Data processing method, discrete representation determination method, and signal generation method

Abstract

Disclosed in one or more embodiments of the present disclosure are a data processing method, a discrete representation determination method, and a signal generation method. The data processing method comprises: acquiring abstract representation data of a first timing signal; searching at least two discrete data spaces for a first discrete data combination corresponding to the abstract representation data; on the basis of the abstract representation data and the first discrete data combination, updating the at least two discrete data spaces; searching the at least two updated discrete data spaces for a second discrete data combination corresponding to the abstract representation data, such that the second discrete data combination can represent a global optimal solution which is determined by using the at least two updated discrete data spaces; and on the basis of the second discrete data combination, determining discrete representation data of the first timing signal.

Description

A data processing method, a discrete representation determination method, and a signal generation method

[0001] CROSS-REFERENCE TO RELATED APPLICATIONS

[0002] This application claims priority to the Chinese patent application filed on September 28, 2023, with application number 202311279768.2 and invention name “A data processing method, discrete representation determination method, and signal generation method”. The entire contents of that application are incorporated by reference into this application. Technical Field

[0003] Embodiments of the present disclosure relate to the field of data processing technology, and in particular, to a data processing method, a discrete representation determination method, a signal generation method, an apparatus, an electronic device, and a computer-readable medium. Background Art

[0004] In some application scenarios, in order to meet certain requirements (such as data transmission speed requirements, data storage space requirements, etc.), a discrete representation of a timing signal can be performed. This makes how to discretely represent a timing signal a technical problem that needs to be solved urgently.

[0005] Summary of the Invention

[0006] One or more embodiments of the present disclosure provide a data processing method, a discrete representation determination method, a signal generation method, an apparatus, an electronic device, and a computer-readable medium.

[0007] In order to achieve the above objectives, the technical solutions provided by the present disclosure are as follows:

[0008] An embodiment of the present disclosure provides a data processing method, the method comprising:

[0009] Acquiring abstract representation data of the first time series signal;

[0010] Searching for a first discrete data combination corresponding to the abstract representation data from at least two discrete data spaces;

[0011] updating the at least two discrete data spaces according to the combination of the abstract representation data and the first discrete data;

[0012] searching for a second discrete data combination corresponding to the abstract representation data from the at least two updated discrete data spaces; the at least two updated discrete data spaces are used to provide a plurality of candidate discrete data combinations, the plurality of candidate discrete data combinations including the second discrete data combination, wherein the gap representation data between the discrete representation data determined using the second discrete data combination and the abstract representation data is smaller than the gap representation data between the discrete representation data and the abstract representation data determined using any other candidate discrete data combination among the plurality of candidate discrete data combinations except the second discrete data combination;

[0013] Discrete representation data of the first time series signal is determined according to the second discrete data combination.

[0014] In a possible implementation manner, the second discrete data combination is determined using a light column search algorithm.

[0015] In a possible implementation manner, the number of the first timing signals is K, where K is a positive integer;

[0016] The method further comprises:

[0017] Determine a target search algorithm corresponding to the k-th first time series signal from at least one candidate search algorithm, where k is a positive integer and k≤K;

[0018] The step of searching for a second discrete data combination corresponding to the abstract representation data from the updated at least two discrete data spaces includes:

[0019] According to the target search algorithm corresponding to the kth first time series signal, a second discrete data combination corresponding to the abstract representation data of the kth first time series signal is searched from the at least two updated discrete data spaces, where k is a positive integer and k≤K.

[0020] In one possible implementation manner, determining a target search algorithm corresponding to the k-th first time series signal from at least one candidate search algorithm includes:

[0021] According to the preset selection probability corresponding to each candidate search algorithm, a target search algorithm corresponding to the k-th first time series signal is determined from at least one candidate search algorithm.

[0022] In one possible implementation, the at least one candidate search algorithm includes a light column search algorithm and a greedy algorithm.

[0023] In a possible implementation manner, the first timing signal is an action sequence signal.

[0024] In one possible implementation, the at least two discrete data spaces include a first discrete data space and a second discrete data space, and there is a difference between at least one candidate discrete data provided by the first discrete data space and at least one candidate discrete data provided by the second discrete data space.

[0025] In a possible implementation manner, the at least one candidate discrete data provided by each of the at least two discrete data spaces is the same.

[0026] In one possible implementation, the abstract representation data includes at least one eigenvector; the first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each of the eigenvectors; the second discrete data combination corresponding to the abstract representation data includes the second discrete data combination corresponding to each of the eigenvectors; for any eigenvector, the gap representation data between the discrete representation data determined by using the second discrete data combination corresponding to the eigenvector and the eigenvector is smaller than the gap representation data between the discrete representation data determined by using any other candidate discrete data combination among the several candidate discrete data combinations except the second discrete data combination corresponding to the eigenvector and the eigenvector.

[0027] In one possible implementation, the data processing model includes an encoding module, a quantization module, and a decoding module; the abstract representation data of the first time series signal is obtained by the encoding module encoding the first time series signal; the discrete representation data of the first time series signal is obtained by the quantization module performing discrete representation processing on the abstract representation data; the decoding module is used to determine the signal reconstruction result corresponding to the first time series signal based on the discrete representation data of the first time series signal.

[0028] In a possible implementation manner, before acquiring the abstract representation data of the first time series signal, the method further includes:

[0029] Acquire the first time series signal from a sample data set;

[0030] After determining the signal reconstruction result corresponding to the first time series signal based on the discrete representation data of the first time series signal, the method further includes:

[0031] The data processing model is updated according to the first time series signal, the signal reconstruction result, the abstract representation data, and the discrete representation data, and the step of obtaining the first time series signal from the sample data set is returned to continue until a preset stop condition is reached.

[0032] An embodiment of the present disclosure provides a method for determining a discrete representation, the method comprising:

[0033] obtaining a second timing signal;

[0034] The discrete representation data of the second time series signal is determined using the encoding module and the quantization module in the pre-built data processing model; the data processing model is determined using the data processing method provided by the embodiment of the present disclosure.

[0035] An embodiment of the present disclosure provides a signal generation method, the method comprising:

[0036] Obtain target discrete representation data;

[0037] A decoding module in a pre-built data processing model is used to generate a time series signal corresponding to the target discrete representation data; the data processing model is determined using the data processing method provided in the embodiment of the present disclosure.

[0038] In a possible implementation manner, the target discrete representation data is determined based on the discrete representation data to be used stored in a preset storage space.

[0039] In a possible implementation manner, the target discrete representation data is the discrete representation data to be used.

[0040] In a possible implementation manner, the target discrete representation data is determined by modifying the discrete representation data to be used.

[0041] An embodiment of the present disclosure provides a data processing device, including:

[0042] A first acquiring unit, configured to acquire abstract representation data of a first time series signal;

[0043] A first searching unit, configured to search for a first discrete data combination corresponding to the abstract representation data from at least two discrete data spaces;

[0044] a first updating unit, configured to update the at least two discrete data spaces according to the combination of the abstract representation data and the first discrete data;

[0045] a second search unit, configured to search for a second discrete data combination corresponding to the abstract representation data from the at least two updated discrete data spaces; the at least two updated discrete data spaces are used to provide a plurality of candidate discrete data combinations, the plurality of candidate discrete data combinations including the second discrete data combination, wherein a gap representation data between the discrete representation data determined using the second discrete data combination and the abstract representation data is smaller than a gap representation data between the discrete representation data and the abstract representation data determined using any other candidate discrete data combination among the plurality of candidate discrete data combinations except the second discrete data combination;

[0046] The first determining unit is configured to determine discrete representation data of the first time series signal according to the second discrete data combination.

[0047] An embodiment of the present disclosure provides a discrete representation determination device, comprising:

[0048] A second acquiring unit, configured to acquire a second timing signal;

[0049] The second determination unit is used to determine the discrete representation data of the second time series signal using the encoding module and the quantization module in the pre-built data processing model; the data processing model is determined using the data processing method provided by the embodiment of the present disclosure.

