DE102020105783A1 - Method of creating a reduced neural network - Google Patents

Abstract

The invention relates to methods for generating a reduced neural network (20) for a control unit (2) of a vehicle (1), comprising the following steps: - initializing the reduced neural network (20) in a first step (11), - calculating each first corresponding output data sets (31a, 31b, 31z) with the aid of a first neural network (30) and in dependence on a respective input data set (32a, 32b, 32z), a total of the respective input data sets (32a, 32b, 32z) being a first set (32 ) forms the input data sets in a second step (12) - determining respective deviations (42a, 42b, 42z) of the first output data sets (31a, 31b, 31z) from respective comparison data sets (41a, 41b, 41z) in a third step (13 ), the respective first output data record (31a, 31b, 31z) corresponding to the respective comparison data record (41a, 41b, 41z), - adapting the reduced neural network (20) as a function a second set of input data sets, which is a subset of the first set, the first output data sets corresponding to the input data sets of the second set having a greater deviation on average than the first output data sets corresponding to the other input data sets of the first set, the other input data sets being a third Form the set of input data sets and the reduced neural network (20) is adapted independently of the third set of input data sets, in a fourth step (16), - repeating steps three and four until an abort criterion is met.

Description

The invention relates to a method for generating a reduced neural network for a control device of a vehicle and / or a vehicle component, a control device for a vehicle and / or a vehicle component, a vehicle, a reduced neural network and a computer program product.

The DE 10 2018 103 113 discloses a control device of a vehicle with a neural network.

• - Initialisieren des reduzierten neuronalen Netzes in einem ersten Schritt,
• - Berechnen von jeweiligen ersten korrespondierenden Ausgangsdatensätzen mithilfe eines ersten neuronalen Netzes und in Abhängigkeit eines jeweiligen Eingangsdatensatzes, wobei eine Gesamtheit der jeweiligen Eingangsdatensätze eine erste Menge der Eingangsdatensätze bildet, in einem zweiten Schritt,
• - Bestimmen von jeweiligen Abweichungen der ersten Ausgangsdatensätze von jeweiligen Vergleichsdatensätzen in einem dritten Schritt, wobei der jeweilige erste Ausgangsdatensatz mit dem jeweiligen Vergleichsdatensatz korrespondiert,
• - Adaptieren des reduzierten neuronalen Netzes in Abhängigkeit einer zweiten Menge der Eingangsdatensätze, die eine Teilmenge der ersten Menge ist, wobei die zu den Eingangsdatensätzen der zweiten Menge korrespondierenden ersten Ausgangsdatensätze im Durchschnitt eine größere Abweichung aufweisen als die zu den übrigen Eingangsdatensätzen der ersten Menge korrespondierenden ersten Ausgangsdatensätze, wobei die übrigen Eingangsdatensätze eine dritte Menge der Eingangsdatensätze bilden und das reduzierte neuronale Netz unabhängig von der dritten Menge der Eingangsdatensätze adaptiert wird, in einem vierten Schritt,
• - Wiederholen der Schritte drei und vier, bis ein Abbruchkriterium erfüllt ist.

The proposed method for creating a reduced neural network for a control device of a vehicle and / or a vehicle component comprises the following steps:

• Initializing the reduced neural network in a first step,
• Calculating respective first corresponding output data sets with the aid of a first neural network and depending on a respective input data set, a total of the respective input data sets forming a first set of the input data sets in a second step,
• Determining in a third step the respective deviations of the first output data sets from the respective comparison data sets, the respective first output data set corresponding to the respective comparison data set,
• - Adapting the reduced neural network as a function of a second set of input data sets, which is a subset of the first set, the first output data sets corresponding to the input data sets of the second set having, on average, a greater deviation than the first corresponding to the other input data sets of the first set Output data sets, the remaining input data sets forming a third set of input data sets and the reduced neural network being adapted independently of the third set of input data sets, in a fourth step,
• - Repeat steps three and four until an abort criterion is met.

The input data records of the third set are not included in the second set. The term “network” is used below for the term “neural network”.

The input data sets, the first output data sets and the comparison data sets can each be vectors. The respective deviation of the first output data sets can be determined, for example, with the aid of a vector standard of a respective difference vector. The respective difference vector is preferably calculated from a difference between the corresponding first output data set and the corresponding comparison data set. The respective first corresponding output data records mean in each case a single first output data record which is calculated as a function of the respective input data record corresponding to this output data record.

Corresponding the respective first output data record with the respective comparison data record means that these two data records are related to the input data record with which the respective first output data record is calculated. For example, the respective comparison data record can be calculated using the reduced network and depending on the input data record with which the first output data record corresponding to this input data record is calculated on the basis of the first network. In this case, the corresponding comparison data record is preferably an output data record of the reduced network.

