Ofertes de projectes
Consulta ofertes d'altres estudis i especialitats
Boltzmann Machines are probabilistic models developed in 1985 by D.H. Ackley, G.E. Hinton and T.J. Sejnowski. In 2006, Restricted Boltzmann Machines (RBMs) were used in the pre-training step of several successful deep learning models, leading to a new renaissance of neural networks and artificial intelligence. In spite of their nice mathematical formulation, there are a number of issues that are hard to compute. This project aims to address any of these issues.
La lingüística explora les regularitats estadístiques de les llengües i les anomenades lleis lingüístiques. En aquest treball es tracta d'explorar experimentalment la didàctica de la llengua i la tecnologia de forma interdisciplinària a l'ensenyament. Com influeix la llengua a la pràctica educativa? Poden ajudar les tecnologies a la millora de la lectoescriptura i la comprensió lectora, dos elements crucials? Idealment caldrà tenir accés a grups d'estudiants de secundària o cicles: un grup experimental i un grup control sobre els que poder realitzar l'estudi.
To date, traditional Deep Learning (DL) solutions (e.g. Feed-forward Neural Networks, Convolutional Neural Networks) have had a major impact in numerous fields, such as Speak Recognition (e.g., Siri, Alexa), Autonomous driving, Computer Vision,etc. It was just recently, however, that a new DL technique called Graph Neural Network (GNN) was introduced, proving to be unprecedentedly accurate to solve problems that are formalized as graphs.
Navigation meshes are necessary to represent the walkable space of an environment so that agents can perform pathfinding and move through them. Current navigation meshes tend to flatten the environmetn to represent it as 2D polygons connected by edges (square cells, triangles or larger convex polygons). This abstraction presents problems when dealing with complex outdor geometry where the terrain may not be completely flat. With this project, we would develop a novel navigation mesh that can keep the complexity of any 3D input geometry, while still generating small graphs
The idea is to extend an SVM to have more than one layer of processing, thus making it capable of extracting deeper relations among the data observations. All the big benefits of using kernel machines would be then maintained, making it a very useful learning algorithm.
The goal of the project is to create an immersive analytics module to explore the results of training experiences.
This project will use training data (+250k, 5 year span) from diabetic subjects from a primary healthcare database which includes diagnosis and drug codes, and various measurements of clinical and analytical variables. The model will predict of three specific clinical outcomes: i) occurrence of comorbidity, ii) achievement of target glycaemic control (defined as glycated haemoglobin <7 %), and iii) change in glucose-lowering treatment.
The ReliefF family of algorithms have a good reputation for feature selection and weighing. These tasks are very useful and almost essential in any data science problem. The project deals with the kernelization of these algorithms, and thus to make it much more flexible and powerful.
Training neural networks require large quantities of annotated data. Obtaining these data is expensive and labor-intensive. In this project, methods that use a mix of annotated and non-annotated data for training object detectors will be studied. The work will be centered on using transformer architectures that use attention mechanisms.
In mobility applications, tracking and identifying the vehicles is of paramount importance. The goal of this work is to develop algorithms for tracking and extracting information about vehicles (type of vehicle, license plate, doors open/closed, etc) in a city using computer vision.
Sports analytics is a key technology in professional and amateur sports. In this project, the goal is to develop a method to track the players in team games (basketball, football, ...) during the complete game.
This is an offer for a PhD Dissertation, which can be initiated as a MS Thesis. We are working to design a domain-specific accelerator, in the context of high-performance computing, for homomorphically-encrypted deep learning inference. Previous cryptography experience is not required, since there will be an expert in the team advising on those matters. The successful candidate will be integrated into an international team of 9 members working toward this purpose on different topics.
In this project, we wish to create a summarized universal representation of packet flows within a computer network. We approach this problem as an Unsupervised Learning problem, where the summarized representation must be as small as possible while minimizing the reconstruction error. In order to build this representation, multiple approaches are to be considered. This includes using traditional techniques such as a Fourier Transformation, and using representations learned Machine Learning models, specifically Autoencoders and sequence models like 1-dimensional CNNs, RNNs
In this project, we wish to create a summarized universal representation of packet flows within a computer network. We approach this problem as an Unsupervised Learning problem, where the summarized representation must be as small as possible while minimizing the reconstruction error. In order to build this representation, multiple approaches are to be considered. This includes using traditional techniques such as a Fourier Transformation, and using representations learned Machine Learning models, specifically Autoencoders and sequence models like 1-dimensional CNNs, RNNs
The goal of the project is to create an application that facilitates the segmentation of medical models using immersive techniques.
Quantum computers promise exponential improvements over conventional ones due to the extraordinary properties of qubits. However, quantum computing faces many challenges relative to the scaling of the algorithms and of the computers that run them. This thesis delves into these challenges and proposes solutions to create scalable quantum computing systems.