[0050] An embodiment of the present disclosure provides a signal generating device, comprising:

[0051] A third acquisition unit is used to acquire target discrete representation data;

[0052] A signal generating unit is used to generate a timing signal corresponding to the target discrete representation data using a decoding module in a pre-built data processing model; the data processing model is determined using the data processing method provided in an embodiment of the present disclosure.

[0053] An embodiment of the present disclosure provides an electronic device, the device comprising: a processor and a memory;

[0054] The memory is used to store instructions or computer programs;

[0055] The processor is used to execute the instructions or computer programs in the memory so that the electronic device performs the data processing method provided by the embodiments of the present disclosure, or performs the discrete representation determination method provided by the embodiments of the present disclosure, or performs the signal generation method provided by the embodiments of the present disclosure.

[0056] An embodiment of the present disclosure provides a computer-readable medium, which stores instructions or computer programs. When the instructions or computer programs are executed on a device, the device executes the data processing method provided by the embodiment of the present disclosure, or executes the discrete representation determination method provided by the embodiment of the present disclosure, or executes the signal generation method provided by the embodiment of the present disclosure.

[0057] An embodiment of the present disclosure provides a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program contains program code for executing the data processing method provided by the embodiment of the present disclosure, or program code for executing the discrete representation determination method provided by the embodiment of the present disclosure, or program code for executing the signal generation method provided by the embodiment of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0058] In order to more clearly illustrate the technical solutions in the embodiments of the present disclosure or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments recorded in the present disclosure. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0059] FIG1 is a flow chart of a data processing method provided by one embodiment of the present disclosure;

[0060] FIG2 is a schematic diagram of a working principle of a model provided by an embodiment of the present disclosure;

[0061] FIG3 is a schematic diagram of a model structure provided by an embodiment of the present disclosure;

[0062] FIG4 is a flowchart of a discrete representation determination method provided by one embodiment of the present disclosure;

[0063] FIG5 is a flow chart of a signal generating method provided by an embodiment of the present disclosure;

[0064] FIG6 is a schematic structural diagram of a data processing device provided by an embodiment of the present disclosure;

[0065] FIG7 is a schematic structural diagram of a discrete representation determination device provided by an embodiment of the present disclosure;

[0066] FIG8 is a schematic structural diagram of a signal generating device provided by an embodiment of the present disclosure;

[0067] FIG9 is a schematic structural diagram of an electronic device provided by an embodiment of the present disclosure. DETAILED DESCRIPTION

[0068] In order to enable those skilled in the art to better understand the solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the drawings in the embodiments of the present disclosure. Obviously, the embodiments described are only part of the embodiments of the present disclosure, not all of the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present disclosure without making any creative efforts shall fall within the scope of protection of the present disclosure.

[0069] Compared with the related art, one or more embodiments of the present disclosure have at least the following advantages:

[0070] In the technical solution provided by one or more embodiments of the present disclosure, abstract representation data of a first time series signal (for example, the encoding result of an action sequence signal) is first obtained; then a first discrete data combination corresponding to the abstract representation data is searched from at least two discrete data spaces; secondly, based on the abstract representation data and the first discrete data combination, the at least two discrete data spaces are updated so that the updated at least two discrete data spaces can better provide several candidate discrete data combinations; then, a second discrete data combination corresponding to the abstract representation data is searched from the updated at least two discrete data spaces so that the gap representation data between the discrete representation data determined using the second discrete data combination and the abstract representation data is smaller than the gap representation data between the discrete representation data determined using any other candidate discrete data combination among the several candidate discrete data combinations except the second discrete data combination and the abstract representation data, so that the second discrete data combination can represent the global optimal solution determined using the updated at least two discrete data spaces; finally, based on the second discrete data combination, the discrete representation data of the first time series signal is determined, thereby achieving the purpose of discrete representation of a time series signal.

[0071] In addition, since the second discrete data combination above is the global optimal solution obtained by searching from at least two discrete data spaces above, the gap representation data between the discrete representation data determined by the second discrete data combination and the abstract representation data above is minimized, so that the discrete representation data is closest to the abstract representation data, and the information loss generated when the abstract representation data is converted into the discrete representation data is minimized, so that the discrete representation data can better represent the detailed information carried by the time series signal, thereby effectively improving the discrete representation accuracy of the time series signal.

[0072] To better understand the technical solutions provided by one or more embodiments of the present disclosure, the data processing methods provided by one or more embodiments of the present disclosure are described below with reference to some figures. As shown in Figure 1, the data processing method provided by one embodiment of the present disclosure includes the following steps S101-S105. Figure 1 is a flow chart of a data processing method provided by one embodiment of the present disclosure.

[0073] S101: Acquire abstract representation data of a first timing signal.

[0074] The first time series signal refers to a signal that requires data processing (e.g., abstract representation processing, discrete representation processing, signal reconstruction processing, etc.); and the present disclosure does not limit the number of the first time series signals. For example, in some application scenarios (e.g., using some time series signals to build a machine learning model with the data processing function), the first time series signal may refer to the sample data used in a round of training; and the number of the first time series signals may be K, so that the K first time series signals can be used as the same batch to complete this round of training. K is a positive integer.

[0075] In addition, the present disclosure does not limit the implementation of the first timing signal mentioned above. For ease of understanding, some examples are provided below for illustration.

[0076] Example 1: In some application scenarios (eg, audio signal processing scenarios), the first time series signal mentioned above may be an audio signal (eg, audio data, voice data, etc.).

[0077] Example 2: In some application scenarios (eg, image sequence signal processing scenarios), the first timing signal mentioned above may be an image sequence signal (eg, a video signal similar to video data, etc.).

[0078] Example 3: In some application scenarios (eg, text signal processing scenarios), the first time-series signal mentioned above may be a text signal (eg, a signal composed of multiple texts arranged in sequence in time).

[0079] Example 4: In some application scenarios (e.g., 3D motion processing scenarios), the first timing signal described above may be a motion sequence signal. The motion sequence signal is used to describe the motion state of an object (e.g., a person, an animal, a robot, etc.); and the motion sequence signal includes at least one frame of motion data, each frame of motion data being used to describe the posture of the object.

[0080] In addition, the present disclosure does not limit the implementation of the action sequence signal in the above paragraph. For example, in some application scenarios, the action sequence signal can be implemented using the signal M shown in the following formulas (1)-(2).

[0081] Where M represents an action sequence signal; p t Represents the t-th frame action data in the M, the p t At the tth time sequence position in the M, the p t It can be used to represent the motion posture of an object in a certain format (for example, the tree structure format presented by the "input motion sequence signal" in Figure 2), where t is a positive integer, t≤T, T represents the number of action data frames in M, and D represents the p t The dimensions of this vector data; Indicates that the p t The angular velocity of the root node in this frame of motion data; Indicates that the p t The linear velocity of the root node in this frame of motion data; J represents the p t The position of the joint point in this frame of motion data in the normalized coordinate system of the root node; j t Indicates that the p t The velocity of the joint point in this frame of motion data in the normalized coordinate system of the root node; Indicates that the size of M is T×D; Indicates that the p t The size is 1×D.

[0082] The above “abstract representation data of the first timing signal” is used to abstractly represent the first timing signal; and the present disclosure does not limit the implementation method of the “abstract representation data of the first timing signal”, for example, the “abstract representation data of the first timing signal” may include at least one eigenvector. It should be noted that the present disclosure does not limit the correlation between the number of the eigenvectors and the number of motion data in the first timing signal, for example, the two are equal. For another example, the number of motion data in the first timing signal is a preset multiple of the number of the eigenvectors (for example, 2 times, 4 times or 8 times, etc.), and the preset multiple can be set according to the actual application scenario.