The control device of the vehicle and / or the vehicle component is set up to control at least one operating parameter of the vehicle and / or the vehicle component. The operating parameter can be, for example, a speed, an acceleration, distances between the vehicle and another vehicle, an engine speed, a braking force or a steering angle. The vehicle and / or the vehicle component can advantageously be used for autonomous and / or partially autonomous driving.

The proposed method and the control device enable control of the autonomous and / or partially autonomous vehicle and / or the vehicle component in various driving situations. Driving situations can be, for example, driving on the motorway and / or driving within cities and / or driving in traffic-calmed areas.

Steps three and four are repeated until a termination criterion is met. Furthermore, a different order than the order of the steps mentioned so far is possible. For example, the second step can be carried out before the first step.

It is possible that the reduced network is only adapted in the fourth step as a function of a single input data record. In this case, the second set has only this one input data record.

The reduced network can be adapted as a function of the second set of input data sets, for example in the event that the second set has only a single input data set, as follows. The reduced network is adapted depending on the first output data set corresponding to the input data set of the second set and / or its deviation. Since this corresponding first output data set or its deviation was calculated using the input data set of the second set, the adaptation takes place indirectly as a function of this input data set. The dependency also means an indirect dependency.

Furthermore, a fourth set of input data sets is to be considered, which summarizes all those input data sets which in the fourth step are repeated for the second set when steps three and four are repeated several times.

The first set is advantageously the set of input data sets that was used for a training session or for a training session and a testing of the first network before the second or first step is carried out. In this sense, the method can also include an initial step in which the first network is generated using the first set of input data records.

The second set and the fourth set advantageously contain less than a third of the input data records of the first set. The first advantage of adapting the reduced network as a function of the second set is that a computing effort during adaptation can be reduced. This differs from a learning method, for example of the first network, in which the first network is adapted as a function of all input data records of the first set. With such a learning method, the first network can be adapted during training as a function of an averaged error, which is calculated as a function of all output data records of the first network, which were calculated using all input data records of the first set.

A second advantage of adapting the reduced network as a function of the second quantity is that the computing effort of the reduced network can be reduced when the reduced network is used in the control unit. This can be brought about in that the reduced network can be less complex than the first network and one dimension of a diversity of the reduced network can be smaller than one dimension of a diversity of the first network. The lower complexity and / or the smaller diversity can have the result that difference quotients that can be formed using the reduced network can be lower than difference quotients that can be formed using the first network. This advantage and the lower computing effort when adapting the reduced network improves the possible uses of the reduced network compared to the first network, in particular when the reduced network is used in a control loop of the control device.

With the respective diversity there is in each case a set of possible data records that the reduced network or the first network can calculate for determining possible output data records of the corresponding network as a function of all possible input data records for the respective networks. In other words, the first set of input data records is based on the diversity of the reduced network or the first network using an internal logic of the reduced network or the first network, i.e. can be mapped to the respective amount of possible data records.

The definition of a manifold used here differs from the mathematical definition of a manifold, in which the manifold must fulfill certain properties, for example that it is a Hausdorff space or locally Euclidean. For example, a manifold can be discontinuous or have holes in the sense of the invention.

According to a first variant, the possible data records can be the possible output data records. According to a second variant, the possible data records can be internal data records of the respective network which are not routed to an output of the network. In the second variant, the respective manifold is referred to below as the internal manifold.

A third advantage of adapting the reduced network as a function of the second set consists in the fact that the input data records of the second set can strongly influence a change in the internal logic of the reduced network when adapting. The adaptation of the reduced network can thus be accelerated. The third advantage can be achieved particularly well if the second set has only a single input data record.

The second set is advantageously formed such that the respective deviations of the first output data sets corresponding to the input data sets of the second set are greater than all individual deviations of the first output data sets of the third set. With this configuration, the third advantage can be implemented particularly well. In this case, namely those input data sets which have a particularly strong influence on the adaptation of the reduced network and which produce high deviations. In this case, the fourth set can be regarded as a set of the input data sets that contains the most relevant input data sets.

Based on this configuration, the third advantage can be achieved even more easily if the second set has only a single input data record. When repeating the fourth step, a single input data record of the first set is preferably not contained in the second set again. This means that different input data sets, which each have a particularly strong influence on the adaptation of the reduced network, are used for the adaptation in succession. This can further accelerate the adaptation.

According to a further development, it is provided that the reduced network has basic functions. In this development, the reduced network is adapted in the fourth step in such a way that at least one further basic function is added to the reduced network. By adding the further basic function, a dimension of the manifold, preferably the internal manifold, of the reduced network is increased. The further basic function is preferably added indirectly as a function of the second set of input data records. For example, the further basic function can be determined with the aid of those first output data sets which correspond to the input data sets of the second set. The first set of input data records can preferably be mapped to the diversity of the reduced network using the basic functions including the further basic function. The advantage of this development is that each time the dimension of the diversity of the reduced network is increased, the internal logic of the reduced network is adapted to information contained in the input data record of the second set and in the first output data record corresponding to this can be.

