Existing approaches to solving combinatorial optimization problems on graphs suffer from the need to engineer each problem algorithmically, with practical problems recurring in many instances. The practical side of theoretical computer science, such as computational complexity, then needs to be addressed. Relevant developments in machine learning research on graphs are surveyed for this purpose. We organize and compare the structures involved with learning to solve combinatorial optimization problems, with a special eye on the telecommunications domain and its continuous development of live and research networks.

No abstract

Joining element design is mainly a manual task resulting in costly and prolonged development trajectories. Current limited automation solutions support engineers, but still lead to repetitive tasks and design iterations. Machine learning finds and exploits patterns in data to predict designs enabling engineers to focus on core competencies. This work proposes a novel methodology to predict joining element locations using machine learning. It describes two approaches to predict specifically spot-weld locations using voxels as data representation. The study presents a regression and classification concept with 3D fully convolutional neural networks. Coordinate-based performance measurements enable to compare and evaluate models regardless of learning tasks or data structures. Results indicate that both concepts can accurately predict joining locations by only considering geometry.

No abstract