EFFICIENT MACHINE LEARNING INFERENCE FOR EMBEDDED SYSTEMS WITH INTEGER BASED RESTRICTED BOLTZMANN MACHINES CLASSIFIERS
Nowadays, there exist many emerging applications for embedded systems, such as computer vision and speech recognition, which heavily rely on machine learning classification. The typical approach to carry out machine learning related tasks in embedded systems is to use the embedded device as a sensor which collects data, and then carry out the classification computations in the cloud. However, this approach presents issues in several aspects such as latency, power consumption, network bandwidth, privacy, and security. Thus, it would be ideal to perform the machine learning computations in the ...
(For more, see "View full record.")