Information that is contained in the internal logic of the first network can also be taken into account when adapting the reduced network and can thus also be represented by the internal logic of the reduced network. If steps three and four are successively repeated, the internal logic of the reduced network can be adapted to the internal logic of the first network with each repetition. It is thus possible to use the reduced network to achieve a performance which is similar to a performance of the first network, the reduced network additionally having the advantages mentioned above. When using the reduced network in the control loop, its performance may even be better than that of the first network for the following reason. The diversity of the reduced network can have a softer course in a possible output space of the reduced network than the diversity of the first network in the possible output space. As a result, the difference quotients of the reduced network described above can be lower, which can make it easier to regulate the vehicle component using the control loop.

Both the basic functions and the further basic function have no input or output, i.e. no argument or function value, but are defined by unchangeable values. The basic function and the further basic function vectors with unchangeable entries are advantageous.

The proposed reduced neural network can be generated according to one of the configurations mentioned above, preferably also according to one of the configurations mentioned below.

The proposed control device is suitable for the vehicle and / or the vehicle component and is set up by a method and / or for carrying out a method according to one of the embodiments already described, preferably also according to one of the embodiments mentioned below. The control device advantageously has the reduced neural network, as can be generated according to one of the above-mentioned, preferably also according to one of the following, configurations after the method has been carried out.

The proposed vehicle includes the control unit. Because the proposed control unit and the proposed vehicle have the reduced network or can generate this when the method is carried out, the same advantages can be achieved with the vehicle and the control unit as with the reduced network.

The proposed computer program product causes a computer, when the computer program product is executed by the computer, to carry out the proposed method, preferably according to the variants mentioned below. The computer program product can be stored in the control device or, in a further variant, can be present as a program on an external data carrier outside the vehicle. The computer is advantageously located in the control unit and can be designed in the form of a microcontroller. By using the computer product, the proposed method on the Computer can be executed, it has the advantages mentioned below for the method.

The first and reduced network can be designed in the form of a respective special data structure which, when loaded from a memory of the computer into a main memory of the computer and called up from the main memory by the computer program product, can initiate the following actions. With the aid of the first or reduced network, a corresponding physical neural first or reduced network with neurons can preferably be simulated, output signals from a plurality of neurons of the corresponding physical network being routed to inputs of further neurons of the corresponding physical network. The neurons of the first and reduced networks thus correspond to the respective neurons of the corresponding physical networks and can be viewed as models with inputs and outputs, preferably only one output.

In addition, the first or reduced network can have respective activation functions and / or threshold values for each neuron of the corresponding network. The activation functions and / or threshold values can also be in the corresponding special data structure, i.e. in the first or reduced network.

• 1 Ein Fahrzeug mit einem Steuergerät und Fahrzeugkomponenten,
• 2 Schritte eines Verfahrens zur Erzeugung eines reduzierten neuronalen Netzes für das Steuergerät nach 1,
• 3 ein erstes neuronales Netz,
• 4 ein Datenfluss zur Berechnung von Abweichungen zwischen Ausgangsdatensätzen des ersten neuronalen Netzes und des reduzierten neuronalen Netzes,
• 5 eine Mannigfaltigkeit des ersten Netzes nach 3,
• 6 das reduzierte neuronale Netz nach einem Initialisieren des reduzierten neuronalen Netzes,
• 7 das reduzierte neuronale Netz nach einem Hinzufügen einer weiteren Basisfunktion,
• 8 das reduzierte neuronale Netz nach Beendigung des Verfahrens nach 2,
• 9 eine Mannigfaltigkeit des reduzierten neuronalen Netzes nach Beendigung des Verfahrens.

The dependent claims describe further advantageous embodiments of the invention. Preferred exemplary embodiments are explained in more detail with reference to the following figures. The figures show schematically in:

• 1 A vehicle with a control unit and vehicle components,
• 2nd Steps of a method for generating a reduced neural network for the control device 1 ,
• 3rd a first neural network,
• 4th a data flow for calculating deviations between output data sets of the first neural network and the reduced neural network,
• 5 a diversity of the first network 3rd ,
• 6 the reduced neural network after initialization of the reduced neural network,
• 7 the reduced neural network after adding another basic function,
• 8th the reduced neural network after completing the procedure 2nd ,
• 9 a diversity of the reduced neural network after completion of the procedure.

1 shows a vehicle according to the invention 1 with a control device according to the invention 2nd . The vehicle 1 can a camera 3rd , a distance sensor 4th , a measuring device 5 for recording operating parameters of the vehicle 1 and a facility 6 for detecting a state of a driver of the vehicle 1 exhibit. The operating parameters can be the above parameters. In addition, the vehicle has at least a first vehicle component 7 , a second vehicle component 8th and a third vehicle component 9 , each of which can be, for example, a drive unit, a chassis component or a safety element, such as an airbag or a belt system.

