Procedure and system for processing predictions in neuronal networks in the presence of enemy disturbances

摘要

A process for processing predictions in the presence of enemy interference in a recording system comprising a processor and, coupled to it, a memory. The processor can be configured to connect to one or more sensors to receive input sizes (x). The processor can be configured to operate a module in memory to implement a neural network. The neural network may have a network function fi, i being network parameters. The process can be run by the processor and may include input sizes (x),comprising at least one specified input size (xo) to generate the respective output sizes, the output sizes being predictions of the neural network and comprising a specified output size y0 corresponding to the specified input size (x0), with y0 = fh (x0). The process may also include the production of a measurement size (m) from a variety of starting sizes including the specified starting size y. The measurement size m may be at or near the specified input size (x0),The following are: (i) a first measurement size M1 corresponding to a gradient of the specified output size y, (ii) a second measurement size M2 corresponding to a gradient of a specified target function derived from a training process for the neural network, or (iii) a third measurement size M3; derived from a combination of M1 and M2; The procedure may also include determining whether the measurement size (m) is equal to or greater than a threshold value. The procedure may also include, where it is determined that the measurement size (m) is equal to or greater than the threshold value,one or more remedial actions are implemented to correct a malfunction. A classification procedure based on the procedure shall also be disclosed. A corresponding recording and/or classification system and a vehicle comprising the recording and/or classification system shall also be disclosed.