Projektämnen / Project themes
Below is a collection of project areas that are offered by our pool of supervisors. Ideally, you identify different project areas of your interest. You can then develop a research project together with your supervisor in the area of your interest!
Your supervisor's role is to help you navigate in your research efforts with focus on generic aspects of problem formulation, project scoping, planning etc. Keep in mind that it is you who come up with a concrete project idea, acquire required competences, obtain necessary resources (e.g. data), read up on the state of the art --- in other words, you take care of the subject related content, as your supervisor may well not be an expert in the field of your study.
Please fill in a KTH form questionnair to submit your preferred topic options by Monday the 20th of January, 2025 - noon (12h, midday).
Use this KTH web form to submit your teams: https://www.kth.se/form/cs-kexjobb-2025
Anyone who submits before Monday Jan 20, 2025, noon will have the same chances of getting supervisors of choice.
You need to send
-) your group information: two names and two KTH email addresses
-) multiple topic area suggestions: ideally 3 areas in ranked order
If you plan to work and write in Swedish, please indicate this in your form, so we find a supervisor who speaks Swedish.
Any other questions? Please send email to kexjobb.csc@ncslab.se. Thank you!
=================================================================
Biologically Detailed Neuron Networks (English)
Large-scale simulations of the brain areas and interconnected neural circuits becomes a common tool in the studies of the brain functions and neurological diseases in neuroscience. In order to set up a simulation, the data pre-processing step is needed involving 3D reconstruction of the morphology of individual neurons, cell placement within the simulated brain volume and deciding the connectivity of the nervous cells based on the proximity of their neurites. Numerical methods, data structures and algorithms for the network setup are needed for serial and parallel processing with great scaling properties.
Alexander Kozlov <akozlov@kth.se>
=================================================================
Computational Modelling and Analysis of Brain Activity (English)
I work in the field of computational neuroscience. We are interested in the activity dynamics and information processing in biological neuronal networks. In particular, we are studying how properties of neural hardware (neurons and synapses) affect the dynamics of neuronal networks. To this end, we use both numerical simulations (computer models) as well as the analytical tools from Physics. In parallel, we are analysizing neuronal activity recorded form behaving animals and characterizing animal behavior into meaningful states. For this work we rely on signal processing tools and machine learning (especially to classify animal behavior). Our approach to data analysis is model-driven not data-driven.
Arvind Kumar <arvkumar@kth.se>
=================================================================
Simulation and Visualisation of Crowd Behaviour (English)
Crowds of realistically-behaving synthetic characters have many application domains, from special effects for entertainment, where they replace expensive extras and stunt performers, to evacuation, urban planning and traffic simulations, where the safety and feasibility of designs can be tested and modified prior to the costly construction of real environments. This project area covers all aspects of 3D characters and embodied AI, including graphical methods for rendering and shading characters, animating the detailed full-body behaviour of individual characters (for example, via LLMs), simulation of collision avoidance for high density crowds, and visualisation methods for crowd datasets. See the following student project page for some examples of previous projects: https://www.csc.kth.se/~chpeters/ESAL/studentprojects.html
Christopher Peters <chpeters@kth.se>
=================================================================
Multi-Agent Strategic Planning (Swedish / English)
In this project we will investigate how a team of intelligent agents can coordinate their actions to achieve a common objective. More specifically, we will study the notion of knowledge: what information does each agent need to keep during the mission and how is it to be updated. In particular, we are interested in how the ability of the team to achieve objectives is increased if the agents keep nested knowledge of the type "agent A knows that agent B knows that agent A knows that the door is closed". We will study these questions in the context of a simple class of games, called multi-player games over finite graphs.
Dilian Gurov <dilian@kth.se>
=================================================================
HPC processor architectures (English)
General-purpose architectures of computing devices like CPUs are becoming increasingly less suitable for high-performance computing. They are not providing sufficient performance and/or consume too much power relative to the delivered compute performance. This development sparked a new interest in research on how to optimise such architectures for different applications. Such research is facilitated by performance models or simulation tools that allow investigating how different types of numerical kernels can benefit from new architectures. Another dimension is the exploration of new instructions like the (meanwhile less new) SIMD instructions for Arm or the RISC-V vector instructions.
Dirk Pleiter <pleiter@kth.se>
=================================================================
Computational Neuroscience and Machine Learning (Swedish / English)
I supervise projects in computational neuroscience, simulations of how neurons operate in the brain. I also supervise projects in machine learning, methods for classification or regression of static (like images) or temporal (like music) data, or methods to obtaining representations like self-supervised learning, self-organizing feature maps or the like.
