Investigation the potential and limitations of Shapley values for explaining machine learning models
MmWave offers ultra high wireless capacities, but these capacities fluctuate wildly. This project aims to develop a proxy for reliable consistent mmWave communication using multiple concurrent mmWave paths.
You will implement and test a web-based design interface, where exercise managers and trainers can plan crisis response exercises. This front-end will interact with an AI-planning module (developed by us) which aids planners when using the front-end. The planned simulation will run in a simulation platform such as Unity.
Memory training exercises are known to have a positive effect on improving the memorability of humans. A memory training task can be as simple as trying to remember a password of increasing complexity or a number of varying length.
his topic focuses on developing automated testing methods for cancer registration and support system at Cancer Registry of Norway.
Developing automated testing tools for telehealth services at Oslo City.
Use machine learning and video processing techniques to automatically find events in sports videos. For example, in football, some "easy" ones are goals, but others like tackles are harder.
A large number of sport videos exists, and generating automatich highlights or summaries is a detious and time consuming task. In this thesis, we aim to automatically generate highlight summaries from soccer games using tools like machine learning, computer vision and video analysis.
Use machine learning and video processing techniques to cut video events in sports videos. E.g., find the best start and stop times for a goal in a soccer video.
Investigate, develop and evaluate data-driven techniques and prototypes that help software engineers build software systems that are autonomously self-healing. These are systems that can understand when they are not operating correctly and, without human intervention, make the necessary adjustments to restore themselves to normal operation.
The goal for this project is to build and maintain a platform which allows for the automatic merging, representation, and analysis of multiple crowdsourced network measurement datasets from different sources. The applications include tracing the evolution of network performance during the COVID-19 pandemic.
To help IT professionals optimize the value for stakeholders, you will develop a set of easy-to-use tool for estimating and monitoring how much potential benefit a system under development will produce. You will design the tool based on recent theoretical results, and you will test and evaluate the tool with actual software professionals.
This project focuses on building digital twin for one or more parts of the seafood supply chain.
This project focuses on building digital twins for Smart Buildings (e.g., Smart Hospitals) and Smart Power Generators (e.g., Wind Turbines) for advanced analyses with AI techniques.
Survey the literature, techniques, tools, and tactics employed by competitors in DARPA's Cyber Grand Challenge, and use selected techniques, tools, and tactics to develop and evaluate a prototype autonomous secure and resilient system.
Develop and evaluate data-driven techniques and prototypes for automatically assessing the security of a software system by analyzing the system's source code for potential security vulnerabilities during the development stage.
Experiencing delay while playing a video game can be very annoying for users. However, games have different sensitivity to delay, in some games like a First Person Shooter game, the existence of delay is very annoying, and in a less sensitive game like a turn-based game it is less annoying. In this thesis, we aim to classify the games based on their delay sensitivity.
Over the recent years Generative Adversarial Models (GANs) have revolutionized the field of artificial intelligence (AI). Given a set of training data, the models can generate outcomes with striking similarities in properties to the training data. E.g. if the training data is a set of images, the models are able to generate new images with the same properties and look stunningly realistic. However the theoretical foundation of GANs is still immature and further understanding is needed. For example, GANs have been criticised to possibly only replicate a limited set of the properties in the training data.
Gastrointestinal (GI) cancers are the leading cause of cancer worldwide. Early detection of abnormalities in the GI tract can reduce the chances of GI cancers and provide successful treatment. Towards fulfilling this goal, we target to build novel generalizable and robust machine learning algorithms that can potentially find the miss-detected lesion that were left out during manual endoscopy and improve the healthcare system by automatically detecting, segmenting, and localizing diseases and other endoscopic findings inside the GI tract.
The detection of abnormalities in the gastrointestinal tract can help to reduce the chances of colorectal cancer and provide successful treatment. Towards fulfilling this goal, we target to build a generalizable and robust machine learning that can improve the healthcare system by automatically segmenting and detecting diseases and instruments inside the GI tract.
This topic is about developing digital twins for various types of cyber-physical systems.
Deep learning models used for image recognition and classification are often so complex that they are effectively black boxes. Investigate and develop methods for explaining these using concepts from game theory
Explain the prediction of a deep neural network analyzing microscopic videos of human semen.
This topic explores the potential of using synthetic data to train machine learning algorithms in fields that don’t have a lot of open annotated datasets, such as medicine.