[0083] Based on the above content, it can be seen that in one possible implementation, when the first timing signal above includes T motion data, the abstract representation data of the first timing signal may include T eigenvectors. In another possible implementation, when the first timing signal above includes T motion data (for example, T motion data shown in formula (1) above), in order to better improve data processing efficiency, the abstract representation data of the first timing signal may include T / 8 eigenvectors (for example, T / 8 eigenvectors shown in formula (3) below). Z={z1,z2,…,z T / 8}=Encoder(M) (3)

[0084] Where Z represents the encoding feature of the action sequence signal M; Encoder(M) represents the encoding process for the action sequence signal M; i Represents the feature vector at the i-th temporal position in the coding feature Z, where i is a positive integer, i≤T / 8.

[0085] It should be noted that the present disclosure does not limit the implementation method of the above-mentioned feature vector. For example, in some application scenarios, the feature vector may refer to a feature vector obtained by encoding processing. Based on this, it can be seen that in some application scenarios, the "abstract representation data of the first time series signal" may refer to the encoding characteristics of the first time series signal (for example, the encoding characteristics shown in formula (3) above), and the encoding characteristics include at least one feature vector. It should be noted that the present disclosure does not limit the method for determining the encoding characteristics of the first time series signal. For example, it can be implemented using any existing or future encoding method (for example, the encoding module shown in Figure 2 or Figure 3, etc.).

[0086] Based on the relevant content of "abstract representation data of the first time series signal" above, it can be known that in some application scenarios, when the first time series signal above is an action sequence signal M (for example, M shown in formula (1) above), and the action sequence signal M includes T frames of action data, the first time series signal can be input into a coding module (for example, the coding module shown in Figure 2 or Figure 3), so that the coding module performs coding processing on the first time series signal to obtain the coding feature Z corresponding to the first time series signal (for example, T / 8 frame 512-dimensional latent space feature, etc.), so that the coding feature Z can abstractly represent the first time series signal, so the coding feature Z can be regarded as the abstract representation data of the first time series signal. It can be seen that in one possible implementation, when the first time series signal is implemented using the action sequence signal M, the abstract representation data of the first time series signal can be implemented using the coding feature corresponding to the first time series signal.

[0087] It should be noted that the present disclosure does not limit the implementation method of the encoding module. For example, in some application scenarios, the encoding module may include several one-dimensional (1-Dimension, 1D) convolutions, rectified linear unit (Rectified Linear Unit, ReLU) activation layers and residual modules.

[0088] It should also be noted that the present disclosure does not limit the implementation methods of “several”. For example, “several” may refer to at least two.

[0089] Based on the relevant content of S101 above, it can be known that in some application scenarios (for example, in scenarios where a round of training process is involved in using some time series signals to build a machine learning model with the data processing function), the number of the first time series signals above can be K, and the abstract representation data of each first time series signal can be obtained (for example, with the help of the encoding module shown in Figure 2 or Figure 3, the abstract representation data of the kth first time series signal is determined, k is a positive integer, k≤K, K is a positive integer), so that the abstract representation data of each first time series signal is used to abstractly represent the corresponding first time series signal, so that certain data processing (for example, discrete representation processing, etc.) can be completed based on these abstract representation data later.

[0090] S102: Searching for a first discrete data combination corresponding to the abstract representation data from at least two discrete data spaces.

[0091] The discrete data space is used to provide at least two candidate discrete data, so that the discrete representation of a certain time series signal (eg, the first time series signal mentioned above) can be determined based on the candidate discrete data selected from these candidate discrete data.

[0092] In addition, the present disclosure is not limited to the implementation method of the above discrete data space. For example, it can be implemented using any existing or future space (such as a dictionary, etc.) that can provide some candidate discrete data.

[0093] In addition, the present disclosure does not limit the implementation of the above "at least two discrete data spaces". For example, the "at least two discrete data spaces" may include Q discrete data spaces (for example, Q learnable dictionaries, etc.), where Q is a positive integer, Q ≥ 2, and Q can be set in advance according to the actual application scenario. The qth discrete data space is used to provide at least two candidate discrete data (for example, the following formula (4) shown in These G candidate discrete data), and the present disclosure does not limit the implementation of the qth discrete data space. For example, it can be implemented using the following formula (4). q is a positive integer, q≤Q, and Q represents the number of discrete data spaces in the "at least two discrete data spaces".

[0094] Where C q represents the qth discrete data space (e.g., the qth learnable dictionary, etc.); represents the g-th candidate discrete data in the q-th discrete data space, and the present disclosure does not limit the For example, it can be implemented using a vector (e.g., a 512-dimensional vector), where g is a positive integer, 1≤g≤G, and G represents the number of candidate discrete data in the qth discrete data space.

[0095] In addition, the present disclosure does not limit the association relationship between any two discrete data spaces in the above “at least two discrete data spaces”. For ease of understanding, the following description is combined with two cases.

[0096] Case 1: In some application scenarios, in order to minimize the usage of storage space, all discrete data spaces in the above “at least two discrete data spaces” share the same set of candidate discrete data.

[0097] Based on the relevant content of Case 1 above, it can be seen that in one possible implementation method, at least one candidate discrete data provided by each discrete data space in the above "at least two discrete data spaces" is the same, so that each discrete data space in the "at least two discrete data spaces" is the same, so that when storing the "at least two discrete data spaces", it is only necessary to allocate storage space according to the storage resource requirements of one discrete data space, which is conducive to minimizing storage resource consumption as much as possible.

[0098] Based on the content of the previous paragraph, it can be seen that in some application scenarios, when the above "at least two discrete data spaces" include Q discrete data spaces, at least one candidate discrete data provided by the qth discrete data space is the same as at least one candidate discrete data provided by the q+1th discrete data space, q is a positive integer, q+1≤Q, so that multiple discrete data spaces can share the same set of candidate discrete data, which is beneficial to reduce storage consumption.

[0099] Case 2: In some application scenarios, in order to maximize the discrete representation accuracy of the timing signal, there may be at least two types of discrete data spaces in the above "at least two discrete data spaces", and different types of discrete data spaces are used to provide different candidate discrete data.

[0100] Based on the relevant content of the above situation 2, it can be seen that in one possible implementation, when the above "at least two discrete data spaces" include a first discrete data space and a second discrete data space, there is a difference between at least one candidate discrete data provided by the first discrete data space and at least one candidate discrete data provided by the second discrete data space, so that there are at least two types of discrete data spaces in the "at least two discrete data spaces". This is conducive to increasing the diversity of candidate discrete data, and thus conducive to improving the accuracy of the discrete representation determined based on the "at least two discrete data spaces". The first discrete data space is used to represent one type of discrete data space in the "at least two discrete data spaces", and the second discrete data space is used to represent another type of discrete data space in the "at least two discrete data spaces".

[0101] Based on the content of the previous paragraph, it can be seen that in some application scenarios, when the above "at least two discrete data spaces" include Q discrete data spaces, at least one candidate discrete data provided by the qth discrete data space is different from at least one candidate discrete data provided by any other discrete data space in the Q discrete data spaces except the qth discrete data space, q is a positive integer, q≤Q, so that different discrete data spaces in the "at least two discrete data spaces" can be used to provide different candidate discrete data, which is beneficial to improve the diversity of candidate discrete data, and further beneficial to improve the accuracy of the discrete representation determined based on the "at least two discrete data spaces".

[0102] The "first discrete data combination" in S102 above is used to represent the results obtained by searching for the abstract representation data of the first timing signal in the at least two discrete data spaces above; and the "first discrete data combination" may include the search results corresponding to each discrete data space in the at least two discrete data spaces.

[0103] It can be seen that in one possible implementation, when the above "at least two discrete data spaces" include Q discrete data spaces, the above first discrete data combination may include the search result corresponding to the first discrete data space, the search result corresponding to the second discrete data space, ... (and so on), and the search result corresponding to the Qth discrete data space. The search result corresponding to the qth discrete data space refers to a candidate discrete data selected from the qth discrete data space when determining the abstract representation of the first time series signal based on the above at least two discrete data spaces. q is a positive integer, q≤Q, and Q is a positive integer.

