The Advanced Network Management and Control laboratory at Electrical Engineering (EE) department of TU/e has two PhD positions in the field of edge network and computing resource allocation using artificial intelligence at the edge (Edge AI).PhD 1 - Edge Network Resource Availability Prediction
PhD 2 - In-network Distributed Resource Planning
Goal and backgroundComputing resources available at or nearby a cyber-physical system have much slower upgrade cycles than the algorithms they serve. This makes maintenance very expensive as physical access by experts becomes necessary for upgrades or complete replacements.
TU/e ,
ASTRON and
Thermo Fisher Scientific join forces to address the problem exploiting novel AI architectures at the edge, like distributed reservoir computing. The project, Autonomous Distribution Architecture on Progressing Topologies and Optimization of Resources (
ADAPTOR - ref.18651), was funded by the
Open Technology Program (OTP) of
NWO Domain Applied and Engineering Sciences (TTW).
ADAPTOR extends the autonomy of these systems by building an intelligent fog/edge solution able to aggregate all resources of interconnected devices into a single distributed pool and assign tasks to it. The goal of ADAPTOR is to accurately predict the availability of resources at any moment and move the task to the best resource. The higher the resource utilization the more effective ADAPTOR solution is. ADAPTOR's diverse use cases, electron-microscope farms and space-based radio telescope swarms, bring challenges on time-sensitivity, tight resource constraints and scalability.
Research ChallengesRequirements derived from the two use cases drive the research challenges of the project. Both cases refer to networks of resources at the edge operating within some kind of isolation. In electro-microscope farms, cost and latency requirements make access to cloud resources prohibitive. Similarly, space-based radio telescope swarms cannot afford data offloading to nearby stationary computing resources, e.g. on Earth, due to power constraints and topology dynamics. It is therefore necessary to harvest unused resources within the farm or swarm.
The project will develop novel distributed AI mechanisms able to predict over time the availability of nearby resources (network and compute). Task migration can then start proactively improving resource utility.
RolePhD 1 will analytically and experimentally study the problem of scheduling local and remote compute/storage and network resources for a limited number of resource locations and workload sources.
- Model the resource availability prediction and allocation problems under dynamic constraints and uncertainties coming from communication and other hardware-related errors.
- Design novel AI schemes for resource availability predictions and brain-inspired models for high prediction throughput and sustainable resource utility.
- Experiment with the project's use cases on both simulated setups and real-world deployments. The work is analytical on neural circuits and machine learning as well as practical on real-world edge computing infrastructures.
PhD 2 will devise efficient and accurate distributed AI mechanisms. This topic will investigate scalable ways of mapping the ADAPTOR architecture to physically distributed devices. In brief:
- Model the impact of underlying communication networks performance on distributed AI mechanisms used for resource allocation and availability predictions.
- Design novel distributed AI schemes able to harvest inherent non-linearities to predict resource availability and improve predictions throughput.
- Experiment with future-proof designs on the project's use cases.
The work is strongly experimental on simulated and real-world distributed computing systems.
Work environmentEindhoven University of Technology (TU/e) is one of Europe's top technological universities, in the heart of one of Europe's largest high-tech innovation ecosystems - the Eindhoven Brainport region. Research at TU/e is a combination of academic excellence and a strong real-world impact through close collaboration with regional and international high-tech industries.
The candidate will be employed within the Electro-Optical Communications Group (ECO), in particular within the
advanced network management and control laboratory. The candidate will strongly interact with the ECO group, which consists of over 70 researchers. This position is embedded within the Center for Wireless Technology (CWT/e) at TU/e which focuses on four programs: Ultra-High Data- Rate Systems, Ultra-Low Power and Internet-of-Things Communication, Terahertz Technology, and Radio Astronomy.