NCCL (pronounced "Nickel") is a stand-alone library of standard collective communication routines for GPUs. It has been optimized to achieve high bandwidth on platforms using PCIe, NVLink, NVswitch, as well as networking using TCP/IP sockets. NCCL supports an arbitrary number of GPUs installed in a single node or across multiple nodes and can be used in either single- or multi-process (e.g., MPI) applications.
Over the last decade it has been shown that waste is cleared from the brain along different fluid pathways. In this project we will use one or more fluid mechanics models (diffusion, convection-diffusion, Navier-Stokes equations, Biot's equations) to answer which mechanism is most likely to explain waste clearance from the brain as seen in experimental data.
Applying Deep Neural Networks for the generation of synthetic health data that protects privacy and promotes healthcare research
Survey experiments in studies of political and electoral behavior using profiles of potential candidates are standard practice in political science. With this type of experiments, the researchers' goal is to test which characteristics voters value most in political candidates. One of the main challenges, however, is to generate realistic profiles to be tested. One of the important components of these profiles are the faces of candidates. A potential option is to use faces of real candidates, but this involves complex legal and ethical issues. Another option is to hire models to represent candidates, but this can be expensive and time-consuming.
Data-driven medical systems for disease prevention rely on multimodal data streams for decision making. The representations of each data stream can be ordered in graphs. In this topic, students will explore graph convolutional neural networks for deciding steps within a medical system to prevent life-style diseases developing in its users.
Construction, evaluation and reasoning using knowledge graphs for software vulnerability assessments.
In this project you will investigate and optimize a variety of machine learning techniques to interact with advanced biophysical mathematical models describing cellular processes in cardiac cells.
Ranging from surveillance, network security, stream mining, and financial analysis streaming applications are emerging and requiring real-time processing and predictive modelling. These applications have challenging properties and ambitious learning demands; the process has to be completely unsupervised, automated, without human intervention for observing or labelling. Lately, Machine learning raised as a powerful modelling and learning tool and it manifested impressive results on several benchmark problems. Is the current state-of-the-art of machine learning capable to handle the challenge of learning and modelling in real-time?
Compare, benchmark, and optimize machine learning frameworks.
Developing methods based on AI techniques design, develop, and test various Drone applications.
This thesis focuses on the development and testing of techniques for localising Narrowband Internet of Things (NB-IoT) devices, by leveraging machine and deep learning techniques on empirical measurements
Message Passing Interface (MPI) is a standardized and portable message-passing system designed by researchers from academia and industry to function on a wide variety of parallel computers.
How can we distribute complex problems on supercomputers that contain many CPU cores and GPUs ? This thesis looks at improving the software that decomposes problems for such parallel heterogeneous systems.
Ordinary differential equation (ODE)-based “ionic” or “membrane” modeling of excitable, living cells is an established formalism in computational physiology permitting insight into their function. In this project, a recent model of the electrophysiology of a non-excitable cell, the chondrocyte (which maintains and sustains cartilage) – will be supported and explored.
Generating segmentation mask of polyps of Gastrointestinal Tract (GI) images collected from endoscopic videos is an important task for analyzing GI tract videos. This segmentation task can be achieved by the state of the art GAN architectures by generating synthetic polyps conditioned on segmentation masks.
The goal is to apply existing quantum search algorithms or develop new quantum search and optimization algorithms to solve classical optimization problems.
Forecasts of precipitation the next minutes and hours ahead are of great value for a wide range of users including the public. At the Norwegian Meteorological Institute such forecasts have been generated for several years using optical flow methods and made available in the yr app and at yr.no. In recent years approaches based on deep learning have also demonstrated promising results. For example, the use of convolutional LSTMs has been proposed by Shi et. al. (2015), while Sønderby et. al. (2020) apply networks with attention-based layers.
Automatic classification of sport news into different categories using state of the art Natural Language Processing (NLP). The project is building upon an existing dataset of Norwegian soccer news.
This topic explores testing reinforcement learning agents, their training, and their environments through automatic generation of test scenarios.
Time series analysis using Residual neural networks. The main task is to explore how different ResNets can be combined to solve complex time series problems.
Developing methods based on AI techniques to discover unforeseen situations in Elevators or Train.
This research is focused on generating sperm segmentations to predict motility and morphology levels of sperms videos in the Visem open dataset. To accomplish this, the CycleGAN architecture will be used with computer-generated synthetic segmented sperm images.
Use machine learning to generate an avatar that responds to audio and video input from a person and generates an answer delivered by a virtual avatar with corresponding facial expressions.
This thesis topic focuses on the methods to develop and test quantum programs.