[0104] In addition, the present disclosure does not limit the implementation of the first discrete data combination above. For example, in some application scenarios, when the abstract representation data of the first time series signal above includes at least one feature vector (for example, z1, z2, ..., z shown in formula (3) above), T / 8 When the i-th eigenvector is a eigenvector (e.g., eigenvectors), the first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each eigenvector. The i-th eigenvector refers to the eigenvector at the i-th temporal position in the abstract representation data; the first discrete data combination corresponding to the i-th eigenvector refers to the discrete data combination obtained by searching for the i-th eigenvector in the above "at least two discrete data spaces" when performing discrete representation processing on the i-th eigenvector. i is a positive integer, i≤I (for example, I=T / 8), and I represents the number of eigenvectors in the abstract representation data.

[0105] In addition, the present disclosure does not limit the method for determining the first discrete data combination. For example, in some application scenarios, the first discrete data combination can be determined by means of a greedy algorithm.

[0106] Based on the above content, it can be seen that in some application scenarios, when the abstract representation data of the first time series signal includes at least one eigenvector, the first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each eigenvector, and the "at least two discrete data spaces" above include Q discrete data spaces, the i-th eigenvector (for example, the z involved in the above formula (3)) i ) can be determined as follows: first, searching for candidate discrete data closest to the i-th eigenvector from the first discrete data space as the search result corresponding to the first discrete data space, and calculating the gap representation data between the search result corresponding to the first discrete data space and the i-th eigenvector as the gap representation data corresponding to the first discrete data space, so that the gap representation data can represent the error generated when the search result corresponding to the first discrete data space is used to determine the discrete representation of the i-th eigenvector; then searching for candidate discrete data closest to the gap representation data corresponding to the first discrete data space from the second discrete data space as the search result corresponding to the second discrete data space, and calculating the gap between the search result corresponding to the second discrete data space and the gap representation data corresponding to the first discrete data space. Characterization data, as the gap characterization data corresponding to the second discrete data space, so that the gap characterization data can represent the error generated when the search results corresponding to the 1-2 discrete data spaces are used to determine the discrete representation of the i-th eigenvector; then search for candidate discrete data closest to the gap characterization data corresponding to the second discrete data space from the third discrete data space, as the search result corresponding to the third discrete data space, and calculate the gap characterization data between the search result corresponding to the third discrete data space and the gap characterization data corresponding to the second discrete data space, as the gap characterization data corresponding to the third discrete data space, so that the gap characterization data can represent the error generated when the search results corresponding to the 1-3 discrete data spaces are used to determine the discrete representation of the i-th eigenvector; ... (and so on, until all Q discrete data spaces are traversed). It can be seen that under this search method, the optimal result (that is, the optimal result that can be determined for the discrete data space under the current circumstances) is searched out from each discrete data space in turn. Wherein, i is a positive integer, i≤I (for example, I=T / 8), and I represents the number of eigenvectors in the abstract representation data.

[0107] Based on the relevant content of S102 above, it can be known that in some application scenarios, when the number of the first time series signals above is K, for the kth first time series signal (for example, M shown in formula (1) above), after obtaining the abstract representation data of the kth first time series signal (for example, Z shown in formula (3) above), a greedy algorithm can be used to search for the first discrete data combination corresponding to the abstract representation data from Q discrete data spaces (for example, the first discrete data combination corresponding to each eigenvector in the abstract representation data), so that the discrete representation effect of the Q discrete data spaces for the kth first time series signal can be determined with the help of the first discrete data combination. Wherein, k is a positive integer, k≤K, and K is a positive integer.

[0108] S103: Update at least two discrete data spaces according to the combination of the abstract representation data and the first discrete data.

[0109] In the present disclosure, for a first time series signal, after determining the first discrete data combination corresponding to the abstract representation data of the first time series signal using the above at least two discrete data spaces, the at least two discrete data spaces can be updated based on the abstract representation data and the first discrete data combination, so that the discrete representation determined for the first time series signal using the updated at least two discrete data spaces is more accurate, thereby enabling online learning processing for the at least two discrete data spaces.

[0110] It should be noted that the present disclosure does not limit the implementation of the above S103. For example, in some application scenarios, S103 can use the online Kmean algorithm to implement online moving average update processing for at least two discrete data spaces (for example, the update processing process shown in formula (5) below).

[0111] Where z i The abstract representation of the first time series signal above is the feature vector at the i-th time series position in the data Z, where i is a positive integer, i≤T / 8; Indicates that z i The qth discrete data in the first discrete data combination corresponding to this feature vector (for example, the z i The index identifier of the candidate discrete data described by the search result corresponding to this feature vector in the qth discrete data space, so that the Used to uniquely identify the z in S102 above i This feature vector is the candidate discrete data obtained by searching in the qth discrete data space; Indicates that the qth discrete data space after update has The candidate discrete data identified by this index; Indicates that the qth discrete data space before the update has This index identifies the candidate discrete data; q is a positive integer, q≤Q; α represents the The corresponding weight, and α can be preset, for example, α = 0.99; 1-α means that z i The corresponding weight.

[0112] It should be noted that the present disclosure is not limited to the implementation, for example, the Can be used to indicate that This index identifies the position of the candidate discrete data in the qth discrete data space above.

[0113] It should also be noted that the present disclosure does not limit the implementation method of S103 above. For example, in some application scenarios, when the abstract representation data of the first timing signal above includes at least one eigenvector, and the first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each eigenvector, the S103 can specifically be: using the first eigenvector and the first discrete data combination corresponding to the first eigenvector, update at least two discrete data spaces; using the second eigenvector and the first discrete data combination corresponding to the second eigenvector, continue to update the at least two discrete data spaces; ... (and so on, until all eigenvectors are traversed), so that the at least two discrete data spaces finally updated can meet the discrete representation requirements of each eigenvector as much as possible.

[0114] For another example, in some application scenarios, when the abstract representation data of the first time series signal includes at least one eigenvector, and the first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each eigenvector, the S103 can be specifically: using the first eigenvector and the first discrete data combination corresponding to the first eigenvector, updating at least two discrete data spaces to obtain at least two updated discrete data spaces corresponding to the first eigenvector, so that the at least two updated discrete data spaces corresponding to the first eigenvector can meet the discrete representation requirements of the first eigenvector as much as possible; using the second eigenvector vector and the first discrete data combination corresponding to the second eigenvector, updating at least two discrete data spaces, and obtaining at least two updated discrete data spaces corresponding to the second eigenvector, so that the at least two updated discrete data spaces corresponding to the second eigenvector can meet the discrete representation requirements of the second eigenvector as much as possible; ... (and so on, until all eigenvectors are traversed), so that at least two updated discrete data spaces corresponding to each eigenvector can be obtained, so that the at least two updated discrete data spaces corresponding to each eigenvector can be used to complete the secondary search processing for the corresponding eigenvector respectively.

[0115] Based on the relevant content of S103 above, it can be known that in some application scenarios, for the above K first time series signals, after obtaining the abstract representation data of each first time series signal and its corresponding first discrete data combination, the above at least two discrete data spaces can be updated based on these abstract representation data and its corresponding first discrete data combination, so that the updated at least two discrete data spaces can more accurately determine the discrete representation of each first time series signal.

[0116] S104: Search for a second discrete data combination corresponding to the abstract representation data from the at least two updated discrete data spaces; the at least two updated discrete data spaces are used to provide a number of candidate discrete data combinations, the number of candidate discrete data combinations including the second discrete data combination, and the gap representation data between the discrete representation data and the abstract representation data determined using the second discrete data combination is smaller than the gap representation data between the discrete representation data and the abstract representation data determined using any other candidate discrete data combination among the number of candidate discrete data combinations except the second discrete data combination.

[0117] The second discrete data combination is used to represent the result of searching for the abstract representation data of the first time series signal in the at least two updated discrete data spaces. It should be noted that the implementation of the second discrete data combination is similar to that of the first discrete data combination described above and will not be further described here for the sake of brevity.

