Detecting iEEG High Frequency Oscillations Using Event-mapped Hyperdimensional Computing

At the Institute of Neuroinformatics (INI), University of Zurich and ETH Zurich, we focus on event-based signal processing and other promising approaches for the hardware embodiment of a low-power bio-inspired data processing unit. We currently have access to in-house neuromorphic hardware platforms alongside control software that allows interactive chip programming via PCs.
We are currently working on models and hardware designs for the detection and localization of high frequency oscillations (HFOs). These high frequency oscillations are being investigated as epileptogenic-zone bio-markers in epilaptic patients. We are interested to analyze them by a combination of event-driven and hyperdimensional vector processing. Our approach is based on standard signal pre-conditioning followed by the generation of asynchronous events (spikes) to be used in neuromorphic processors. This project will investigate the use of such events in the context of hyperdimensional vector processing: these events will be converted to binary vectors that represent inputs for a hyperdimensional vector processing unit. The goal is to use the hyperdimensional vector processing framework in conjuction wjth the event-based representation to distinguish high frequency oscillations from background noise. This project mainly focuses on converting events to hyperdimensional vectors and designing an encoder to further combine/compare such large patterns in a pre-defined software platform as depicted in Figure 1. In case of successful HFO detection, the applicant can exploit existing hyperdimensional computing libraries to develop final HDL code for hardware implementation.