Quantum computers promise exponential improvements over conventional ones due to the extraordinary properties of qubits. However, quantum computing faces many challenges relative to the scaling of the algorithms and of the computers that run them. This thesis delves into these challenges and proposes solutions to create scalable quantum computing systems.
To date, traditional Deep Learning (DL) solutions (e.g. Feed-forward Neural Networks, Convolutional Neural Networks) have had a major impact in numerous fields, such as Speak Recognition (e.g., Siri, Alexa), Autonomous driving, Computer Vision,etc. It was just recently, however, that a new DL technique called Graph Neural Network (GNN) was introduced, proving to be unprecedentedly accurate to solve problems that are formalized as graphs.
Recent advances in the field of Reinforcement Learning (DRL) are rising a lot of attention due to its potential for automatic control and automatization. Breakthroughs from academia and the industry (e.g, Stanford, DeepMind and OpenAI) are demonstrating that DRL is an effective technique to face complex optimization problems with many dimensions and non-linearities. However, to train a DRL agent in large optimization scenarios still remains a challenge due to the computational intensive operations during backpropagation.
This project aims to analyze the prediction capability of Optical Coherence Tomography Angiography (OCTA) images for Diabetes Mellitus (DM) and Diabetic Retinopathy (DR,) in a large high-quality image dataset from previous research projects carried out in the field of Ophthalmology (Fundacio¿ La Marato¿ TV3, Fondo Investigaciones Sanitarias, FIS). OCTA is a newly developed, non-invasive, retinal imaging technique that permits adequate delineation of the perifoveal vascular network. It allows the detection of paramacular areas of capillary non perfusion and/or enlargement of the foveal avascular zone (FAZ), representing an excellent tool for assessment of DR.
The UPC has a set of sensors installed in different buildings that measure CO2 as well as temperature. Although the data can be downloaded, it is a massive dataset that is difficult to analyze. The goal of the project is to build a visual analytics tool that facilitates the exploration of such data to answer complex questions such as whether a peak of CO2 has been produced in any classroom, or which are the most comfortable (i. e., temperature wise) classrooms to teach.
Dimensionality reduction algorithms transform high-dimensional data to lower dimensions, usually 2D or 3D. Analyzing the results of such algorithms typically is carried out using 2D plots and measures. The goal of the project is the creation of an application that facilitates the visual comparison and exploration of those spaces.
This is an offer for a PhD Dissertation, which can be initiated as a MS Thesis. We are working to accelerate homomorphically-encrypted deep learning inference via different efforts, in this case FPGA accelerators. Previous cryptography experience is not required, since there will be an expert in the team advising on those matters. The successful candidate will be integrated into an international team of 9 members working toward this purpose on different topics.
This is an offer for a PhD Dissertation, which can be initiated as a MS Thesis. We are working to accelerate homomorphically-encrypted deep learning inference via different efforts, in this case GPU accelerators. Previous cryptography experience is not required, since there will be an expert in the team advising on those matters. The successful candidate will be integrated into an international team of 9 members working toward this purpose on different topics.
Quantum computers promise exponential improvements over conventional ones due to the extraordinary properties of qubits. However, quantum computing faces many challenges relative to the scaling of the algorithms and of the computers that run them. This thesis delves into these challenges and proposes solutions to create scalable quantum computing systems.
Recent advances in the field of Reinforcement Learning (DRL) are rising a lot of attention due to its potential for automatic control and automatization. Breakthroughs from academia and the industry (e.g, Stanford, DeepMind and OpenAI) are demonstrating that DRL is an effective technique to face complex optimization problems with many dimensions and non-linearities. However, to train a DRL agent in large optimization scenarios still remains a challenge due to the computational intensive operations during backpropagation.
La inteligencia artificial (IA) ha avanzado mucho en los últimos años, en especial con el desarrollo de aplicaciones de lenguaje natural (como GPT-3 y subsiguientes, ChatGPT, Bard...), de resolución de problemas matemáticos (Wolfram Alpha, Photomath...) y de generación de imágenes (Dall-e-2, Stable diffussion, Midjourney...). En este contexto múltiples preguntas son posibles. Algunas: - ¿Se puede o se deben incorporar esta tecnologías a la educación? - En caso afirmativo, ¿cómo y en qué debemos cambiar? - ¿Qué relación hay entre la ética y el uso de la IA en educación?