2nd shows individual steps of the inventive method for generating a reduced neural network 20th for the control unit 2nd of the vehicle 1 . A first step 11 involves initializing the reduced neural network 20th . A second step 12th comprises calculating respective first corresponding output data sets ( 31a , 31b , 31z) with the help of a first neural network 30th and depending on a respective input data record ( 32a , 32b , 32z ), whereby a total of the input data sets ( 32a , 32b , 32z ) a first set 32 the input data records ( 32a , 32b , 32z ) forms.

A third step 13 comprises determining, preferably using a comparison module 44 , of respective deviations 42 of the first output data records 31 of respective comparison data sets 41 , in the 4th exemplary with 42a , 42b , 42z designated. The respective first output data record in which 4th for example with 31a , 31b , 31z designated, corresponds to the respective comparison data set, in 4th for example with 41a , 41b , 41z designated.

In the 3rd and 4th data records shown that have the same letter correspond to each other. This means that these data records can belong to a single event, in particular to a single point in time, in which, for example, a recording of sensor data of the vehicle 1 was carried out.

The first set 32 first sensor data 33a , 33b , 33z who, for example, with the camera 3rd were recorded, second sensor data 34a , 34b , 34z that with the distance sensor 4th were recorded, third sensor data 35a , 35b , 35z using the measuring device 5 were captured, and fourth Sensor data 36a , 36b , 36z that with the establishment 6 were recorded.

The first output records 31 can, for example, first control parameters 37a , 37b , 37z to control the first vehicle component 7 , second control parameters 38a , 38b , 38z to control the second vehicle component 8th and third control parameters 39a , 39b , 39z to control the third vehicle component 9 exhibit.

A fourth step 14 of the method includes adapting the reduced neural network 20th depending on a second set of input data sets. The second set is a subset of the first set 32 . The first output data records corresponding to the input data records of the second set have on average a greater deviation than those to the other input data records of the first set 32 corresponding first output data sets. The remaining input data sets form a third set of the input data sets. The reduced neural network 20th is in the fourth step 14 adapted regardless of the third set of input data sets. In other words, the reduced network 20th indirectly depending on those input data records of the first set 32 adapted, which on average the largest deviations of the first corresponding output data sets 31 produce. The dependency can also be indirect, as described above.

The third step 13 and the fourth step 14 are repeated until a termination criterion 15 is satisfied. All input data records in the fourth step are advantageous 14 are included in the second set when repeating the steps 13 , 14 collected in the form of a fourth quantity.

According to a first variant, the control device 2nd set up to carry out the individual steps of the method according to the invention according to one of the variants already mentioned or mentioned below. According to a second variant, the control device 2nd generated with the aid of the method according to the invention according to the variants already mentioned or mentioned below.

1 still shows a computer program product 10th which, when it is executed by a computer, causes the computer to carry out a method according to the variants mentioned above or mentioned below. The computer program product 10th can in the control unit 2nd be saved or in a further variant as a program on an external data carrier outside the vehicle 1 available. The computer is advantageously located in the control unit and can be designed in the form of a microcontroller. By using the computer product 10th the proposed method can be carried out on the computer, it has the advantages already mentioned for the method.

According to a preferred embodiment, the respective comparison data record 41a , 41b , 41z with the help of the reduced network 20th and depending on the corresponding input data record 32a , 32b , 32z calculated. Another embodiment provides that the first neural network 30th with the help of the first output data sets 31 and the first lot 32 the input data records were trained.

wobei $a_i\left(\overrightarrow{\mu }\right)$

ein jeweiliger Koeffizient ist, dessen Wert von seinem Argument $\overrightarrow{{\mu }_J}$

abhängt, φ_i Basisfunktionen sind und $\overrightarrow{y_J}$

ein Ausgangsvektor des ersten neuronalen Netzes 30 ist. Der Ausgangsvektor $\overrightarrow{y_J}$

des ersten Netzes 30 kann, je nachdem welcher der Eingangsdatensätze der ersten Menge 32 an einem Eingang des Netzes 30 anliegt, in Form einer der ersten Ausgangsdatensätze 31a, 31b oder 31z vorliegen. Dabei kann der Vektor $\overrightarrow{{\mu }_J}$

einer der Eingangsdatensätze 32a, 32b oder 32z sein. Alle möglichen Ausgangsvektoren $\overrightarrow{y_J},$

die das erste Netz 30 in Abhängigkeit von allen möglichen Eingangsdatensätzen $\overrightarrow{{\mu }_J}$

