The recent advances in spiking neural networks (SNNs), standing as the next generation of artificial neural networks have demonstrated clear computational benefits over traditional frame/image-based neural networks. In contrast to more traditional artificial neural networks (ANNs), SNNs propagate spikes, i.e., sparse binary signals, in an asynchronous fashion. Using more sophisticated neuron models, such brain-inspired architectures can in principle offer more efficient and compact processing pipelines, leading to faster decision making using low computational and power resources, thanks to the sparse nature of the spikes. A promising research avenue is the combination of SNNs with event cameras (or neuromorphic cameras), a new imaging modality enabling low-cost imaging at high speed. Event cameras are also bio-inspired sensor recording only temporal changes in intensity. This generally reduces drastically the amount of data recorded and in turn can provide higher frame rates, as most static/background objects (when seen by the camera) can be discarded. Typical applications of this technology include detection and tracking of high-speed objects, surveillance, and imaging/sensing from highly dynamic platforms.
While exploiting fully SNN architectures will require new hardware platforms, most of the computer vision tasks above can be addressed with traditional computing resources, provided that SNNs are implemented efficiently and trained appropriately. This project proposes to investigate a recent family of SNNs referred to as probabilistic SNNs (PSNNs) which rely on statistical formulations of the neuron models and offer new ways to design SNN architectures, as well as train them (as deep Bayesian networks). The development of spiking architectures finds applications beyond event cameras, as it can be used for applications where data can be interpreted as asynchronous streams of “events”. For instance, in the context of RF/EO sensing, SNNs could in the future be used to process series of detection events (actual targets but also clutter detections. Such solutions however require a better understanding of the training processes applicable to SNNs.
The student will be hosted within the School of Engineering and Physical Sciences (EPS) at Heriot-Watt University, within the group of Dr. Altmann. The student will benefit from the support of the UDRC consortium (Altmann Co-I, leader of WP 1.1 on scalable inference), computational resources available within the School of Engineering and Physical Sciences and the proximity of several UDRC Co-Is at Heriot-Watt University and the University of Edinburgh. Beyond the main applications using event cameras, the student will also benefit from the expertise of HWU in SPAD-based sensing, which is another promising modality well adapted for SNNs. As an HWU PhD student, the student will be supported by the HWU Postgraduate Research Development Program which offers a collection of interactive workshops, online courses and other training opportunities to develop as a successful researcher.
All applicants must have or expect to have a 1st class MPhys, MEng, MSci or equivalent degree in electrical engineering, applied mathematics, physics, computer science, or a related discipline, by Summer 2023. Selection will be based on academic excellence and research potential, and all short-listed applicants will be interviewed (in person or via Teams). A stipend will be provided for the 4-year duration of the Scholarship. This will be at the rate set by UKRI on an annual basis (based on Oct. to Sept. year). The current stipend rates can be found here
The screening of candidates will start in February 2022 and the first selection process will take place on 28/02. If the scholarship is not allocated, the screening will continue until the position is filled.
Expressions of interest and full applications (single PDF including a motivation letter, full transcript of record, CV, and names of 2 references) should be sent by email to Dr. Altmann ([Email Address Removed])