[0118] It can be seen that in a possible implementation, when the abstract representation data of the first time series signal above includes at least one feature vector (for example, z1, z2, ..., z shown in formula (3) above), T / 8 When the second discrete data combination corresponding to the abstract representation data includes the second discrete data combination corresponding to each eigenvector, and for any eigenvector, the difference characterization data between the discrete representation data determined by the second discrete data combination corresponding to the eigenvector and the eigenvector is less than the difference characterization data between the discrete representation data determined by any other candidate discrete data combination except the second discrete data combination corresponding to the eigenvector among the above candidate discrete data combinations, so that the second discrete data combination corresponding to the eigenvector can represent the global optimal solution determined for the eigenvector using the at least two updated discrete data spaces. Among them, the second discrete data combination corresponding to the i-th eigenvector refers to the discrete data combination obtained by searching for the i-th eigenvector from the above "at least two updated discrete data spaces" when performing discrete representation processing on the i-th eigenvector. i is a positive integer, i≤I (for example, I=T / 8), and I represents the number of eigenvectors in the abstract representation data.

[0119] In addition, for the second discrete data combination above, in order to better improve the discrete representation accuracy, the second discrete data combination can be determined with the help of a light column search algorithm so that the second discrete data combination can represent the global optimal solution obtained by searching the abstract representation data of the first timing signal from at least two discrete data spaces updated above (such as the global optimal solution shown in formula (6) below).

[0120] Where z i The eigenvector at the i-th time series position in the abstract representation data Z representing the first time series signal, where i is a positive integer and i≤T / 8; Represents the qth discrete data space (for example, at least according to z i The candidate discrete data selected from the qth discrete data space obtained by combining this feature vector and its corresponding first discrete data, etc.; Indicates the index identifier of the selected candidate discrete data.

[0121] Based on the above content, it can be seen that in some application scenarios, when the above "at least two discrete data spaces after update" are used to provide several candidate discrete data combinations, the second discrete data combination corresponding to the abstract representation data of the first time series signal above can refer to a candidate discrete data combination existing in the several candidate discrete data combinations that can determine the discrete representation closest to the abstract representation data (that is, a candidate discrete data combination that satisfies the global optimal solution condition shown in formula (6) above), so that the difference representation data between the discrete representation data determined by using the second discrete data combination and the abstract representation data is smaller than the difference representation data between the discrete representation data determined by using any other candidate discrete data combination in the several candidate discrete data combinations except the second discrete data combination and the abstract representation data, so that the second discrete data combination can represent the global optimal solution obtained by searching for the abstract representation data from the at least two discrete data spaces after update. Wherein, the candidate discrete data combination refers to a set of candidate discrete data that can be provided by the "at least two discrete data spaces after update"; and the implementation method of the candidate discrete data combination is similar to the implementation method of the first discrete data combination above. For the sake of brevity, it is not repeated here.

[0122] Based on the relevant content of S104 above, it can be known that in some application scenarios, when the number of the first time series signals above is K, for the kth first time series signal, a light column search algorithm can be used to search for a second discrete data combination corresponding to the abstract representation data of the kth first time series signal from the at least two discrete data spaces updated above, so that the discrete representation determined using the second discrete data combination can retain as much detailed information carried by the kth first time series signal as possible, so that the discrete representation data of the kth first time series signal can be subsequently determined based on the second discrete data combination. Wherein, k is a positive integer, k≤K, and K is a positive integer.

[0123] S105: Determine discrete representation data of the first time series signal according to the second discrete data combination.

[0124] The discrete representation data of the first time series signal refers to the discrete representation of the first time series signal; and the present disclosure does not limit the determination process of the discrete representation data of the first time series signal. For example, it can be implemented using the following formula (7).

[0125] Where, The abstract representation data Z representing the first time series signal above corresponds to the discrete representation of the feature vector at the i-th time series position; Represents the qth discrete data space (for example, at least according to z iThe candidate discrete data selected from the qth discrete data space obtained by combining this feature vector and its corresponding first discrete data, etc.; Indicates the index identifier of the selected candidate discrete data.

[0126] Based on the relevant contents of S101 to S105 above, it can be seen that for the data processing method provided by the embodiment of the present disclosure, the abstract representation data of the first time series signal (for example, the encoding result of an action sequence signal) is first obtained; then, a first discrete data combination corresponding to the abstract representation data is searched from at least two discrete data spaces; secondly, based on the abstract representation data and the first discrete data combination, the at least two discrete data spaces are updated so that the updated at least two discrete data spaces can better provide a number of candidate discrete data combinations; then, a second discrete data combination corresponding to the abstract representation data is searched from the updated at least two discrete data spaces. According to the present invention, a method for obtaining a discrete representation data of a first time series signal and a second time series signal is adopted to obtain a discrete representation data of the first time series signal. The method comprises the steps of: determining a discrete representation data of the first time series signal and a discrete data space based on the second discrete data combination; and determining a discrete representation data of the first time series signal based on the second discrete data combination. The method comprises the steps of: determining a discrete representation data of the first time series signal and a discrete data space based on the second discrete data combination; and determining a discrete representation data of the first time series signal based on the second discrete data combination. The method comprises the steps of: determining a discrete representation data of the first time series signal and a discrete data space based on the second discrete data combination; and determining a discrete representation data of the first time series signal based on the second discrete data combination. The method comprises the steps of:

[0127] In addition, since the second discrete data combination above is the global optimal solution obtained by searching from at least two discrete data spaces above, the gap representation data between the discrete representation data determined by the second discrete data combination and the abstract representation data above is minimized, so that the discrete representation data is closest to the abstract representation data, and the information loss generated when the abstract representation data is converted into the discrete representation data is minimized, so that the discrete representation data can better represent the detailed information carried by the time series signal, thereby effectively improving the discrete representation accuracy of the time series signal.

[0128] In addition, the embodiments of the present disclosure do not limit the execution subject of the data processing method provided by the embodiments of the present disclosure. For example, the data processing method provided by the embodiments of the present disclosure can be applied to a terminal device or a server. For another example, the data processing method provided by the embodiments of the present disclosure can also be implemented with the help of the data interaction process between the terminal device and the server. Among them, the terminal device can be a smart phone, a computer, a personal digital assistant (PDA), a tablet computer, etc. The server can be a stand-alone server, a cluster server, or a cloud server.

[0129] In fact, in some application scenarios (for example, using the K first time series signals mentioned above to complete a model training, etc.), in order to better improve the discrete representation effect, when using at least two discrete data spaces updated above to determine the discrete representation of the K first time series signals, a search algorithm (for example, a light column search algorithm) can be used to complete the discrete representation of a part of the first time series signals, and another search algorithm (for example, a greedy algorithm) can be used to complete the discrete representation of another part of the first time series signals.

[0130] Based on the above content, the present disclosure also provides a possible implementation of the above S104. In this implementation, when the number of the above first timing signals is K, and K is a positive integer, the S104 may specifically include the following steps 11 and 12.

[0131] Step 11: Determine a target search algorithm corresponding to the k-th first time series signal from at least one candidate search algorithm, where k is a positive integer and k≤K.

[0132] The candidate search algorithm refers to a search algorithm that can be selected when determining the discrete representation of a first time series signal using the at least two discrete data spaces updated above.

[0133] Furthermore, the present disclosure does not limit the implementation of the above "at least one candidate search algorithm." For example, in some application scenarios, the "at least one candidate search algorithm" may include at least a light column search algorithm. In other application scenarios, the "at least one candidate search algorithm" may include at least a light column search algorithm and a greedy algorithm.

[0134] The target search algorithm corresponding to the kth first time series signal refers to the search algorithm required to determine the discrete representations of the K first time series signals using the at least two discrete data spaces updated above. It can be seen that in some application scenarios, the target search algorithm corresponding to the kth first time series signal can be a beam search algorithm or a greedy algorithm. Here, k is a positive integer, k≤K.

[0135] In addition, the present disclosure does not limit the implementation method of the above step 11. For example, it can be specifically as follows: according to the preset selection probability corresponding to each candidate search algorithm, determine the target search algorithm corresponding to the k-th first time series signal from at least one candidate search algorithm. The preset selection probability corresponding to the d-th candidate search algorithm is used to represent the possibility of selecting the d-th candidate search algorithm when determining the discrete representation of a first time series signal using the at least two discrete data spaces updated above, d is a positive integer, d≤D, D is a positive integer, and D represents the number of candidate search algorithms in the "at least one candidate search algorithm".

