Abstract
The assignee for this patent application is Aistorm Inc. (Cupertino, California, United States).Reporters obtained the following quote from the background information supplied by the inventors:“Analog machine learning holds the promise of significant powe r savings, performance improvement,accuracy improvement and noise reduction com pared to digital implementations. Generally, machinelearning requires the imple mentation of a weighted summer, which may have a linear shift called a bias,as well as an activation or decision means to produce an output. Each output may be connected to manyother weighted summers and these connections may need to be d ynamically programmable, for examplein recurrent networks, or virtual neuron sc hemes, as do the weights and bias values in all networks.
NETL
NSTL