Ofertes de projectes

La interacció en entorns de Realitat Virtual és complexa perquè els dispositius amb els quals s'ha d'operar són diferents, i a la vegada permeten obtenir informació sobre la posició de l'usuari, la orientació del seu cap, etc.
L'objectiu del projecte és la implementació i avaluació de diferents mètodes de manipulació i interacció que siguin intuïtius i fàcils d'usar en entorns de realitat virtual per models mèdics. També s'estudiarà la viabilitat i usabilitat de l'aplicació per a ser utilitzada en diagnòstic.

A thorough prediction of specific parameters expected during a sailing race is a valuable information for a sailor, as it completely conditions his/her tactics during the race. With the aim of developing a decision support system valid for the Olympic Classes Sailing Venues, the following components are being developed and integrated together into one single web-based platform: 1. Wind component 2. Waves component 3. Oceanic current component 4. Boat Performance component. The present master thesis proposal is related with the wind component. The goal is to develop a methodology/procedure to analyse the 'recorded wind dataset' and recognize significant wind patterns, in other words, characteristic features of the wind speed and direction related to the other weather parameters of the day (air pressure, air and water temperature, cloud coverage, etc..) and to the geographical position of the specific racing area. A similar analysis should be performed on the 'weather prediction model dataset', to find correlations between predicted and measured values. This step is fundamental for the validation of the weather model that will be used daily during the Olympic Games.

Traditional classification schemes for wind patterns are based on meteorological experience and human interpretation of synoptic weather charts. However, it is very useful to have approaches based on clustering that can automatically induce wind patterns based on collected/predicted data, as well as the characteristic features of these patterns and their evolution through the day. Moreover, because data measuring is being performed in different locations of the race areas, these automatic clustering methods will be key to identify different behaviours of the wind depending on the area for the same pattern. All these will allow:

    -          A detailed analysis to determine the representativeness of the wind fields encountered in the measuring period, their frequency of occurrence, timing, rate of evolution, and transition probabilities.

    -          Consequently, a more thorough prediction of the conditions expected before a sailing race, which is, as mentioned, a highly valuable information for the sailor.

    During the previous master projects the clustering module has been completely developed, and it has been applied to data coming from weather prediction models and to data collected on the sea (for Tokyo 2020 Olympic Games). This has been done applying traditional clustering techniques. Also, since these data are recorded daily and present a time-series behaviour, more advanced and novel sequential clustering techniques have been applied to these data in order to extract more complex sequential wind patterns. The current project aims then at:

-      Improving the clustering modules, including the automatic computation of the optimum number of clusters

-      Applying additional machine learning techniques in order to take into account available key parameters, such as geolocation

-      Develop a methodology for comparing/combining the results obtained with both data sets (weather models and actual data)

-      Develop a prediction module that suggests the forecasted pattern and behaviour for a racing day given the weather model data and eventually the actual data early in this day

-      Generalising the environment so that it can be used not only in Marseille (the Olympic venue for Paris2024), but in any particular location where wind patterns are interesting and data are available

    The format of data collected on the sea is the one provided by the Austrian Sailing federation who is the owner of the on-board wind measurement system, and the format of weather prediction models dataset is the one used by operational meteorological centres, and suggested by the World Meteorological Organization, for the storage and the exchange of gridded weather and climate models fields (grib2 or NetCDF).

    Although this is a modality A thesis, interaction/collaboration with TriM company is highly possible depending on the evolution of the work.

Current quantum hardware, which has on the order of tens to hundreds of physical qubits, is insufficient to run these important quantum algorithms. Scaling these devices even to a moderate size with low error rates has proven extremely challenging. Due to these challenges, as well as developing technology for communicating between different quantum chips, we expect quantum hardware to scale via a modular approach similar to how a classical computer can be scaled increasing the number of processors not just the size of the processors.

Goal:

The goal of the thesis is to understand the communications pattern of a set of quantum computers communicating in a modular approach. This will be achieved by a quantum simulator running a set of relevant quantum algorithms and generating a dataset of the required communications. Then analyzing this dataset and proposing early solutions, both from a computer architecture perspective as well as from the communication technologies needed for this. In addition we will also consider ML-based solution for scheduling, therefore the dataset generated becomes a training set for a  Graph Neural Network-based solution.