En aquest treball es proposa dur a terme una anàlisi sobre dades obertes de facturació de fàrmacs pel CatSalut en el període 2016-2023 centrat en la descripció de les dades i la visualització de la informació rellevant, fent èmfasi en la detecció d'usos diferents dels fàrmacs segons la regió sanitària, el sexe o la franja d'edat. Caldrà integrar dades procedents de diferents fonts. També es desenvoluparà un model de predicció de l'ús de fàrmacs anticoagulants i el seu impacte econòmic al nostre territori.
Estudiar las opciones y estándares para incorporar información de "provenance" sobre datos médicos, enfocándose principalmente a estándares (formatos y/o protocolos). Se partirá de la experiencia del grupo en "provenance" para imágenes o para información genómica, aplicándola a datos médicos, como los representados en documentos HL7 o recursos FHIR. Además, se analizará cómo garantizar que la información (metadatos) de Provenance no es modificada y no afecta a la privacidad de nadie. Al final se haría un demostrador.
The main objective of this TFG is to study different network architectures to analyse multimodal data for 3D object detection.
Quantum computing holds immense promise, but optimizing quantum algorithms remains a challenge. In this research project, we'll explore how Deep Learning can improve optimization in quantum systems. By leveraging neural networks, we aim to accelerate practical applications in areas like quantum cryptography and material simulation. You'll join an interdisciplinary team, since the work will be carried in collaboration with the Quantum Information Group at IMDEA Networks in Madrid. They will offer the expertise in Quantum Physics while we work on the Deep Learning part..
LiDAR sensors, crucial in robotics and autonomous driving, emit laser beams to estimate the 3D position of reflected points, stored as a "point cloud". Deep Learning techniques are used in Computer Vision for tasks like image classification, object detection, and semantic segmentation. Recently, these techniques have been applied to point clouds instead of images, expanding applications, especially in autonomous navigation, due to LiDAR's precision and robustness. The goal of this Master's thesis is to investigate Deep Learning techniques for the Semantic Segmentation.
In this project, a survey and analysis of different available methods for post-quantum cryptography will be carried out. Then, a selected set of methods suited for securing next generation communication networks will be implemented and their performance evaluated against classical cryptography methods.
Aquest projecte de final de grau es centra en l'exploració i desenvolupament de mètodes avançats d'intel·ligència artificial aplicats a Brain Computer Interface (BCI), amb l'objectiu de millorar la precisió, eficiència i accessibilitat d'aquestes tecnologies.
The work to do includes: - Analysis of existing specifications for health/genomics consent (FHIR, GA4GH, ...) - Analysis of GA4GH's DUO and other related ontologies for access control - Analysis of how ODRL and other related rights management languages could be used for consents management - Investigate how XACML could be used to manage health/genomics consent information. Existing technologies will be investigated (many of them under development) to provide added value on analysis and proposals for use, standardization and iteroperability approaches.
In this thesis, the candidate will study the acceleration of Graph Neural Networks (GNNs). In particular, given the large design space of dataflows for GNNs and the wide variety of graph datasets that exist, the key question is which dataflow to choose as a function of the characteristics of the dataset. In this work, we will study this from a data-driven perspective, simulating the dataflows for different graph datasets and performing a regression that estimates speed of a given dataflow for unseen datasets, to then choose the best one prior to inference.
This degree thesis aims at designing and implementing a versatile deep learning model for real-time, advanced textile classification using Near-Infrared (NIR) spectroscopy in recycling facilities. It focuses on textile composition prediction (cotton, polyester, viscose...), but it is open to additional models, such as for RGB cameras. This adaptable approach aims not only to enhance the recycling process and material sorting accuracy, but also to ensure sustainability by smoothly incorporating the technology into the daily operations of actual recycling facilities.
In this project, we will study the use of deep learning tools such as HoVer-Net and GNNs for analyzing histhopatological images obtained using different kinds of stains (typically: HER-2, Ki67, RE, RP). This study will be conducted intially on breast cancer samples and possibly extended to other types of tissues.
Quantum computers promise exponential improvements over conventional ones due to the extraordinary properties of qubits. However, quantum computing faces many challenges relative to the scaling of the algorithms and of the computers that run them. This thesis delves into these challenges and proposes solutions to create scalable quantum computing systems.
La lumbàlgia és un dels problemes de salut més freqüents a atenció primària (suposa 25% de les consultes de AP) i que genera més peticions a la hospitalària. L'objectiu del TFG es utilitzar les dades de tipus text dels informes generats durant la visita del pacient, així com les dades de tipus imatge de radiografies, TACS, etc. per tal d'entrenar un model que sigui capaç de predir l'unitat de derivació del cas, superant l'èxit actual de les visites derivades; és a dir, intentant reduïr les visites inoportunes a unitats que no el poden tractar.
The goal of the project is to create an application that is able to read football data and creates a set of views that facilitate the visual analysis of such data.