des ersten Netzes 30 errechnen kann, sind für ein einfaches Anwendungsbeispiel in 5 in einem dreidimensionalen Raum in Form einer Kurve 51 dargestellt. Die Kurve 51 kann in diesem Sinne eine Mannigfaltigkeit des ersten neuronalen Netzes 30 ausbilden, auf die alle möglichen Eingangsdatensätze, die als Eingabe für das erste neuronale Netz 30 verwendet werden können, mit Hilfe der internen Logik des ersten Netzes 30 auf diese Mannigfaltigkeit abgebildet werden können.According to a simple example, there is a single output data record, such as the first output data record 31a , 31b or 31z , in the form of a vector with three entries, as described in 3rd is shown. Furthermore, the input data records of the first set 32 have a large number of entries, ie up to 20,000 entries, for example. An internal logic of the first neural network 30th can be represented using the following formula: $$\overrightarrow{y_J}={\displaystyle \sum _{i=1}^n\left[{\alpha }_i\left(\overrightarrow{{\mu }_J}\right)\cdot {\phi }_i\right];\text{With}\overrightarrow{y_J}=\left(\begin{array}{l}{y}_1\\ y_{\mathrm{2nd}}\\ y_{\mathrm{3rd}}\end{array}\right),}$$

in which $a_i\left(\overrightarrow{\mu }\right)$

is a respective coefficient whose value is from its argument $\overrightarrow{{\mu }_J}$

depends, φ _{i are} basic functions and $\overrightarrow{y_J}$

an output vector of the first neural network 30th is. The output vector $\overrightarrow{y_J}$

of the first network 30th can, depending on which of the input data records of the first set 32 at an entrance to the network 30th is present in the form of one of the first output data records 31a , 31b or 31z available. The vector $\overrightarrow{{\mu }_J}$

one of the input records 32a , 32b or 32z his. All possible output vectors $\overrightarrow{y_J},$

the the first network 30th depending on all possible input data sets $\overrightarrow{{\mu }_J}$

of the first network 30th can be calculated for a simple application example in 5 in a three-dimensional space in the form of a curve 51 shown. The curve 51 can in this sense a diversity of the first neural network 30th train on the all possible input data sets, which as input for the first neural network 30th can be used using the internal logic of the first network 30th can be mapped to this diversity.

5 shows individual points as an example 51a , 51b , 51z that on the curve 51 lie and accordingly with the first output data set 31a , 31b respectively 31z correspond. Every time with the help of an input data record and the first neural network 30th If a further first output data record is calculated, another point can be computer-aided in which 5 inserted three-dimensional space and thereby the curve 51 be approximated.

The following is the generation of the reduced network 20th to be discribed. After the first step 11 points the reduced network 20th preferably an initial basis function 21 , also referred to below as Ψ ₀ , and preferably an initial weight 22 , hereinafter referred to as β ₀ . The reduced network is exemplary and not restrictive 20th a first exit 61 , a second exit 62 and a third exit 63 and preferably as many inputs 64 , like a single input record of the first set 32 of entries. The initial basic function 21 and the initial weight 22 are preferably determined with the aid of a random generator.

The second quantity advantageously has the fourth step when it is carried out, and in particular also each time it is carried out repeatedly 14 only a single input data record. The respective deviation of the first output data record corresponding to this single input data record of the second set is advantageously greater than all individual deviations of the remaining first output data records of the first set. In this case, the reduced network can be adapted in the fourth step according to a “greedy algorithm”.

For example, is the deviation 42a larger than any other deviations in the third step 13 are calculated, such as the deviations 42b to 42z , the second set can be replaced by the one with the first output data set 31a corresponding input data record 32a be formed because the first output data set 31a with the deviation 42a corresponds. Corresponding means that the deviation 42a with the help of the first output data record 31a is calculated. In this case, the remaining first output data records are the first output data records 31b to 31z . In this example, the input data records form accordingly 32b to 32z the third set of input records 32 . Adapting the reduced network 20th takes place according to this simple example in the fourth step 14 depending on the second set of input data sets 32 which in this example are given by the individual input data record 32a is formed, and regardless of the third set of input data sets 32 .

wobei ${\beta }_i\left(\overrightarrow{\mu }\right)$

eines der mehreren Gewichte 82 ist, dessen Wert von seinem Argument $\overrightarrow{{\mu }_J}$

abhängt, ψ_i Basisfunktionen des reduzierten Netzes 20 sind und $\overrightarrow{y_{J_{red}}}$

ein Ausgangsvektor des reduzierten Netzes 20, wie beispielsweise der erste Ausgangsdatensatz 31a, ist.Adapting the reduced network 20th can in a preferred embodiment by adding a further basic function 71 , hereinafter also referred to as another basic function ψ _{i + 1} , and advantageously by adding a further weight 72 , hereinafter also called further weight β _{i + 1} , as in 7 is shown. Accordingly shows 6 the reduced network 20th after the first step 11 and before the fourth step 14 and the 7 the reduced network 20th after the fourth step 14 . After repeating the steps several times 13 , 14 can the reduced network 20th , as in 8th shown several basic functions 81 and several weights 82 exhibit. The weights 82 can preferably depending on one at the inputs 64 created input data record of the first set 32 their respective values change and preferably assume different values for the same input data record. An internal logic of the reduced network 20th can be represented by the following formula: $${\overrightarrow{y}}_{J_{red}}={\displaystyle {\sum }_{i=1}^N\left[{\beta }_i\left({\overrightarrow{\mu }}_J\right)\cdot {\mathrm{\psi }}_i\right];\text{With}\overrightarrow{y_{J_{red}}}=\left(\begin{array}{l}{y}_{1_{red}}\\ y_{{\mathrm{2nd}}_{red}}\\ y_{{\mathrm{3rd}}_{red}}\end{array}\right),}$$

