Inhalt des Dokuments
Computing on the edge
While large-scale machine learning problems are nowadays often solved on large cloud-based GPU units, other tasks such as the solution of middle size convex optimization problems can be currently solved on low-cost microcontrollers that are closer to the sensors and physical objects, provided that a proper formulation and implementation is performed.
The main goal of this project is to achieve low-energy, low-cost but highly intelligent operation of smart buildings which also reduce communication costs by computing on the edge. To achieve that goal, automatic code generation techniques will be developed, which starting from a high-level description of a problem, automatically formulate it in an optimal way for a given hardware platform. Low-cost microcontrollers, FPGAs, and standard CPUs will be explicitly considered. Cloud computing, whether in an online cloud or in a private network of each building will still have a relevant role by doing periodic updates that improve performance on the long term, rather than by performing the normal operation of the buildings.