The goal of the project is the creation of an interactive application that facilitates the exploration of hiking tracks across mountains. The dataset contains multiple tracks that consist on tracks (sets of positions with timestamps) that have been gathered. The objective would be to create a tool that analyzes those tracks in the context of an Open Street Map background.
Development of an AI based intrusion detection system in 5G environments. In this TFG, Deep Learning and Transfer Learning methods will be analysed to improve the accuracy and precision of cyber-attacks detection in 5G networks. Some of the limitation of existing AI based Intrusion Detection Systems are related to the limited computing capabilities of the edge devices and public available 5G datasets, which are scarce and unbalanced.
Computing systems are ubiquitous in our daily life, to the point that progress is intimately tied to the improvements brought by new generations of the processors that lie at the heart of these systems. A common trait of current computing systems is that their internal data communication has become a fundamental bottleneck and traditional interconnects are just not good enough. This thesis aims to study how we can speed up architectures with CPUs, GPUs, and ML accelerators thanks to unconventional (e.g. wireless) interconnects.
Computing systems are ubiquitous in our daily life, to the point that progress is intimately tied to the improvements brought by new generations of the processors that lie at the heart of these systems. A common trait of current computing systems is that their internal data communication has become a fundamental bottleneck and traditional interconnects are just not good enough. This thesis aims to study how we can speed up architectures with CPUs, GPUs, and ML accelerators thanks to unconventional (e.g. wireless) interconnects.
The goal of this project is to explore the application of Learning to Rank algorithms to develop effective trading strategies based on investor sentiment towards a group of stocks. The primary focus is on employing ListNet, a Neural Network algorithm designed for list-wise ranking, to analyse and rank stocks according to their performance influenced by investor sentiment and other news indicators.
This thesis aims to explore the possibilities of the new and less studied variant of neural networks called Graph Neural Networks (GNNs). While convolutional networks are good for computer vision or recurrent networks are good for temporal analysis, GNNs are able to learn and model graph-structured relational data, with huge implications in fields such as quantum chemistry, computer networks, or social networks among others.
This thesis aims to explore the possibilities of the new and less studied variant of neural networks called Graph Neural Networks (GNNs). While convolutional networks are good for computer vision or recurrent networks are good for temporal analysis, GNNs are able to learn and model graph-structured relational data, with huge implications in fields such as quantum chemistry, computer networks, or social networks among others.
The purpose of the project is to create an interactive tool for the visual analysis of gender gaps in different variables across countries or regions.
This thesis aims to explore the possibilities of the new and less studied variant of neural networks called Graph Neural Networks (GNNs). While convolutional networks are good for computer vision or recurrent networks are good for temporal analysis, GNNs are able to learn and model graph-structured relational data, with huge implications in fields such as quantum chemistry, computer networks, or social networks among others.
Recent advancements in nanotechnology have enabled the concept of the "Human Intranet", where devices inside and on our body can sense and communicate, opening the door to multiple exciting applications in the healthcare domain. This thesis aims to delve into the computing, communication, and localization aspects of the "Human Intranet" and how to practically realize them in the next decade.
Recent advancements in nanotechnology have enabled the concept of the "Human Intranet", where devices inside and on our body can sense and communicate, opening the door to multiple exciting applications in the healthcare domain. This thesis aims to delve into the computing, communication, and localization aspects of the "Human Intranet" and how to practically realize them in the next decade.
Recent advancements in nanotechnology have enabled the concept of the "Human Intranet", where devices inside and on our body can sense and communicate, opening the door to multiple exciting applications in the healthcare domain. This thesis aims to delve into the computing, communication, and localization aspects of the "Human Intranet" and how to practically realize them in the next decade.
Recent advancements in nanotechnology have enabled the concept of the "Human Intranet", where devices inside and on our body can sense and communicate, opening the door to multiple exciting applications in the healthcare domain. This thesis aims to delve into the computing, communication, and localization aspects of the "Human Intranet" and how to practically realize them in the next decade.
Quantum computers promise exponential improvements over conventional ones due to the extraordinary properties of qubits. However, quantum computing faces many challenges relative to the scaling of the algorithms (including quantum machine learning algorithms) and of the computers that run them. This thesis delves into these challenges and proposes solutions to create scalable quantum computing systems.
This TFG consists of the data collection and analysis of the inference of ML-based systems in order to define energy labels to classify ML-based systems usage. Such energy labels will be integrated in the GAISSA tooling (https://gaissa.upc.edu/en). Users can be: (i) ML engineers reporting energy consumption of deploying ML-based systems; (ii) citizens when it comes to decide which ML-based to utilise for a particular objective with optimal energy efficiency. Contextual elements (ML task, cost, location, time, and hardware used) will be considered for the energy labels.
