WASP at the Department of Computer and Information Science (IDA)

Photo iStock/12521104 In comparison to 4G, there is a plethora of new applications envisioned for 5G, ranging from traditional efficient broadband communication links to more strictly reliable communication links suitable for critical communication in industrial applications or massive communication that is typically associated to the Internet of Things.

In this project we develop solutions to increase the autonomy of 5G networks with the goal to improve network performance, reduce operation complexity, and increase resilience.

Employing probabilistic machine learning models in the project is a ground for an optimal decision making, in particular to satisfy the ultra-reliability requirements where uncertainty quantifications are crucial. The project aims to use probabilistic classification and regression frameworks to guarantee flexibility and autonomy of 5G components, and also use spatial and temporal models in order to take into account dynamic aspects such as device velocity, device movement path or daily/monthly variations of the network workload.

To autonomously adapt to various environmental changes and disturbances, the project aims to embed online learning components into our methods. Efficiency and scalability are challenges in the efficient development of the autonomous 5G environment, and development of appropriate big data processing methods is yet another aim of the project.

Publications

• Caroline Svahn, Oleg Sysoev, Mirsad Čirkić, Fredrik Gunnarsson and Joel Berglund, Inter-Frequency Radio Signal Quality Prediction for Handover, Evaluated in 3GPP LTE, 2019 IEEE 89th Vehicular Technology Conference.
• Caroline Svahn and Oleg Sysoev (2022), Selective Imputation of Covariates in High Dimensional Censored Data, Journal of Computational and Graphical Statistics.

External partner

Ericsson AB

Foto iStock/Wenjie Dong The rapid deployment of streaming sensors have made spatiotemporal data increasingly common. In this project we develop probabilistic models for spatiotemporal data with applications in the field of transportation.

The main focus is on developing computationally efficient Bayesian methods for learning, prediction and decision-making for spatiotemporal network-structured data.

Subprojects

1. A Bayesian Dynamic Stochastic Block Model for Large-Scale Multilayered Networks with Applications to Airline Network Prediction.
2. Traffic Flow Modeling and Prediction using Gaussian Processes with Road Topology Structure.

Extern partner

Stockholms Lokaltrafik, SL

Publications

• Héctor Rodriguez Déniz, Mattias Villani (2022), Robust Real-Time Delay Predictions in a Network of High-Frequency Urban Buses, IEEE transactions on intelligent transportation systems (Print).
• Héctor Rodriguez Déniz, Mattias Villani, Augusto Voltes-Dorta (2022), A multilayered block network model to forecast large dynamic transportation graphs: An application to US air transport, Transportation Research Part C: Emerging Technologies
• Héctor Rodriguez Déniz, Erik Jenelius, Mattias Villani (2017), Urban Network Travel Time Prediction via Online Multi-Output Gaussian Process Regression, 2017 IEEE 20TH INTERNATIONAL CONFERENCE ON INTELLIGENT TRANSPORTATION SYSTEMS (ITSC)

Photo metamorworks

Probabilistic models and deep learning are two very successful branches of machine learning, with complementary properties. In this project, we will develop theory and methods related to the interplay between these technologies, enabling us to take advantage of the strengths of both types of methods.

Probabilistic models—where unobserved variables are viewed as stochastic and dependencies between variables are encoded in joint probability distributions—are widely used in the areas of statistics and machine learning.

Probabilistic models come with many desirable properties: they enable reasoning about the uncertainties inherent to most data; they can be constructed hierarchically to build complex models from simple parts; they provide a natural safeguard against overfitting; and they allow for fully coherent inferences over complex structures from data.

Deep learning is a different branch of machine learning which has recently had remarkable success in a range of different applications related to computer vision, natural language processing and more. In most cases, the deep neural networks (DNNs) are trained discriminatively using supervised learning, with large sets of (annotated) training data.

Unfortunately, the advancement of deep learning has come at a price—DNNs often lack many of the desirable properties of probabilistic models, such as uncertainty quantification and structure exploitation over well-defined probabilistic priors.

In this project, we will develop theory and methods related to the interplay between probabilistic models and deep learning. We intend to develop both new models and new inference and learning algorithms for applications where unobserved variables are naturally characterized using, for instance, probabilistic graphical models or stochastic processes, whereas data is from some domain where deep learning has been successful (e.g., images).

We will consider applications to clinical decision support, active learning, weak supervision, and semi-supervised learning in the context of dynamical systems. However, the family of problems that involve such interplay between deep learning and probabilistic models is general, and we expect the results from the project to be widely applicable.

External partner

Chalmers University of Technology:

• Lennart Svensson (PI)
• Jakob Lindqvist (WASP-MLX PhD student)