[0136] It can be seen that in some application scenarios, when at least one candidate search algorithm includes a beam search algorithm and a greedy algorithm, if the preset selection probability corresponding to the beam search algorithm is 0.5 and the preset selection probability corresponding to the greedy algorithm is 0.5, then the process of determining the target search algorithm corresponding to the k-th first time series signal may specifically be: according to the preset selection probability corresponding to the beam search algorithm and the preset selection probability corresponding to the greedy algorithm, determine the target search algorithm corresponding to the k-th first time series signal from the beam search algorithm and the greedy algorithm, so that the target search algorithm is the beam search algorithm with a probability of 0.5 and the greedy algorithm with a probability of 0.5. Wherein, k is a positive integer, k≤K.

[0137] Based on the relevant content of step 11 above, it can be known that for the above K first timing signals, the target search algorithm corresponding to each first timing signal can be determined from these candidate search algorithms according to the preset selection probabilities corresponding to some candidate search algorithms, so that the selection distribution of these target search algorithms is similar to the distribution described by the preset selection probabilities corresponding to these candidate search algorithms.

[0138] It should be noted that the present disclosure does not limit the execution time of the above step 11. For example, the execution time of the above step 11 can be earlier than the execution time of the following step 12.

[0139] Step 12: According to the target search algorithm corresponding to the kth first time series signal above, search for a second discrete data combination corresponding to the abstract representation data of the kth first time series signal from at least two discrete data spaces updated above, where k is a positive integer, k≤K.

[0140] In the present disclosure, for the kth first time series signal described above, after obtaining the target search algorithm corresponding to the kth first time series signal, the target search algorithm corresponding to the kth first time series signal can be used to search for a second discrete data combination corresponding to the abstract representation data of the kth first time series signal from the at least two updated discrete data spaces described above, so that the second discrete data combination meets the search result described by the target search algorithm. Here, k is a positive integer, k≤K.

[0141] Based on the relevant content of steps 11 to 12 above, it can be seen that in some application scenarios, for the K first time series signals above, different search algorithms can be used to search for the second discrete data combination corresponding to the abstract representation data of each first time series signal from at least two discrete data spaces updated above. This is conducive to improving the diversity of the second discrete data combination, thereby helping to improve the discrete representation effect.

[0142] In fact, in some application scenarios (e.g., model building), the discrete representation determination process described by the data processing method above can be used to train a model with discrete representation determination capabilities. Based on this, the present disclosure also provides a model training process, which can specifically include all or part of the steps in steps 21-23 below.

[0143] Step 21: Obtain a first time series signal from the sample data set.

[0144] The sample data set refers to the data required for use during model training (e.g., the "input motion sequence signal" shown in FIG2 ). The present disclosure is not limited to this sample data set; for example, the sample data set may include multiple sample data. The sample data refers to the data that needs to be processed by the model during model training. The implementation of the sample data is similar to that of the first time series signal described above and will not be further described here for the sake of brevity.

[0145] Based on the above content, it can be seen that in some application scenarios (for example, training a model that can discretely represent motion sequence signals), the above sample data set may include several sample motion sequence signals so that the model training process can be completed based on these sample motion sequence signals. Among them, the sample motion sequence signals refer to motion sequence signals that can be used as samples in the model training process (for example, the "input motion sequence signal" shown in Figure 2).

[0146] It should be noted that the present disclosure does not limit the method for obtaining the sample data set in the above paragraph. For example, in some application scenarios, when the sample data set includes several sample motion sequence signals, the method for obtaining the sample data set can be specifically: after obtaining the object motion data set (for example, a human motion data set, an animal motion data set, a robot motion data set, etc.), each motion data sequence in the object motion data set can be converted into a sample motion sequence signal (for example, the "input motion sequence signal" shown in Figure 2, etc.) according to a preset data conversion method, and based on these converted sample motion sequence signals, the sample data set is determined so that the sample data set includes these sample motion sequence signals. Among them, the preset data conversion method is used to convert each frame of motion data in the motion data sequence in a certain format (for example, a parameter format, etc.) into a data signal expressed in a graph structure (for example, the signal shown by the above formulas (1)-(2)); and the preset data conversion method can be set in advance according to the actual application scenario.

[0147] Based on the above content, it can be seen that in some application scenarios, if one wants to train a model with the function of discrete representation processing for motion sequence signals, one can first obtain a large amount of object motion data so that these object motion data can cover various object motion states; then, these object motion data are converted and processed to obtain some sample motion sequence signals (for example, the signals shown by the above formulas (1)-(2)), so that the sample motion sequence signals can represent the object posture of each frame with the help of the linear velocity of the root node, the angular velocity of the root node, and the position and velocity of the key points in the normalized coordinate system of the root node.

[0148] Based on the relevant content of step 21 above, it can be known that for the current round of training process, K first time series signals can be obtained from the sample data set (for example, K sample data that have not yet participated in the model training process are randomly extracted from the sample data set as the K first time series signals), so that the K first time series signals can be subsequently used as a batch to participate in the current round of training process. Wherein, K is a positive integer.

[0149] Step 22: Determine the signal reconstruction result corresponding to the first time series signal using a data processing model. The data processing model includes an encoding module, a quantization module, and a decoding module. The encoding module is configured to encode the first time series signal to obtain abstract representation data of the first time series signal. The quantization module is configured to perform discrete representation processing on the abstract representation data of the first time series signal to obtain discrete representation data of the first time series signal. The decoding module is configured to determine the signal reconstruction result corresponding to the first time series signal based on the discrete representation data of the first time series signal.

[0150] The data processing model is used to perform data processing (eg, abstract representation processing, discrete representation processing, signal reconstruction processing, etc.) on the input data of the data processing model.

[0151] Furthermore, for the data processing model described above, the data processing model may include an encoding module, a quantization module, and a decoding module. The input data for the quantization module includes the output data of the encoding module, and the input data for the decoding module includes the output data of the quantization module. For a better understanding, these modules are described below.

[0152] The encoding module is used to perform encoding processing on the input data of the encoding module (for example, a time series signal) to obtain an abstract representation of the input data (for example, the encoding feature shown in formula (3) above); and the present disclosure does not limit the implementation method of the encoding module. For example, it can be implemented using any existing or future module with encoding function (for example, the encoding module shown in Figure 2 or Figure 3). It can be seen that in some application scenarios, when the input data of the encoding module includes the kth first time series signal, the encoding module can be used to perform encoding processing on the kth first time series signal to obtain abstract representation data of the kth first time series signal, so that the abstract representation data can abstractly represent the kth first time series signal, where k is a positive integer, k≤K, and K is a positive integer.

[0153] It should be noted that the present disclosure does not limit the implementation of the encoding module described above. For example, it may include several 1D convolutions, ReLU activation layers, and residual modules. In addition, the present disclosure does not limit the working principle of the encoding module. For example, the working principle can be implemented using any of the implementations of S101 described above.

[0154] The quantization module is used to perform discrete representation processing on the input data of the quantization module (for example, an abstract representation of a time series signal) to obtain a discrete representation of the input data; and the working principle of the quantization module is: for any latent feature in the input data (for example, the above z i ), the discrete representation of the latent feature (for example, the discrete representation data shown in formula (7) above) can be determined with the help of the objective function shown in formula (8) below. It can be seen that in one possible implementation, the working principle of the quantization module can be implemented using any of the implementations of S102-S104 above.