ARAP course instructors offer the opportunity to carry out Bachelor's Theses (TFGs) in the field of reinforcement learning. While we welcome initiatives proposed by candidates, we specifically seek individuals to undertake a Bachelor's Thesis within the scope of a European project, focusing on the application of reinforcement learning to cope with stringent bitrate, latency and jitter constraints for time-sensitive Massive Machine-to-Machine communications, in the domain of 5G-advanced, 6G, and Wi-Fi networks.
This TFG aims to significantly improve the capabilities of PETGEM, an open-source 3D electromagnetic modeler primarily written in Python, by integrating new features, and supporting object-oriented programming principles. This project is undertaken in collaboration with the Wave Phenomena Group of the Barcelona Supercomputing Center and the Programming Models Group of the Universitat Politecnica de Catalunya, adding a significant real-world context to the research.
The syntactic structure of a sentence can be represented as a tree where vertices are words and arcs indicate syntactic dependencies between words. Syntactic dependency parsing is the branch of computational linguistic concerned with the extraction of syntactic dependency structures from raw text. This research proposal is focused on unsupervised syntactic dependency parsing, i.e. methods to extract syntactic dependency structures from unlabelled data. This projects consists of implementing simple unsupervised parsers and evaluating them on human languages and other species
Incorporar els principis biomecànics a les aules de tecnologia a nivell de secundària
The syntactic structure of a sentence can be represented as a tree where vertices are words and arcs indicate syntactic dependencies between words. Syntactic dependency parsing is the branch of computational linguistic concerned with the extraction of syntactic dependency structures from raw text. This research proposal is focused on unsupervised syntactic dependency parsing, i.e. methods to extract syntactic dependency structures from unlabelled data. This projects consists of implementing simple unsupervised parsers and evaluating them on human languages and other species
Por un lado, se trata de argumentar la importancia de una visión holística de la sostenibilidad en la educación, es decir, aquella que contemple la totalidad del sistema o del problema, y no se centre en resolver pequeñas partes integrantes del problema, perdiendo la visión de conjunto necesaria para alcanzar una comprensión más profunda del problema. En el aula, se trata de fomentar dos competencias transversales: 1) La sostenibilidad y compromiso social. 2) El pensamiento crítico. Para ello se pueden usar diferentes herramientas y estrategias, como el debate en clase.
The GESSI research group has a web application called SOPCOM that allows the definition and composition of software engineering processes. This project will consist on adding some functionalities to the application and testing the application for improving its reliability and usability. This project will start from the current version of the SOPCOM tool developed in a TFG. Technologies: Angular, TypeScript, NodeJS, Angular Material, PHP, MySQL, Docker, Postman
This master thesis project aims to elevate the capabilities of PETGEM, a cutting-edge open-source 3D electromagnetic modeler written primarily in Python. Conducted in collaboration with the Wave Phenomena Group of the Barcelona Supercomputing Center, the Programming Models Group of the Universitat Politecnica de Catalunya, and the Argonne National Laboratory and University of Chicago, this research adds a substantial real-world context to the development.
La realitat augmentada ofereix moltes possibilitats pedagògiques. La proposta de TFM consisteix a buscar una eina d'AR útil per la docència i plantejar-ne l'aplicació en una unitat didàctica/unitat formativa.
The goal of the project is the creation of an application that uses LLMs or NLP technologies to generate visualizations from data stored in databases. The objective would be to add a layer of natural language for data querying.
Application of machine learning techniques to improve weather forecasting, optimize data processing and implement various ML algorithms into existing monitoring systems for a vast number of applications in the field of atmospheric sciences
This project will be done in collaboration with Telefonica Research. Telefonica Research is a diverse, multidisciplinary and international group of scientists who dare to push the frontiers of knowledge and prepare for the upcoming challenges on communications and the Internet.
This project will be done in collaboration with Telefonica Research. Telefonica Research is a diverse, multidisciplinary and international group of scientists who dare to push the frontiers of knowledge and prepare for the upcoming challenges on communications and the Internet.
S'ofereix una beca d'iniciació a la recerca de 20 hores/setmana amb un salari aprox. de 600 Euros/mes per realitzar el TFM en el marc del projecte GRAPHSEC.
Es tracta de dissenyar una app basada en aprenentatge profund i transformadors (transformers) que sigui capaç d'escoltar com un estudiant que estudia una llengua pronuncia un text, notar on la pronúncia es desvia més de l'adequada i pronunciar-li bé per què l'estudiant ho senti.
Es tracta de crear un software analitzador de paraules clau (keywords) en els títols dels articles científics dels darrers anys per tal de trobar tendències, associacions i de fer prediccions sobre el futur, sempre en el camp de Machine Learning/Data Science.
