thesis, such methods will be ported to the Clang infrastructure of the LLVM compiler project, one of the two most widely used compiler frameworks nowadays. Moreover, these methods will be applied in large scale applications/industrial use cases. The work will be performed in the Barcelona Supercomputing Center (BSC) in collaboration with the European Space Agency (ESA).
The world economy is organized in complex multilayer networks of interactions, in which countries are connected with each other through different interactions, such as trade and financial investment. This networked substrate is the ultimate responsible for the propagation of economic shocks, such financial crisis or economic isolation due to the imposition of sanctions or trade tariffs. In this work, we will analyze the multiplex network reconstructed using yearly data of bilateral trade and financial positions between countries, available at the IMF.
We will try to answer to the following research question: did the global trade-investment network become more resilient in time?
The objectives of the work will be:
- study the evolution of the network across time (compare different years)
- test the resilience of the network to shock propagation, by considering multi-layer percolation processes and different models of shock propagation.
Additional information: https://www.nature.com/articles/s41598-019-49173-2
Rust as a safe language has been increasingly considered for use in space systems. In this thesis, a set of open source benchmarks for on-board space applications which have been developed at the Barcelona Supercomputing Center for the European Space Agency (ESA) will be ported to Rust and their performance and programmability will be evaluated compared to implementations in other not-memory safe languages such as C. The work will be performed in the Barcelona Supercomputing Center.
Rust as a safe language has been increasingly considered for use in space systems. In this thesis, a set of open source benchmarks for on-board space applications (https://obpmark.github.io/) which have been developed at the Barcelona Supercomputing Center (BSC) for the European Space Agency (ESA) will be ported to Rust and their performance and programmability will be evaluated compared to implementations in other not-memory safe languages such as C on embedded computing platforms. The work will be performed in the Barcelona Supercomputing Center.
This Final Degree Project (TFG) aims to improve machine learning pipelines maintainability, evolvability and replication, and ease their transition from experimentation to production. To do so, it proposes the use of object-oriented concepts to define suitable architectural elements needed to design Python machine learning pipelines in Jupiter Notebook. To demonstrate the feasibility of the proposal, an example in the domain of tweet sentiment analysis is presented.
Machine learning is a branch of artificial intelligence which focuses on the use of data and algorithms to produce machine learning models to imitate intelligent human behaviour. A machine learning pipeline is the code that builds a machine learning model. Usually, machine learning pipelines (defined in the experimentation stage) are written in Python using Jupiter Notebook (a server-client application that allows editing and running notebook documents via a web browser). The rapid development of these pipelines in an experimental stage together with the lack of application of software engineering best practices make it difficult to maintain, evolve and replicate the pipeline for the construction of the corresponding machine learning models. Therefore, the transition of machine learning pipelines from experimentation to production is currently anecdotal. This Final Degree Project (TFG) aims to improve machine learning pipelines maintainability, evolvability and replication, and ease their transition from experimentation to production. To do so, it proposes the use of object-oriented concepts to define suitable architectural elements needed to design Python machine learning pipelines in Jupiter Notebook. Such architectural elements will support the design of mantenible, evolvable and replicable Python machine learning pipelines in Jupiter Notebook, easing their transition from experimentation to production. To demonstrate the feasibility of the proposal, an example in the domain of tweet sentiment analysis is presented.
GPUs are increasingly considered for safety critical systems such as autonomous driving. NVIDIA GPUs are more popular, but have a closed infrastructure. In this thesis we will focus on AMD GPUs which support a more open environment. We will use an AMD GPU simulator to understand and evaluate software for safety critical systems.
GPUs are increasingly considered for safety critical systems such as autonomous driving, avionics and space. NVIDIA GPUs are more popular and more studied compared to other GPUs, but have a closed infrastructure. In this thesis we will focus on AMD GPUs which support a more open environment and therefore are a better fit for safety critical systems (https://upcommons.upc.edu/handle/2117/349157, https://upcommons.upc.edu/handle/2117/348962). We will use an AMD GPU simulator in order to model an Embedded AMD GPU and use it to understand and evaluate GPU software for safety critical systems. The work will be performed at the Barcelona Supercomputing Center (BSC) in close collaboration with industrial partners from the aerospace domain.
© Facultat d'Informàtica de Barcelona - Universitat Politècnica de Catalunya - Avíso legal sobre esta web
Esta web utiliza cookies propias para ofrecerle una mejor experiencia y servicio. Si continúa la navegación, entendemos que acepta nuestra política de cookies..