M&CS participating in Flagship Telecom

Earlier this week, telecom and ICT service provider KPN and TU Eindhoven have announced a collaboration to develop high-quality technology for the telecom sector. Software, data analytics and, for example, artificial intelligence are playing an increasingly important role in this sector. The strategic cooperation focuses on four centers of TU/e, namely: Center for Wireless Technology (CWTe Lab), Data Science Center Eindhoven (DSC/e), Smart Cities Program, and the Institute for Photonic Integration.

The Data Mining group, headed by prof.dr. Mykola Pechenizkiy, part of our Department of Mathematics and Computer Science and the Data Science Center Eindhoven, will participate in the Flagship Telecom. They are going to work on the project called ‘Extending Machine Learning from Core to Edge’.

DLCE: Deep Learning on the Core and on the Edge
Recent progress in Machine Learning (ML) using Deep Learning methods has brought significant breakthroughs in areas such as Computer Vision, Speech Recognition and Natural Language Processing. Many of these solutions execute complex models over input data that requires significant compute capacity. This is commonly implemented in large compute clusters that reside in the Cloud.

However, as the complexity of the applications grow (i.e. streaming video processing, virtual/augmented reality, mobile solutions) the computer network infrastructure between the point of service and the compute infrastructure is becoming a bottleneck. With the advancements of hardware at the edge of the network, where the actual service is needed, the possibility to move some or most of the computational load is becoming evident.

 Thus, we aim to study methods for extending Deep Learning models from the core of the network into models that exist both on the edge and on the core. The main goal is formulated as: taking into account network factors such as available resource on the edge, connectivity, bandwidth, latency and jitter, provide reliable, efficient and scalable Deep Learning based solutions to a multitude of connected devices.

Relevant topics are:
- DLCE Reliability: provide better guarantees for quality of service and graceful degradation in case of lack of resources 
- DLCE Efficiency: utilize the available resources at the edge such that the delivered quality is maximized
- DLCE Scalability: distribute larger capabilities to the edge and enable better scalability
- DLCE Improved experience: empower edge devices to provide to provide low latency interaction by distributing more service components to the edge