Quantum computing holds immense promise, but optimizing quantum algorithms remains a challenge. In this research project, we'll explore how Deep Learning can improve optimization in quantum systems. By leveraging neural networks, we aim to accelerate practical applications in areas like quantum cryptography and material simulation. You'll join an interdisciplinary team, since the work will be carried in collaboration with the Quantum Information Group at IMDEA Networks in Madrid. They will offer the expertise in Quantum Physics while we work on the Deep Learning part..
Topological Deep Learning TLD is a new neural network architecture intended to extract knowledge from complex data structures. In particular TLD operates with hypergraphs, simplicial complexes and cell complexes. Right now, TLD is a new trend in the field of deep learning, with relevant applications in information compression, biology and chemistry. The goal of this project is to extend the IGNNITION framework (1) to support relevant use-cases in TLD. IGNNITION is a popular framework with thousands of users per month: https://ignnition.org/
The goal of the project is to create an immersive analytics module to explore the results of training experiences.
Topological Deep Learning TLD is a new neural network architecture intended to extract knowledge from complex data structures. In particular TLD operates with hypergraphs, simplicial complexes and cell complexes. Right now, TLD is a new trend in the field of deep learning, with relevant applications in information compression, biology and chemistry. The goal of this project is to extend the IGNNITION framework (1) to support relevant use-cases in TLD. IGNNITION is a popular framework with thousands of users per month: https://ignnition.org/
S'ofereix una beca d'iniciació a la recerca de 20 hores/setmana amb un salari aprox. de 600 Euros/mes per realitzar el TFM en el marc del projecte Eprivo.eu.
S'ofereix una beca d'iniciació a la recerca de 20 hores/setmana amb un salari aprox. de 600 Euros/mes per realitzar el TFM en el marc del projecte Eprivo.eu.
Improving the public transportation systems is paramount to promote a sustainable urban mobility. Gathering information about factors like the time that people are waiting at the bus stop, or the number of persons that enter or exit from a vehicle, can help urban planners in the design of the public transportation network. This project aims at using computer vision to obtain data such as waiting times, etc. The methodology will consist of installing cameras at a few bus stops and then use person detection and tracking to get the data.
In mobility applications, tracking and identifying the vehicles is of paramount importance. The goal of this work is to develop algorithms for tracking and extracting information about vehicles (type of vehicle, license plate, doors open/closed, etc) in a city using computer vision.
DR algorithms are a very popular option for gathering insights on complex datasets. They are extensively used in many areas including biology, data visualization, and so on. The purpose of this project is to evaluate the adequacy of different DR algorithms to document representations for the exploratory analysis of large corpus.
L'ús de noves tecnologies és cada cop més freqüent en els projectes de patrimoni cultural. Una d'aquestes tecnologies permet aconseguir informació multiespectral de pintures per tal de poder analitzar informació sobre pigments que poden donar resposta a dubtes actuals per part dels historiadors de l'art. En aquest TFG es proposa crear una eina de visualització de dades que permeti a historiadors de l'art poder visualitzar i analitzar la informació multiespectral per respondre dubtes actuals.
DR algorithms produce outputs that are intended to represent high-dimensional relationships in a lower dimensional space. However, getting to understand how different are the distributions of high- and low-dimensional sets is difficult. The goal of this project is to produce a visual analytics system that provides visual tools to analyze how this process changes throughout the execution of the algorithm.
This thesis aims to explore the possibilities of the new and less studied variant of neural networks called Graph Neural Networks (GNNs). While convolutional networks are good for computer vision or recurrent networks are good for temporal analysis, GNNs are able to learn and model graph-structured relational data, with huge implications in fields such as quantum chemistry, computer networks, or social networks among others.
Recent advancements in nanotechnology have enabled the concept of the "Human Intranet", where devices inside and on our body can sense and communicate, opening the door to multiple exciting applications in the healthcare domain. This thesis aims to delve into the computing, communication, and localization aspects of the "Human Intranet" and how to practically realize them in the next decade.
The goal of the project is to create a foundation model for precision health. To do so, data from the UK Biobank will be used, which is the largest dataset up to date containing genetic, lifestyle and health data from half a million UK participants. The first step is to encode both genotypic and phenotypic data in the same joint embedding space. The following step consists of using the generated embeddings to create a model that predicts phenotypes from genotypic data.
This is an offer for a TFG. We are working to accelerate homomorphically-encrypted deep learning inference via different efforts, in this case GPU accelerators. Previous cryptography experience is not required, since there will be an expert in the team advising on those matters. The successful candidate will be integrated into an international team working toward this purpose on different topics.