Today, network administrators lack a tool capable of modeling networks accurately and fast enough to research new protocols, test new configurations, perform what-if analyses, etc. Recent advances in Deep Learning have shown how Graph Neural Networks can be a powerful tool to create these network models known as Network Digital Twins. More specifically, RouteNet has become the reference GNN architecture for network modeling that, given a network topology, a routing scheme, and a traffic matrix, can predict several important performance metrics (delay, jitter, and losses) for administrators.
However, despite these advances, network administrators still lack a usable user interface to interact with RouteNet. The main objective of this project is to build a full-stack web app capable of interacting with RouteNet. The web app should be able to present a network diagram, a traffic load (in a matrix form), and the network metrics (delay, jitter, and losses) returned by RouteNet. In addition, the web should be interactive, allowing the users to draw the network diagram as well as assign properties like link capacity or the scheduling policy to the different elements found in the graph. Finally, the web should be able to interact with the model via a designed API that takes as arguments the topology of the network and the traffic demands, send them to RouteNet, and finally return the output metrics.

More information about RouteNet: https://github.com/BNN-UPC/RouteNet-Erlang

LoRa is a communication technology for IoT applications. It is typically used to send small amounts of data from the sensor node to a gateway (this is the so-called LoRaWAN architecture). But another network topology is also possible, to do mesh networking with LoRa among sensor nodes, e.g. for wildlife preservation:

https://www.hackster.io/mithun-das/mahout-save-the-elephants-819b54

For doing LoRa mesh we have designed a protocol. https://upcommons.upc.edu/bitstream/handle/2117/347444/A_minimalistic_distance_vector_routing_protocol_for_LoRa_mesh_networks.pdf?sequence=1

We have an on-going implementation for the TTGO T-Beam ESP32 LoRa boards (e.g. https://tienda.bricogeek.com/arduino-compatibles/1502-ttgo-t-beam-esp32-wifi-gps-neo-6m-lora-868-mhz.html).  

Technical details: Code is in C++. The implementation leverages RTOS to perform the functions.

https://github.com/LoRaMesher/LoRaMesher

The thesis can work out specific topics of the LoRaMesher design and implementaion according to interest. For instance protocol design, networking, embedded systems or performance evaluation. It could also be interesting to develop a certain application which integrate the LoRa mesh implementation as a library.

The goal of this thesis is to enable the application of Deep Reinforcement Learning techniques for optimization problems in very large and complex scenarios, in this case LEO satellite communication networks. Inspired by the last related work from OpenAI (https://openai.com/blog/evolution-strategies/), we will explore time-efficient alternatives to train a Graph Neural Network based DRL agent on large optimization scenarios in a reasonable amount of time.

PETGEM is an open-source 3D electromagnetic modeler with support for HPC architectures. It solves Maxwell's equations using a high-order vector finite element method. PETGEM is mostly written in Python. On the other hand, PETSc-DMPlex is a set of data management objects and routines provided by the PETSc library that encapsulate the topology of unstructured meshes in an acyclic graph format. It provides a high-level API to create algebraic structures (Mat and Vec objects) from the given mesh topology and manages communication between them in PDE-based applications.

The main objective of this work is the integration of PETSc's data management API for unstructured meshes (DMPlex) with the PETGEM code for electromagnetic modeling in geophysics. This integration will improve the interoperability and portability of PETGEM by allowing the framework to utilise PETSc's interfaces for various common mesh formats and removing legacy dependencies. It will also improve performance due the use of DMPlex's runtime mesh decomposition capabilities, which removes I/O bottlenecks due to reading mesh decompositions from disk.

Since both projects (PETSc-DMPlex and PETGEM) are under ongoing development, this task requires collaboration with both development teams to communicate software engineering requirements as well as bug reports and patches for problems found during development. Then, this work will be done in collaboration with Dr Matt Knepley (PETSc developer at Argonne National Laboratory and University of Chicago) and Dr. Octavio Castillo-Reyes (PETGEM developer and postdoctoral researcher at Barcelona Supercomputing Center)

Graph Neural Networks (GNN) have been recently proposed to learn, model and generalize over graph structured data. Computer Networks are fundamentally graphs, and many of its relevant characteristics -such as topology and routing- are represented as graph-structured data. 

GNN are a central tool to apply ML techniques to Computer Networks. GNN can learn the relationship of complex network characteristics and build relevant models that can be useful to plan and manage a network. In combination with Deep-Reinforcement Learning (DRL) techniques, GNN can help developing autonomous network optimization mechanisms that will result in unprecedented performance, achieving the ultimate vision of self-driving networks.

The Barcelona Neural Networking Center (https://bnn.upc.edu) is a new research initiative of UPC with the main goal of carrying out fundamental research in the field of Graph Neural Networks applied to Computer Networks, and providing education and training to the new generation of Computer Networking students.

 TinyML is a growing community that does machine learning on microcontrollers https://www.tinyml.org. 

