PhD candidate in Embedded Systems Design for Distributed Deep Learning

Updated: 11 months ago
Deadline: 20 Sep 2019

The Parallel Computing Systems group at the Informatics Institute of the University of Amsterdam and the Leiden Embedded Research Center at the Leiden Institute of Advanced Computer Science of Leiden University are looking for a joint PhD candidate in the area of embedded systems design for deep learning applications. The aim is to work towards a joint doctorate degree from both the University of Amsterdam and Leiden University.

Deep Learning (DL) algorithms are an extremely promising instrument in artificial intelligence, achieving very high performance in numerous recognition, identification, and classification tasks. Even though DL has gained significant importance, it is still very challenging to implement these algorithms on resource-constrained embedded devices, thereby preventing their pervasive adoption in a vast scope of new Internet of Things (IoT) applications and markets. Thus, a step forward is needed towards implementation of the on-line execution of DL algorithms (called inference) in a distributed manner on several resource-constrained embedded devices in order to enable a shift to the edge computing paradigm which is an integral part of the IoT concept. More specifically, when DL is moved at the edge of IoT, severe performance requirements must coexist with tight constraints in terms of power/energy consumption, available processing and memory resources on small embedded devices (sensor nodes, microcontrollers, small single-board computers like ODROID and Raspberry Pi, etc.), posing the need for a distributed and heterogeneous computing platform interconnecting several of these small embedded devices. Unfortunately, designing DL algorithms such that they can be executed on this kind of distributed platforms would require advanced skills and significant manual effort, also considering that DL algorithms are primarily designed to improve only precision, without considering the aforementioned limitations of the devices that will execute the inference and the communication costs due to data exchange among the interconnected devices. The research of the PhD candidate will therefore focus on methods and techniques for automated analysis and design of distributed DL algorithms when targeting efficient implementation of their inference tasks on the aforementioned type of distributed platforms.

For more information and to apply, please see

View or Apply

Similar Positions