Model calculation unit and control unit for calculating a partial derivative of an RBF model

Patent metadata

Field	Value
Publication number	US-11360443-B2
Application number	US-201716329922-A
Country	US
Kind code	B2
Filing date	Sep 4, 2017
Priority date	Sep 7, 2016
Publication date	Jun 14, 2022
Grant date	Jun 14, 2022

Abstract

A model calculation unit for calculating a gradient with respect to a certain input variable of input variables of a predefined input variable vector for an RBF model with the aid of a hard-wired processor core designed as hardware for calculating a fixedly predefined processing algorithm in coupled functional blocks, the processor core being designed to calculate the gradient with respect to the certain input variable for an RBF model as a function of one or multiple input variable(s) of the input variable vector of an input dimension, of a number of nodes, of length scales predefined for each node and each input dimension, and of parameters of the RBF function predefined for each node.

First claim

What is claimed is:

1. A model calculation unit for calculating a gradient with respect to a certain input variable of input variables of a predefined input variable vector of a radial basis function (RBF) model with the aid of a hard-wired processor core designed as hardware for calculating a fixedly predefined processing algorithm in coupled function blocks, the model calculation unit comprising:
the processor core configured to calculate the gradient with respect to the certain input variable for the RBF model as a function of multiple input variables of the input variable vector of an input dimension greater than 1, of a number of nodes, of respective length scales predefined for each node and each input dimension, and of parameters of the RBF function predefined for each node, the processor core configured to:
calculate a first term which, for the certain input variable, indicates a difference between: (i) a value of a first node of the nodes, the first node corresponding to the certain input variable, and (ii) the certain input variable; and
calculate a second term which, for a second node of the nodes, indicates a result of an exponential function multiplied with a parameter of the RBF function predefined for the second node, wherein an argument of the exponential function corresponds to a negative of a sum of products, the products being ascertained for all input dimensions, wherein each product for each respective input dimension of the input variable vector being determined by multiplying together the respective length scale predefined for the respective input dimension and a squared difference between: (i) a value of a respective third node of the nodes corresponding to the respective input dimension, and (ii) an input variable of the multiple input variables which corresponds to the respective input dimension;
wherein a value proportional to the gradient is obtained based on a sum of second products, the second products being ascertained for all of the nodes, each second product for each node of the nodes being determined by multiplying together the first term, the second term, and the respective predefined length scale for the node.

2. The model calculation unit as recited in claim 1 ,
wherein the processor core is configured to calculate the gradient by multiplying the value proportional to the gradient by a multiplication product of: (i) 2, (ii) a value of a predefined input transformation variable which is assigned to the certain input variable, and (iii) a predefined output transformation variable.

3. The model calculation unit as recited in claim 1 , wherein the processor core is configured to calculate also the output variable of the RBF model on an input variable vector including the certain input variable, in addition to the gradient.

4. The model calculation unit as recited in claim 1 ,
wherein the processor core includes a state machine and one or multiple processing operation blocks, the one or multiple operating blocks including a multiply-accumulate (MAC) block and an exponential function calculation block, and a memory for storing the one or multiple input variable(s) of the input variable vector, the nodes, the length scales, the parameters predefined for each node and the output variable.

5. The model calculation unit as recited in claim 1 , wherein the calculation of the gradient is activated or deactivated by predefining a selection variable.

6. A control unit, comprising:
a microprocessor; and
at least one model calculation unit for calculating a gradient with respect to a certain input variable of input variables of a predefined input variable vector of a radial basis function (RBF) model with the aid of a hard-wired processor core designed as hardware for calculating a fixedly predefined processing algorithm in coupled function blocks, the model calculation unit comprising:
the processor core configured to calculate the gradient with respect to the certain input variable for the RBF model as a function of multiple input variables of the input variable vector of an input dimension greater than 1, of a number of nodes, of respective length scales predefined for each node and each input dimension, and of parameters of the RBF function predefined for each node, the processor core configured to:
calculate a first term which, for the certain input variable, indicates a difference between: (i) a value of a first node of the nodes, the first node corresponding to the certain input variable, and (ii) the certain input variable; and
calculate a second term which, for a second node of the nodes, indicates a result of an exponential function which is multiplied with a parameter of the RBF function predefined for the second node, wherein an argument of the exponential function corresponds to a negative sum of products, the products being ascertained for all input dimensions, wherein each product for each respective input dimension of the input variable vector being determined by multiplying together the respective length scale predefined for the respective input dimension and a squared difference between (i) a value of a respective third node of the nodes corresponding to the respective input dimension, and (ii) an input variable of the multiple input variables corresponding to the respective input dimension;
wherein a value proportional to the gradient is obtained based on a sum of second products, the second products being ascertained for all of the nodes, each second product for each node of the nodes being determined by multiplying together the first term, the second term, and the respective predefined length scale for the node.

7. The control unit as recited in claim 6 , wherein the control unit is an integrated circuit.

8. A method of using a control unit, comprising:
providing a control unit, the control unit including a microprocessor, and at least one model calculation unit for calculating a gradient with respect to a certain input variable of input variables of a predefined input variable vector of a radial basis function (RBF) model with the aid of a hard-wired processor core designed as hardware for calculating a fixedly predefined processing algorithm in coupled function blocks, the model calculation unit comprising:
the processor core configured to calculate the gradient with respect to the certain input variable for the RBF model as a function of multiple input variables of the input variable vector of an input dimension greater than 1, of a number of nodes, of respective length scales predefined for each node and each input dimension, and of parameters of the RBF function predefined for each node, the processor core configured to:
calculate a first term which, for the certain input variable, indicates a difference between: (i) a value of a first node of the nodes, the first node corresponding to the certain input variable, and (ii) the certain input variable; and
calculate a second term which, for a second node of the nodes, indicates a result of an exponential function multiplied with a parameter of the RBF function predefined for the second node, wherein an argument of the exponential function corresponds to a negative sum of products, the products being ascertained for all input dimensions, wherein each product for each respective dimension of the input variable vector being determined by multiplying together of the respective length scale predefined for the respective input dimension and a squared difference between: (i) a value of a respective third node of the nodes corresponding to the respective input dimension, and (ii) an input variable of the multiple input variables which corresponds to the respective input dimension;
wherein a value proportional to the gradient is obtained based on a sum of second products

Assignees

• Bosch Gmbh Robert

Inventors

• Guntoro Andre
• Markert Heiner
• Schiegg Martin

Classifications

• G06N3/048

  Activation functions · CPC title

• F02D41/2416

  Interpolation techniques · CPC title

• G06F7/5443

  Sum of products (for applications thereof, see the relevant places, e.g. G06F17/10, H03H17/00) · CPC title

• G05B13/04 Primary

  involving the use of models or simulators · CPC title

• F02D2041/1433

  using a model or simulation of the system · CPC title