Erik Fransén <erikf@kth.se>
=================================================================
Accelerated Parallel Computing with GPU and Quantum Computing (English)
Accelerated parallel computing combines emerging technologies such as GPU, Quantum Computing, and heterogeneous memory to address compute-intensive and memory-intensive workloads. Recent data centers leverage specialized accelerators like RISC-V vector units, GPUs, and emerging quantum co-processors, along with CXL-attached memory pooling to accelerate workloads with high computation or memory needs. Meanwhile, quantum computing has emerged as a revolutionary paradigm capable of enabling novel applications. State-of-the-art quantum computing frameworks include IBM's Qiskit, Google's Cirq, Rigetti’s PyQuil, and Nvidia's cuQuantum. Quantum computing has found applications in areas such as random number generation, Shor’s algorithm for integer factorization, quantum optimization, and quantum neural networks. Existing applications of quantum computing include random number generators, quantum arithmetics, quantum walks, quantum Fourier transform, Shor’s algorithm for integer factorization, Quantum Approximate Optimization Algorithm (QAOA), quantum simulation of the Schrödinger Equation, and quantum neural networks, to name a few. A project example is to design and develop an accelerated application on a new accelerator or memory type, either using real hardware or a simulator and then compare the performance and resource usage. Another project example is to design and develop a quantum algorithm using existing programming interfaces and compare its accuracy and performance with classical approaches. Other examples are the study of noise impact (using noise models from the simulators) on a given quantum algorithm and the comparison of different quantum computing paradigms, e.g., qubit-based vs. continuous variable vs. quantum annealing approaches, for solving the same problem.
Ivy Bo Peng <bopeng@kth.se>
=================================================================
Robot planning algorithms (English)
In our research, we design planning, decision making, and control algorithms for autonomous robots. Our goal is to enable safe, purposeful and trustworthy autonomous behavior. The core of our approach is the use of formal methods-inspired techniques to achieve (probabilistic) guarantees that robots behave as they are expected. In the kexjobb projects, we would like to focus on planning for robots that interact with dynamic environment; with each other, with people, or dynamic objects. Example topics may include integration of short-term predictions in long-term planning, multi-robot coordination with unknown robots’ states, etc.
Jana Tumova <tumova@kth.se>
=================================================================
DigiMat - lead your own project in the largest online course at KTH! (Swedish / English)
Watch the trailer to get an overview: http://digimat.tech/digimat/
Links to an external site.
DigiMat is a unique educational program, unifying math and programming, from preschool to top academic and professional, developed by leading scientists. Drive and build your own projects and initiatives globally based on interactive simulation activities you develop in web-based editable Digital Math JavaScript or Python.
Johan Jansson <jjan@kth.se>
=================================================================
Neuronal Networks and Neuromorphic Computing (English)
We develop brain-inspired algorithms and hardware for real-world systems, typically embedded sensors, human-machine interfaces, or robots. In the kexjob course projects we want to focus on vision sensors and simulated or real mobile robots for control tasks. Examples for research projects are algorithm design for fast visual classification and tracking of objects, gesture recognition, and motor control (wheeled or legged robots). We might include machine learning on collected data sets when appropriate.
Jörg Conradt <conr@kth.se>
=================================================================
Machine Learning in Computer Vision and Biomedical Image Analysis (English)
My research area focuses on computer vision and in particular on applications in biology and medicine. We are interested in how machine intelligence can help precisely diagnose conditions, reduce patient risk, choose effective treatments, and further our understanding of biological systems. In addition, we are also interested in traditional problems in computer vision such as segmentation, detection, classification, generative models, as well as network architectures, learning approaches, etc.
Kevin Smith <ksmith@kth.se>
=================================================================
Machine Learning in Finance or Unsupervised Representational Learning (Swedish / English)
Model financial markets, time series, and other financial data using machine learning. Examples of problems in this domain are dimensionality reduction, clustering, change detection, and volatility estimation. Please note that while "let's try to predict the stock market" can be a valid project if you have a unique approach, you should then plan your project such that you will be able to write an interesting report even in the likely scenario where the outcome is negative. Other modeling objectives are preferred.
Unsupervised learning can be thought of as a way to model the statistics of a data distribution by learning an encoding that captures relevant features, while rejecting noise. This is an important and cutting edge topic in machine learning, where it is valuable to be able to learn even in the absence of labeled data. Representational learning has also been studied in computational neuroscience, as a model for how the brain processes sensory information, for example in the primary visual cortex. Examples of approaches to this problem are a group of methods called "efficient coding", and others that use estimates of mutual information between the input data and the learned codes as learning objectives. An example project under this topic would be to implement one or two representational learning methods from the literature, and to analyze how the resultant encoders behave in a given data domain.
Martin Rehn <rehn@kth.se>
=================================================================
Technology Enhanced Learning (English)
With each advance in technology we have an opportunity to apply it to the domain of education. The surprising results from recent advances in AI are one such timely example. Educators are looking at both the positive and negative implications, such as threats to learning and assessment on one hand, and automatically generating, explaining and assessing tasks on the other hand. Projects in this space will investigate positive and negative implications in the field of computer science education.
Reading:
*) Shanahan, M. (2022). Talking About Large Language Models. arXiv preprint arXiv:2212.03551.
*) Finnie-Ansley, J., Denny, P., Becker, B. A., Luxton-Reilly, A., & Prather, J. (2022). The Robots Are Coming: Exploring the Implications of OpenAI Codex on Introductory Programming. In Australasian Computing Education Conference (pp. 10-19).
*) Sarsa, S., Denny, P., Hellas, A., & Leinonen, J. (2022). Automatic Generation of Programming Exercises and Code Explanations Using Large Language Models. In Proceedings of the 2022 ACM Conference on International Computing Education Research-Volume 1 (pp. 27-43).
Ric Glassey <glassey@kth.se>
=================================================================