Wireless Sensing
I worked as a research assistant at the Technical University of Munich for two years. During this period, I conducted research on leveraging Channel State Information (CSI) from commercial wireless devices to non-invasively monitor vital signs, such as respiratory and heart rate. (aka: getting your heart beat from your WiFi signal 🙂 )
To accomplish this, I developed a framework that integrated CSI, acceleration, and ECG data collection from multiple different sources, coupled with real-time visualization and evaluation for vital sign analysis. Moreover, I explored multiple machine learning techniques to classify distinct movement patterns from WiFi signals. A detailed description and the project’s technical report can be found here.
Quantum Computing
The ZX-Calculus
Additionally, one of my foci has been quantum computing. After completing a foundational university course on the topic, I immediately started out on building my own quantum state vector simulator. As part of a seminar, this personal project eventually evolved into an interactive application for visualizing the ZX-Calculus, a generalized, alternate representation for quantum circuits.
I then chose to do a guided research project on leveraging this diagrammatic notation to automatically optimize the underlying quantum circuits. More details as well as the report can be found here.
Tensor Network Contraction
Subsequently, I also decided to write my master’s thesis on quantum circuit optimization. I worked on improving the performance of a method creating a circuit for Hamiltonian simulation using Riemannian optimization. The existing approach worked similar to other optimization methods such as gradient descent combined with backpropagation to approximate an arbitrary unitary time evolution operator with a brickwall quantum circuit. This approach was based on tensor network contraction which worked well for small systems but failed due to high computational complexity on larger systems.
My contribution was redesigning, implementing, and optimizing this contraction algorithm. Using a different approach as well as a structured caching methodology I was able to reduce the problem for a 16-qubit system from a complexity in the range of multiple exaoperations to a realistic runtime. This allows the optimization procedure to be performed for 2-dimensional systems with 16 qubits resulting in shorter circuits compared to traditional methods for Hamiltonian simulation. The abstract as well as the final thesis can be found here.