[0155] Where C q represents the qth learnable discrete data space, and the C q The learning process may include at least the learning process shown in S102-S103 above, so as to learn the C q (For example, the updated C shown in formula (5) above q), the objective function shown in the above formula (8) can be converted into the above formula (6), so that the multi-objective optimization problem shown in the formula (8) can be converted into a combinatorial optimization problem. Therefore, in order to better improve the discrete representation accuracy, the combinatorial optimization problem shown in the formula (6) can be solved with the help of a global optimal solution determination algorithm to obtain a discrete representation of a signal; and the present disclosure does not limit the implementation method of the global optimal solution determination algorithm. For example, in order to better improve the training speed, a standard light column search algorithm can be used to solve the combinatorial optimization problem shown in the formula (6) to obtain a discrete representation of a signal.

[0156] The decoding module is used to perform signal reconstruction processing on the input data of the decoding module (for example, a discrete representation of a time series signal) to obtain a corresponding reconstructed signal. Moreover, the present disclosure does not limit the implementation method of the decoding module. For example, the decoding module can be implemented using any existing or future module with decoding function (for example, the decoding module shown in Figure 2 or Figure 3). For another example, the decoding module can be implemented using the following formulas (9)-(10).

[0157] Where Z quant discrete representation data representing a first time series signal; The discrete representation corresponding to the eigenvector at the i-th time sequence position in the abstract representation data Z of the first time sequence signal, that is, the discrete representation data Z of the first time sequence signal quant The discrete representation of the i-th time sequence position in the Decode(Z quant ) represents the discrete representation data Z for the first timing signal quant Performing decoding processing to obtain a signal reconstruction result corresponding to the first time series signal; Indicates the signal reconstruction result corresponding to the first timing signal.

[0158] Based on the relevant content of step 22 above, it can be known that in some application scenarios, for the current round, after obtaining K first time series signals, the data processing model can be used to perform data processing on each first time series signal to obtain the data processing results corresponding to each first time series signal (for example, abstract representation data, discrete representation data, and signal reconstruction results, etc.), so that the data processing performance of the data processing model can be determined based on these first time series signals and their corresponding data processing results.

[0159] Step 23: Update the data processing model based on the first timing signal, the signal reconstruction result corresponding to the first timing signal, the abstract representation data of the first timing signal, and the discrete representation data of the first timing signal, and return to continue executing the above step 21 and its subsequent steps until the preset stop condition is reached.

[0160] Among them, the preset stopping condition refers to the condition reached at the end of the multi-round training process of the above data processing model; and the present disclosure does not limit the preset stopping condition. For example, it can be set according to the actual application scenario. For another example, in some application scenarios, the preset stopping condition may at least include: the loss characterization data of the data processing model is lower than the preset loss threshold. For another example, in some application scenarios, the preset stopping condition may at least include: the change rate of the loss characterization data of the data processing model is lower than the preset change rate threshold. For another example, in some application scenarios, the preset stopping condition may at least include: the number of updates of the data processing model reaches a preset number threshold.

[0161] It can be seen that in one possible implementation, the above preset stop condition may be: the loss characterization data of the data processing model is lower than the preset loss threshold, or the loss characterization data of the data processing model is lower than the preset loss threshold, or the number of updates of the data processing model reaches the preset number threshold.

[0162] The above “loss characterization data of the data processing model” is used to characterize the performance of the data processing model; and the present disclosure does not limit the determination process of the “loss characterization data of the data processing model”, for example, it may specifically include the following steps 231-233.

[0163] Step 231: Determine a geometric reconstruction loss according to the first time series signal and a signal reconstruction result corresponding to the first time series signal.

[0164] Among them, the geometric reconstruction loss is used to measure the degree of consistency between the first timing signal above (for example, the "input motion sequence signal" shown in Figure 2) and the timing signal reconstructed by the data processing model for the first timing signal (for example, the "reconstructed motion sequence signal" shown in Figure 2), so that the geometric reconstruction loss can represent the signal reconstruction performance of the above data processing model.

[0165] In addition, the present disclosure does not limit the determination process of the above geometric reconstruction loss. For example, it can be implemented with the help of the following formula (11).

[0166] Where, represents the geometric reconstruction loss; represents the signal reconstruction result corresponding to the kth first time series signal; M krepresents the kth first time series signal; ‖·‖1 represents the L1 loss function.

[0167] Step 232: Determine coding consistency loss according to the abstract representation data of the first time series signal and the discrete representation data of the first time series signal.

[0168] Among them, the coding consistency loss is used to measure the degree of consistency between the abstract representation data of the first time series signal above and the discrete representation data of the first time series signal, so that the coding consistency loss can represent the discrete representation performance of the above data processing model.

[0169] In addition, the present disclosure does not limit the determination process of the above coding consistency loss. For example, it can be implemented with the help of the following formula (12).

[0170] Where, represents the coding consistency loss; Z represents the abstract representation data of the first time series signal; Z quant represents the discrete representation data of the first time series signal; ‖·‖1 represents the L1 loss function; β represents the pre-set weight; sg(·) represents the gradient stop operator, indicating that the gradient is not returned here.

[0171] In addition, the present disclosure does not limit the correlation between the execution time of step 232 and the execution time of step 231. For example, the execution time of step 232 may be almost the same as the execution time of step 231. For another example, the execution time of step 232 may be earlier than the execution time of step 231. For another example, the execution time of step 231 may be earlier than the execution time of step 232.

[0172] Step 233: Determine the loss representation data of the above data processing model based on the above geometric reconstruction loss and the above encoding consistency loss.

[0173] It should be noted that the present disclosure does not limit the implementation method of the above step 233. For example, the step 233 can specifically be: directly combining the above geometric reconstruction loss and the above coding consistency loss to obtain loss characterization data of the data processing model, so that the loss characterization data includes the geometric reconstruction loss and the coding consistency loss.

[0174] For example, the above step 233 may specifically be: adding the above geometric reconstruction loss and the above coding consistency loss to obtain loss representation data of the data processing model, so that the loss representation data includes the sum of the geometric reconstruction loss and the coding consistency loss.

[0175] For example, the above step 233 can be specifically as follows: first add the above geometric reconstruction loss and the above coding consistency loss to obtain the loss sum; then combine the loss sum, the geometric reconstruction loss and the coding consistency loss to obtain the loss representation data of the data processing model, so that the loss representation data includes the loss sum, the geometric reconstruction loss and the coding consistency loss.

[0176] Based on the relevant contents of steps 231 to 233 above, it can be known that in some application scenarios, the loss characterization data of the above data processing model is determined based on the above geometric reconstruction loss and the above coding consistency loss, so that the loss characterization data can represent the performance of the data processing model in terms of signal reconstruction and discrete representation, thereby enabling the loss characterization data to better represent the performance of the data processing model.

[0177] In addition, the present disclosure does not limit the implementation method of the model updating process in the above step 23. For example, it may specifically include the following steps 234 and 235.

[0178] Step 234 : Determine loss representation data of a data processing model according to the first time series signal, a signal reconstruction result corresponding to the first time series signal, abstract representation data of the first time series signal, and discrete representation data of the first time series signal.

[0179] It should be noted that the relevant content of the "loss characterization data of the data processing model" in the above step 234 can be found above. For the sake of brevity, it will not be repeated here.

[0180] Step 235: Update the data processing model based on the loss characterization data of the above data processing model.

[0181] In the present disclosure, after obtaining the loss characterization data of the above data processing model, the loss characterization data can be used to update the data processing model (for example, the geometric reconstruction loss in the loss characterization data is used to update the parameters involved in the encoding module and the decoding module in the data processing model, and the encoding consistency loss in the loss characterization data is used to update the parameters involved in the quantization model in the data processing model, etc.), so that the updated data processing model has better data processing performance (for example, signal reconstruction performance, discrete representation performance, etc.), so that it can subsequently return to continue executing the above step 21 and its subsequent steps based on the updated data processing model, so as to start the next round of training process.

[0182] Based on the relevant content of steps 234 to 235 above, it can be seen that in some application scenarios, for the current round, after obtaining the first time series signal, the signal reconstruction result corresponding to the first time series signal, the abstract representation data of the first time series signal, and the discrete representation data of the first time series signal, the loss of the data processing model can be calculated based on these four types of data, so that the parameters in the data processing model can be updated based on the loss to make the updated model have better performance.