The detection and measurement of fruit size is of great interest to estimate the crop and predict harvest resources. Nowadays, different sensors, such as LiDAR, are able to register fruit trees into a 3D map of the environment. The main goal of this thesis will be to explore and design new deep learning architectures based on Graph or 3D Neural Networks to detect fruits.
This project aims to assess the capability of machine learning to establish a relationship between genomic sequences and corresponding images (e.g., MRI and DXA scans of different parts of the body, photographs) from individuals. The goal would be to develop a generative AI model capable of producing images based on an individual's genetic sequence
Graph drawing is a field that encompasses two disciplines: graph theory and visualization. Traditionally, graph drawing has focused on producing designs that are understandable and aesthetically pleasing. In this project, we want to consider another aspect that aims to produce layouts with a uniform distribution of nodes, at the expense of possibly sacrificing some aesthetic properties of the layout.
The focus of this project is to address the problem of Temporal Action Segmentation, which consists in temporally segmenting and classifying fine-grained actions in untrimmed videos. Successfully addressing this problem can yield substantial advancements in video understanding and benefit various applications domains, including robotics, surveillance and many more.
Digital pathology involves the study and diagnosis of diseases through high-resolution digital images (Whole-Slide Images or WSI) obtained from biopsy captured with a scanning device. In the DigiPatics project, we focus on immunohistochemistry imaging, which consists on staining the cellular tissue with antibodies to detect the presence of specific biomarkers for lung cancer diagnosis. The final objective is to establish the relation between positive and negative nuclei in the tumoral areas and help doctors in their diagnosis work
The goal of the project is to create an interactive application that enables the exploration of a Super Mario dataset that contains gameplays.
The goal of the project is to create an interactive exploratory analysis tool for music data. The exploration may include sound, lyrics, or both.
S'ofereix una beca d'iniciació a la recerca de 20 hores/setmana amb un salari aprox. de 600 Euros/mes per realitzar el TFM en el marc del projecte Eprivo.eu.
La lingüística quantitativa explora les regularitats estadístiques de les llengües i les anomenades lleis lingüístiques. Com afecten la llei de Zipf, la llei de brevetat o l'efecte de freqüència a l'aprenentatge? En aquest treball es tracta d'explorar experimentalment la influència (o no) de les lleis lingüístiques a l'ensenyament-aprenentatge. Com influeix la llengua a la pràctica educativa? Es tracta de fer una anàlisi computacional i lingüística sobre la producció oral en els processos d'ensenyament i aprenentatge.
This degree thesis aims at comparing and selecting algorithms for tumorous area detection and segmentation in Whole Slide Images (WSI) of breast cancer biopsies for the DigitPatICS project. A new H&E dataset is being annotated at ICS for this task. Some additional objectives are proposed such as pre- and post-processing steps (relevant tissue detection, tiling size, resolution and overlap, label post-processing for WSI reconstruction, visualization modes), and the role of H&E and nuclear stains registration for the quantification of the latter in WSIs of breast cancer.
S'ofereix una beca d'iniciació a la recerca de 20 hores/setmana amb un salari aprox. de 600 Euros/mes per realitzar el TFM en el marc del projecte GRAPHSEC.
Notebook platforms like Jupyter and Google Colab facilitate Python-based data science prototyping. However, most data sourcing and preparation, which takes up 80% of a data scientist's time, is done manually. The DTIM group's ODIN system seeks to automate this process. This project aims to integrate ODIN's capabilities into notebook platforms for faster data discovery and preparation.
The DMAG (Distributed Multimedia Applications Group) of the IMP (Information Modeling and Processing) research group of the UPC has developed Reference Software (validated as an ISO International Standard) for JPEG Systems, a set of standard applications to provide extra facilities to JPEG images. This work intends to develop software applications on top of that Reference Software to demonstrate its use. Examples include protection of regions of images using JPEG Systems Privacy and Security standard or use of JPEG Snack standard with audio, images, etc.
UPC and Nestlé are offering a new position to develop the TFG in the field of Machine Learning and Cybersecurity. This TFG will be fully funded (internship) and carried out in collaboration with the Global Security Operations Center of Nestlé and UPC.
UPC and Nestlé are offering a new position to develop the TFM in the field of Machine Learning and Cybersecurity. This TFM will be fully funded (internship) and carried out in collaboration with the Global Security Operations Center of Nestlé and UPC.
Some esports games have APIs that allow the extraction of detailed data on matches. However, these datasets have mostly been analyzed with a statistics perspective. The goal of this project is to create a series of visual analytics tools to gather insights on how people perform in League of Legends.
The proposed challenge is to implement an estimation tool for integrated circuit design, which aims to quickly assess key metrics like area, power, timing, and congestion without the need for a full physical design.