in which ${\beta }_i\left(\overrightarrow{\mu }\right)$

one of the several weights 82 is, the value of its argument $\overrightarrow{{\mu }_J}$

depends, ψ _i basic functions of the reduced network 20th are and $\overrightarrow{y_{J_{red}}}$

an output vector of the reduced network 20th , such as the first output data record 31a , is.

According to a variant, when repeating the steps 13 , 14 the further basic function ψ _{i + 1} , that of the reduced network 20th is added, is designed as a vector and advantageously the first output data set of the first output data sets 31 be the one whose deviation is greatest. According to the above example, the further basic function ψ _{1 can be} the first output data set 31a be when the fourth step 14 is carried out for the first time. In this case, the output vector of the reduced network 20th the values of the first output data record 31a assume if at the inputs 64 Signals, the information of the entries of the input data record 32a included.

Advantageously, the further weight β _{i + 1} that is added to the reduced network in the fourth step is adapted at least to information contained in the input data record that is contained in the second set. In a further development, the respective further basic function and preferably also the additional weight can be determined using a “reduced basic method”.

It is particularly advantageous when repeating the fourth step 14 a single input record of the first set 32 , based on the previous example of the input data record 32a , not included again in the second set.

The second set advantageously contains the input data record whose corresponding first output data record exhibits the greatest deviation. This corresponding first output data record can be in the form of the further basic function of the reduced network 20th to be added. Based on the previous example, here is the first output data record 31a meant. After the steps 13 , 14 have been repeated several times, the added further basic functions can represent a trace of the most relevant first output data sets.

The termination criterion 15 can be fulfilled, for example, if the deviation of the first output data set which corresponds to the input data set of the second set lies below a predetermined first limit. The deviation can be calculated in the form of the above-mentioned vector standard of the above-mentioned difference vector. The vector standard can be the maximum standard, for example. It is also possible that there is a maximum deviation of all first output data records 31 can be approximated using an upper bound. As a result, a computing time for executing the method can be reduced.

The proposed method enables a number of basic functions ψ _{j of} the reduced network 20th by a factor of greater than 1 less than a number of basic functions (φ _{i of} the first network 30th is after the steps 13 , 14 repeated so many times until the termination criterion 15 is satisfied. The factor can be, for example, in a range between ten and one thousand. As a result, a diversity of the reduced network can be less than the diversity of the first network. The diversity of the reduced network can preferably be approximated using a total of the added basic functions. All basic functions of the reduced network are preferred 20th linearly independent of each other. This can further increase the factor.

The proposed method enables the diversity of the first network 30th sufficiently accurate using the basic functions of the reduced network 20th to approximate. Sufficiently precise means that the termination criterion is met, in particular if a number of the input data records of the fourth set is less than a third of a number of input data records of the first set.

betrachtet werden, die das reduzierte Netz 20 in Abhängigkeit von allen möglichen Eingangsdatensätzen $\overrightarrow{{\mu }_J}$

berechnen kann. Die Mannigfaltigkeit des reduzierten Netzes 20 ist beispielhaft in 9 in einem dreidimensionalen Raum in Form einer Kurve 91 dargestellt, wobei die Kurve 91 einen weicheren Verlauf als die Kurve 51 hat. In einer weiteren Ausgestaltung kann die jeweilige Abweichung abgeschätzt werden. Dadurch kann eine Rechenzeit eingespart werden.The diversity of the reduced network can be seen as a set of all possible output vectors $\overrightarrow{y_{J_{red}}}$

be considered the reduced network 20th depending on all possible input data sets $\overrightarrow{{\mu }_J}$

can calculate. The diversity of the reduced network 20th is exemplary in 9 in a three-dimensional space in the form of a curve 91 shown with the curve 91 a softer course than the curve 51 Has. In a further embodiment, the respective deviation can be estimated. This can save computing time.

In a further embodiment, for each of the input data records of the fourth set, respective second corresponding output data records are made using the reduced network 20th calculated. The second output data sets can be used to distribute the second output data sets in an output space of the reduced network 20th , hereinafter referred to as reduced output space, for analysis of the reduced network 20th or be calculated for analysis of the fourth set. The distribution is preferably visualized around at least a first area 92 to be determined in the reduced output space in which there is a high density of second output data sets compared to an average density of the second output data sets in the reduced output space. Similarly, at least a second area 93 are recorded in which there is a low density of second output data records compared to the average density. In the second area 93 points the reduced network 20th a comparatively small and in the first area 92 a comparatively high sensitivity to changes in the inputs 64 input data records of the first set.

