Unlock instant, AI-driven research and patent intelligence for your innovation.

Method for configuring a neural network

A technology of neural network and hardware, applied in the field of configuring neural network

Pending Publication Date: 2021-06-22
ROBERT BOSCH GMBH
View PDF0 Cites 0 Cited by
  • Summary
  • Abstract
  • Description
  • Claims
  • Application Information

AI Technical Summary

Problems solved by technology

[0003] The reason for this could be, for example, that the functionality of the inference hardware is only accessible through the Framework, but it could also be in the inference hardware itself if it has a random element

Method used

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
View more

Image

Smart Image Click on the blue labels to locate them in the text.
Viewing Examples
Smart Image
  • Method for configuring a neural network
  • Method for configuring a neural network
  • Method for configuring a neural network

Examples

Experimental program
Comparison scheme
Effect test

Embodiment Construction

[0027] The core idea of ​​the present invention is to configure or organize the neural network so that the neural network can be robustly trained for reasoning hardware. In this way, hardware / software errors of the inference hardware, the technical details of which are often unknown or not fully known, can advantageously be compensated for. In this way, extensive checks of the exact hardware properties of the inference hardware can advantageously be dispensed with.

[0028] Without the proposed method, one would have to technically spend a lot to simulate the response of the inference hardware also on the training hardware.

[0029] The present invention enables the training of neural networks on training hardware for dedicated embedded inference hardware which is not fully disclosed or which, despite being disclosed, seriously delays technical research and development activities on inference hardware due to its high level of complexity, This may be due, for example, to limit...

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

PUM

No PUM Login to View More

Abstract

A method for configuring a neural network 11. The method includes: feeding image data 1 to the neural network 11 implemented on a training hardware 10; feeding the image data 1 to a neural network 21 implemented on an inference hardware 20; ascertaining a deviation between output data 12 of the training hardware 10 and output data 22 of the inference hardware 20; and ascertaining noise parameters R for the neural network 11 in such a way that after feeding the image data 1 to the neural network 11 implemented on the training hardware 10 and after feeding image data 1 to the neural network 21 implemented on the inference hardware 20, the output data 22 of the inference hardware 20 and the output data 12 of the training hardware 10 are bit-identical.

Description

technical field [0001] The invention relates to a method for configuring a neural network. The invention also relates to a method for training a neural network having noise parameters which are determined according to the proposed method. The invention also relates to a computer program. The invention also relates to a machine-readable storage medium. Background technique [0002] The current DNN (English deep neural network (deep neural network)) reasoning hardware is mainly a compromise between chip area and performance. Because floating-point computer devices are more expensive than digital signal processors (DSPs), DSPs are used in most cases for DNN calculations. A common practice in this context is to train the DNN offline, where the inference hardware is provided with a fixed-point value DNN. Since trade-offs are often made on the hardware side in order to save chip area and operating costs, it often happens that arithmetic operations are performed that are not un...

Claims

the structure of the environmentally friendly knitted fabric provided by the present invention; figure 2 Flow chart of the yarn wrapping machine for environmentally friendly knitted fabrics and storage devices; image 3 Is the parameter map of the yarn covering machine
Login to View More

Application Information

Patent Timeline
no application Login to View More
Patent Type & Authority Applications(China)
IPC IPC(8): G06N3/08G06N3/04G06N3/063
CPCG06N3/082G06N3/084G06N3/063G06N3/045G06N5/04G06N3/08G06V10/95G06F18/2148G06F18/217G06F18/243
Inventor J·E·M·梅奈特
Owner ROBERT BOSCH GMBH