Understanding the evolution of the cyclists along a single stage of races such as Tour de France requires knowing their relative position throughout many hours. The goal is to take a private dataset that includes samples of all riders every 10 seconds, to create a visual analytics tool that can describe the evolution of the stages of the race.
UPC and Nestlé are offering a new position to develop the TFM in the field of Machine Learning and Cybersecurity. This TFM will be fully funded (internship) and carried out in collaboration with the Global Security Operations Center of Nestlé and UPC.
S'ofereix una beca d'iniciació a la recerca de 20 hores/setmana amb un salari aprox. de 600 Euros/mes per realitzar el TFM en el marc del projecte GRAPHSEC.
Quantum computing holds immense promise, but optimizing quantum algorithms remains a challenge. In this research project, we¿ll explore how Deep Learning can improve optimization in quantum systems. By leveraging neural networks, we aim to accelerate practical applications in areas like quantum cryptography and material simulation. You¿ll join an interdisciplinary team, since the work will be carried in collaboration with the Quantum Information Group at IMDEA Networks in Madrid. They will offer the expertise in Quantum Physics while we work on the Deep Learning part..
Topological Deep Learning TLD is a new neural network architecture intended to extract knowledge from complex data structures. In particular TLD operates with hypergraphs, simplicial complexes and cell complexes. Right now, TLD is a new trend in the field of deep learning, with relevant applications in information compression, biology and chemistry. The goal of this project is to extend the IGNNITION framework (1) to support relevant use-cases in TLD. IGNNITION is a popular framework with thousands of users per month: https://ignnition.org/
The syntactic structure of a sentence can be represented as a tree where vertices are words and arcs indicate syntactic dependencies between words. Syntactic dependency parsing is the branch of computational linguistic concerned with the extraction of syntactic dependency structures from raw text. This research proposal is focused on unsupervised syntactic dependency parsing, i.e. methods to extract syntactic dependency structures from unlabelled data. This projects consists of implementing simple unsupervised parsers and evaluating them on human languages and other species
The main objective of this work is to investigate and implement efficient strategies for the parallel construction of sparse matrices derived from FE analysis within the PETGEM workflow. This development aims to provide fundamental insights into the interoperability and portability of PETGEM by enabling the framework to utilize exascale HPC architectures.
The goal of the project consists in the design of off-line RL algorithms that leverage knowledge about the symmetry of the environment and system, and to obtain better policies with fewer data.
El projecte consisteix en usar tècniques de visió per computador per a l'aprenentatge i reconeixement d'un dataset de senyals de trànsit.
Adaptation of state-of-the-art AI models for native integration into desktop applications. The performance and suitability of different frameworks (Pytorch C++, GGML, TensorRT, ONNX Runtime, etc.) will be evaluated with respect to different platforms (Windows and macOS) and hardware resources (CPU, NVIDIA GPUs, Apple Silicon).
This research proposal addresses the critical need for efficient and accurate tumor segmentation in neuroimaging, particularly utilizing Magnetic Resonance Imaging (MRI) data. The project aims to develop a robust and automated approach for tumor delineation. The proposed methodology integrates state-of-the-art MRI neuroimaging techniques with advanced artificial learning methods.
La tecnoética estudia los aspectos morales y éticos de la tecnología en la sociedad, en un sentido amplio: nos incumben tanto el diseño como la creación y el uso de la tecnología. En este TFM se propone que el estudiante de máster plantee la tecnoética mediante la creación de materiales educativos para la enseñanza secundaria o la formación profesional, utilizando metodologías activas y canales narrativos como el cine, la literatura o las series.
Development of a desktop application in C++ that enables users to leverage the latest advances in Computer Vision and Generative AI to facilitate the creation and editing of 2D animations.
This work is included in the research project LABINQUIRY, in a teaching environment. The goal is to develop a system capable of collecting and organise documents (pairs question-answer) expessed in natural language. These documents are generated within the interaction between the teacher and the students.
La proposta consisteix a analitzar algunes de les eines d'AI com a possibles eines per a utilitzar en l'aula de Tecnologia o en alguna de les famílies de la formació professional. Analitzar especialment la seva utilitat pedagògica.
The goal of the project is to create an application that facilitates the segmentation of medical models using immersive techniques.
This thesis aims to explore the possibilities of the new and less studied variant of neural networks called Graph Neural Networks (GNNs). While convolutional networks are good for computer vision or recurrent networks are good for temporal analysis, GNNs are able to learn and model graph-structured relational data, with huge implications in fields such as quantum chemistry, computer networks, or social networks among others.
On som
C/Jordi Girona Salgado,1-3
08034 BARCELONA Espanya
Tel: (+34) 93 401 70 00
© Facultat d'Informàtica de Barcelona - Universitat Politècnica de Catalunya - Avís legal sobre aquest web - Configuració de privadesa