Based on those input records that match the second output records that are in the first area 92 lie, correspond, can be a subspace of a possible input space of the reduced network 20th are formed, which limits these input data records. New input data sets can be formed in this subspace and new output data sets corresponding to these new input data sets can be recorded, for example during test drives with the vehicle 1 . The new input data sets can be read out using further first control parameters for controlling the first vehicle component 7 , further second control parameters for controlling the second vehicle component 8th and third control parameters for controlling the third vehicle component 9 can be obtained from appropriate control units.

With the help of the new input data sets and the new output data sets, the reduced network can 20th be optimized. To do this, the step 12th again, preferably once, repeated and the steps 13 , 14 can be repeated several times. The new input data records preferably form the first set of input data records and the new output data records the comparison data records.

Furthermore, for each of the input data records of the fourth set, respective third output data records can be made using the first network 30th are calculated and a respective second deviation between the second and the corresponding third output data records are determined. Critical areas of the reduced diversity can be formed on the basis of the second deviations, in which the second deviations are higher than an average of the second deviations. For the critical areas you can do the same as for the first area 92 further new input data records and output data records are determined and the reduced network 20th can be optimized in the same way using this additional new data. To do this, the step 12th again preferably be repeated once and the steps 13 , 14 are repeated again several times, the further new input data records preferably forming the first set of input data records and the further new output data records being used as comparison data records.

Furthermore, the critical areas can be recorded as areas in which the reduced network 20th is not operated. This can ensure security when using the reduced network 20th in the control unit 2nd increase.

In the following, another application example will be considered, in which the first network 30th and the reduced network can each be used as a classifier. The input records 32 can be image records of an object that the camera 3rd detected. The values y ₁ , y ₂ , y ₃ can each be probability values that indicate the probability with which the object belongs to a first, second or third class. So the values are y ₁ , y ₂ , y ₃ assigned to the corresponding classes. The first class can include a variety of people, the second class a variety of vehicles and the third class a variety of other objects. In this example, the first barrier can be ten percent and the respective deviation can be the maximum norm of the difference vector.

With the proposed method it is possible to use the reduced network 20th in such a way that it has a similarly good hit rate when classifying the object, but the above-mentioned factor is approximately ten or higher. Depending on the corresponding values y _1red , y _2red , y _3red that the reduced network 20th when using the reduced network 20th in the control unit 2nd can output, the control unit 2nd activate the airbag. Because of the reduced reduced diversity compared to the diversity of the first network 30th can the control unit 2nd the values y _1red , y _2red , y _3red Calculate faster and activate the airbag faster depending on these values. It is also possible that the reduced network 20th noisy input data better than the first network 30th can filter and thereby better results when classifying objects using the camera 3rd are recorded.

In a further embodiment, the reduced network 20th and with at least one test input data record, a first test output data record and using the first network 30th and the test input data record generates a second test output data record. Furthermore, in this embodiment, a comparison can be made between the first test output data set and the second test output data set. Depending on a result of the comparison, at least one piece of information from the first and / or the second test output data record flows into further data processing. This embodiment preferably provides that both the reduced network 20th as well as the first network 30th in the vehicle 1 , especially in the control unit 2nd , are integrated and when the vehicle is in operation 1 be used.

Based on the above example, the test input data record can be one of the image data records and the first test output data record can be by the values y _1red , y _2red , y _3red and the second test _{output data set} by the values y ₁ , y ₂ , y ₃ be trained. A first or a second classification of the object is advantageously carried out with the aid of a decision algorithm on the basis of the first and the second test output data record. Based on the example above, the object is assigned either the first, second or third class in the first and in the second classification. The decision algorithm can provide that in each case that class is selected whose assigned probability value is the highest. The comparison can then be carried out in the form of a comparison of a result of the first classification with a result of the second classification.

In this case, the comparison between the first test output data record and the second test output data record is carried out indirectly on the basis of the comparison between the result of the first and the second classification. According to one In another variant, the result of the first classification as the first test output data record and the result of the second classification as the second test output data record can be viewed and the comparison can be carried out directly.

Depending on a result of the comparison, at least one piece of information from the first and / or the second test output data record flows into the further data processing.

If in a first case the result of the first classification is equal to the result of the second classification, ie the same class was assigned to the object in both cases, the result of the first and / or second classification can be passed on to a further component 101 to trigger one of the vehicle components 7 , 8th , 9 , in particular the security element. The other component 101 can be a control unit of one of the vehicle components 7 , 8th , 9 or an activation module of the control unit 2nd his. In this case, further data processing is carried out with the help of the additional component 101 carried out, the information of the first and / or the second test output data set preferably the respective highest value of the values y _1red , y _2red , y _3red or the values y ₁ , y ₂ , y ₃ can be.