Microcontrollers are sometimes the only choice when the power supply is limited, e.g. in for battery- or solar-powered applications. Application examples of TinyML are "wildlife" observation, such as: https://mybirdbuddy.com/ https://opencollar.io/, but tiny machine learning is also in domestic, healthcare and industrial applications.

While Arduino Uno boards are well known for all kinds of hobbyist microcontroller projects, for machine learning the more powerful 32 bit microcontrollers and development boards are used, such as the Arduino Portenta H7 or Arduino Nano 33 BLE Sense. These are able to run machine learning applications. To get the first information on this topic have a look at TensorFlow Lite for Microcontrollers: https://www.tensorflow.org/lite/microcontrollers.

In this project we would like to explore on-device machine learning model training specifically doing it by federated learning. We would like to use the Arduino Portenta H7, the most powerful microcontroller board from the Arduino family. We have Arduino Portenta H7 that can be used for the project. 

We have done previous work and code available where we showed that it is feasible.

https://www.mdpi.com/2079-9292/11/4/573

In the project it would be interesting to explore the capacities of the Portant H7, like using the two cores of the board, use at least one the sensors (camera, accelaration, microphone,...), and use the network connectivity (Wifi, BLE, LoRa)...

One concrete case would be to use for the communication part of federated learning the Wifi connectivity of the Arduino Portenta H7 to build a "cluster" (inspired by clusters of Raspberry Pi) where all the nodes are interconnected and can participate in federated learning. 

The project can start with already working code (in C/C++, Python) that we have from previous work on TinyML with on-device training and federated learning. https://upcommons.upc.edu/bitstream/handle/2117/353756/160036.pdf

Cybersecurity is becoming an increasingly important challenge for all companies and individuals alike. While big names used to be the main targets in the past, as people's lives move online, anyone is nowadays a potential target for any kind of cyber-attack, ranging from phishing to ransomware or serious privacy issues. In order to fight against those ever-evolving threats, Machine Learning is increasingly being used behind the scenes to design better systems that are able of self-learning to boost detection rates and boost overall resilience to unknown attacks. As AI-based solutions penetrate products across the industry, a new kind of threat that is often overlooked is becoming more and more prominent and dangerous: adversarial machine learning (AML).

AML focuses on designing specific inputs to deceive a previously trained Machine Learning models into misclassifying them for a specific purpose. One of the main flaws of any state-of-the-art Machine Learning or Deep Learning algorithms is that they assume that the nature of the data they receive is systematically benign, which is generally the case but does not hold true when an adversarial input is received. The motivation behind altering a ML model into thinking that, for example, a new sample is benign when in fact is malicious can range from pure research to more serious real-life issues such as an autonomous car wrongly classifying a stop sign (and thus provoking a fatal accident) or a wrongly diagnosed disease because of a slightly manipulated magnetic resonance image.

This problem is no exception for Cybersecurity where companies wrongly assume that once the last AI-based product is deployed in their network, their employees are safe...

In the scope of a XPrize challenge, a group of researchers is developing a set of technologies for the identification of animal and vegetation species in tropical forests. The result must be a list of animals and plants that have been identified. The goal of the project is the creation of a multi-view interactive visualization tool that facilitates the inspection of the results. 

The tool will ideally be web based, using technologies such as Vega, JavaScript or similar. 

Internet services are known to collect large amounts of personal information (PI). As a result, more than half of EU citizens are concerned about their online privacy. In this context, data brokers are companies devoted to collect and sell PI to other companies. Data brokers are often implemented as third-party trackers, which allow them to gain visibility across the Internet. Recent works show that collected PI is not only used for targeted advertising, but also for more obscure practices, such as price discrimination, background scanning, phishing or identity theft.

In this project, you will collaborate in the development of EPRIVO, the first European-wide online privacy observatory. EPRIVO will continuously scan the Internet, from multiple locations across Europe, in the search of third-party trackers that do not respect basic privacy rights and current EU regulations (GDPR 2016/679). Quantitative results will be published through an online service developed within the project. Such results will be useful to Internet users, policy makers, website owners and researchers.

The main objective of this project is to apply the most recent advances in Deep Learning, including Deep Reinforcement Learning and Graph Neural Networks, to uncover the particular methods used to track Internet users and collect PI.

This project will be useful for both Internet users and the research community, and will produce open source tools, real data sets, and publications revealing most privacy attempting practices.

Some preliminary results of our work in this area were recently published in Proceedings of the IEEE (IF: 9.237) and featured in a Wall Street Journal article.

More info at:

http://personals.ac.upc.edu/pbarlet/papers/web-tracking.survey2015.pdf

http://blogs.wsj.com/digits/2015/08/04/7-ways-youre-being-tracked-online-and-how-to-stop-it/