If, in a second case, the result of the first classification is different from the result of the second classification, ie a different class has been assigned to the object, the control unit can be used 2nd preferably a first alternative and / or a second alternative can be implemented.

According to the first alternative of the second case, there can be at least one difference between at least a first pair of values y _1red , y ₁ , a second pair of values y _2red , y ₂ and / or a third pair of values y _3red , y ₃ be determined. Preferably, at least the difference of the pair of values is determined, which is at least the highest value of the values y _1red , y _2red , y _3red has, hereinafter called critical pair of values.

Depending on the result of the first classification, the critical pair of values can do the values y ₁ , y ₂ or y ₃ exhibit. If the critical pair of values has a value that is the second highest of the values y ₁ , y ₂ and y ₃ , the control unit advantageously forwards the result of the first classification to the further component 101 to trigger one of the vehicle components 7 , 8th , 9 further. In this case it is assumed that the reduced network 20th a better classification than the first network 30th has made. In this case, further data processing is carried out with the help of the additional component 101 carried out, the information of the first test output data set being the respective highest value of the values y _1red , y _2red , y _3red is.

According to the second alternative of the second case, an emergency program can be triggered, in particular if the critical pair of values has a value that is the lowest of the values y ₁ , y ₂ and y ₃ is. In this case, the result of the first and the second classification differ greatly. The emergency program can provide that the vehicle 1 is braked or a warning sign in a cockpit of the vehicle 1 is activated. In this case, further data processing can take place in the control unit 2nd take place and trigger the emergency program. In this case, information from the first and second test output data records is included in the further data processing because the different results of the classifications trigger the emergency program.

QUOTES INCLUDE IN THE DESCRIPTION

This list of documents listed by the applicant has been generated automatically and is only included for the better information of the reader. The list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions.

Patent literature cited

• DE 102018103113 [0002]

Claims (10)

Method for generating a reduced neural network (20) for a control unit (2) of a vehicle (1) and / or a vehicle component (7, 8, 9) with the following steps: - initializing the reduced neural network (20) in a first step (11), - Calculation of respective first corresponding output data sets (31a, 31b, 31z) with the aid of a first neural network (30) and depending on a respective input data set (32a, 32b, 32z), an entirety of the respective input data sets (32a, 32b, 32z) being one forms the first set (32) of the input data sets in a second step (12) - Determining respective deviations (42a, 42b, 42z) of the first output data sets (31a, 31b, 31z) from respective comparison data sets (41a, 41b, 41z) in a third step (13), the respective first output data set (31a, 31b, 31z) corresponds to the respective comparison data record (41a, 41b, 41z), - Adapting the reduced neural network (20) as a function of a second set of input data sets, which is a subset of the first set, the first output data sets corresponding to the input data sets of the second set having, on average, a greater deviation than those of the other input data sets of the first Set of corresponding first output data sets, the remaining input data sets forming a third set of input data sets and the reduced neural network (20) being adapted independently of the third set of input data sets, in a fourth step (16), - Repeat steps three and four until an abort criterion is met. Procedure according to Claim 1 The respective comparison data record (41a, 41b, 41z) is calculated with the aid of the reduced neural network (20) and depending on the input data record (32a, 32b, 32z) with which the first corresponding output data record is calculated. Procedure according to Claim 1 or 2nd , whereby the respective deviation is estimated using an upper bound. Method according to one of the preceding claims, wherein the reduced neural network (20) has basic functions (81) and the reduced neural network (20) is adapted such that in the fourth step at least one further basic function (71) corresponds to the reduced neural network (20 ) is added and by adding the further basic function (71) one dimension of a manifold of the reduced neural network (20) is increased. Method according to one of the preceding claims, wherein all the input data records contained in the second set when the fourth step is repeated a number of times are collected in the form of a fourth set of input data records and respective second corresponding output data records corresponding to the respective input data records of the fourth set using the reduced neural network (20 ) are calculated and a distribution of the second output data sets in an output space of the reduced neural network (20) for analysis of the reduced neural network or for analysis of the fourth set is calculated on the basis of the second output data sets. Method according to one of the preceding claims, with the aid of the reduced neural network and with at least one test input data record, a first test output data record and with the aid of the first neural network and the test input data record a second test output data record are generated and a comparison between the first test output data record and the second test output data record is carried out and in Depending on a result of the comparison, at least one piece of information from the first and / or the second test output data record is included in further data processing. Reduced neural network (20), established by a method according to one of the Claims 1 to 6 . Control device (2) for a vehicle (1) and / or a vehicle component (7, 8, 9), set up by a method and / or for carrying out a method according to one of the Claims 1 to 6 . Vehicle (1) comprising a control device (2) according to Claim 8 . A computer program product (10) comprising a program that, when executed by a computer, causes the computer to perform a method according to one of the Claims 1 to 6 perform.

Images