[0183] Based on the relevant content of steps 21 to 23 above, it can be seen that in some application scenarios, for the data processing model, the data processing model can use multiple discrete data spaces (for example, multi-layer shared dictionaries, etc.) and global optimal solution search algorithms (for example, light column search algorithms, etc.) for discrete representation processing. This can effectively reduce the loss of details after the hidden layer features pass through the quantization model during the encoding and reconstruction process, thereby effectively improving the discrete representation accuracy of the time series signal, and thus retaining more detail information carried by the time series signal, which is conducive to better reconstruction of these details, and thus is conducive to increasing the compression degree of the input time series signal of the data processing model on the premise of ensuring that the accuracy requirements are met (for example, a 9.8-second time series signal can be directly compressed to 1.22 seconds, etc.), thereby better reducing the model calculation complexity.

[0184] In fact, in some application scenarios (for example, scenarios with discrete representation processing requirements), for a trained data processing model, the data processing model can be used to perform the task of performing discrete representation processing on any time series signal. Based on this, the present disclosure also provides a discrete representation determination method, as shown in Figure 4, which may include the following S401-S402. Figure 4 is a flowchart of a discrete representation determination method provided in an embodiment of the present disclosure.

[0185] S401: Acquire a second timing signal.

[0186] Among them, the second timing signal refers to a signal that needs to be discretely represented; and the present disclosure does not limit the second timing signal. For example, the implementation of the second timing signal is similar to the implementation of the first timing signal above. For the sake of brevity, it will not be repeated here.

[0187] In addition, the present disclosure does not limit the method for obtaining the second timing signal mentioned above. For example, it can be a timing signal received from other devices, a timing signal uploaded by a user, or a timing signal involved in other tasks.

[0188] S402: Determine discrete representation data of the second time series signal using an encoding module and a quantization module in a pre-built data processing model; the data processing model is determined using any implementation of the data processing method provided in the present disclosure.

[0189] Among them, the discrete representation data of the second timing signal refers to the discrete representation processing result of the second timing signal; and the present disclosure does not limit the discrete representation data of the second timing signal. For example, the implementation method of the discrete representation data of the second timing signal is similar to the implementation method of the discrete representation data of the first timing signal above. For the sake of brevity, it will not be repeated here.

[0190] In addition, the present disclosure does not limit the process of determining the discrete representation data of the second time series signal mentioned above. For example, it can be specifically as follows: first, the encoding module in the pre-constructed data processing model performs encoding processing on the second time series signal to obtain the abstract representation data of the second time series signal, so that the abstract representation data can represent the first time series signal; then, the quantization module in the data processing model performs discrete representation processing on the abstract representation data of the second time series signal to obtain the discrete representation data of the second time series signal.

[0191] It should be noted that the present disclosure does not limit the working principle used by the quantization module in the above paragraph. For example, in some application scenarios, in order to improve the efficiency of discrete representation, the quantization module can implement discrete representation processing with the help of a global optimal solution search process (for example, the global optimal solution search process shown in S104 above). For another example, in other application scenarios, the quantization module can implement discrete representation processing using the discrete representation process shown in S102-S105 above. In this way, discrete representation processing can be achieved with two search processes + one spatial update process.

[0192] Based on the relevant contents of S401 to S402 above, it can be known that for some application scenarios (for example, scenarios with discrete representation processing requirements, etc.), after obtaining a time series signal (for example, a motion sequence signal similar to the "input motion sequence signal" in Figure 2, etc.), the time series signal can be directly discretely represented with the help of a trained data processing model to obtain discrete representation data of the time series signal, thereby achieving the purpose of discrete representation processing for any time series signal. Among them, because the data processing model has good discrete representation performance, the discrete representation determined by the data processing model for a time series signal can better represent the detailed information carried by the time series signal, so that the time series signal can more accurately represent the time series signal, thereby effectively reducing the loss of details caused by the discrete representation processing of the time series signal, thereby helping to improve the discrete representation effect.

[0193] In addition, since the quantization module in the trained data processing model is implemented with the help of a global optimal solution search algorithm (for example, a light column search algorithm), sufficient detail information can be retained in the discrete representation process involved in the data processing model, thereby enabling the compression degree of the data processing model for the time series signal to be increased as much as possible while meeting certain accuracy requirements (for example, a 9.8-second time series signal can be directly compressed to 1.22 seconds, etc.), thereby enabling the discrete representation efficiency to be improved as much as possible while meeting certain accuracy requirements.

[0194] Furthermore, the embodiments of the present disclosure do not limit the execution entity of the discrete representation determination method provided in the embodiments of the present disclosure. For example, the discrete representation determination method provided in the embodiments of the present disclosure can be applied to a terminal device or a server. For another example, the discrete representation determination method provided in the embodiments of the present disclosure can also be implemented through the data exchange process between the terminal device and the server.

[0195] In fact, in some application scenarios (for example, scenarios with time series signal generation requirements), for a trained data processing model, the data processing model can be used to perform the time series signal generation task. Based on this, the present disclosure also provides a signal generation method, as shown in Figure 5, which may include the following S501-S502. Figure 5 is a flowchart of a signal generation method provided in an embodiment of the present disclosure.

[0196] S501: Obtain target discrete representation data.

[0197] The target discrete representation data refers to the discrete representation data required to be used in the signal generation process.

[0198] In addition, the present disclosure does not limit the acquisition process of the above target discrete representation data. For example, in some application scenarios, the target discrete representation data may refer to discrete representation data input by the user in a certain manner.

[0199] For example, in some application scenarios, the target discrete representation data is determined based on the discrete representation data to be used stored in a preset storage space. The preset storage space refers to a storage space used to store some discrete representation data. Furthermore, the present disclosure does not limit the implementation of the preset storage space. For example, it can be implemented using any existing or future storage space (e.g., cache, memory, etc.). The discrete representation data to be used refers to the discrete representation data already stored in the preset storage space.

[0200] Furthermore, the present disclosure does not limit the process for determining the target discrete representation data described in the preceding paragraph. For example, in some application scenarios, the target discrete representation data may be determined by reading the discrete representation data to be used from a preset storage space as the target discrete representation data. Thus, in one possible implementation, the target discrete representation data may be the discrete representation data to be used stored in the preset storage space described above.

[0201] In fact, in order to better meet the signal generation requirements, the present disclosure also provides a possible implementation method of the above-mentioned target discrete representation data determination process. Under this implementation method, the target discrete representation data determination process can be specifically as follows: first read the discrete representation data to be used from the preset storage space; then modify the discrete representation data to be used to obtain the target discrete representation data, so that the target discrete representation data is more in line with the signal generation requirements, which is conducive to improving the diversity of timing signals. It should be noted that the present disclosure does not limit the implementation method of the modification. For example, the modification can be implemented with the help of user operations. For another example, the modification can be implemented with the help of a pre-built neural network model with modification function.

[0202] Based on the content in the above paragraph, it can be seen that in some application scenarios, the above target discrete representation data can be determined by modifying the discrete representation data to be used stored in the above preset storage space. This is conducive to obtaining a variety of high-quality timing signals, thereby better meeting the needs of timing signal diversity.

[0203] S502: Generate a timing signal corresponding to the target discrete representation data using a decoding module in a pre-built data processing model; the data processing model is determined using any implementation of the data processing method provided in the present disclosure.

[0204] The time series signal corresponding to the target discrete representation data refers to the result obtained by performing signal generation processing based on the target discrete representation data.

[0205] In addition, the present disclosure does not limit the determination process of the above-mentioned "timing signal corresponding to the target discrete representation data". For example, it can be specifically as follows: after obtaining the target discrete representation data, the decoding module in the pre-built data processing model can be used to process the target discrete representation data to obtain the timing signal corresponding to the target discrete representation data.

[0206] Based on the relevant content of S501 to S502 above, it can be seen that for some application scenarios (such as scenarios with signal generation requirements), after obtaining a discrete representation data, it is possible to directly use the trained data processing model to perform signal reconstruction processing on the discrete representation data to obtain a new time series signal, so that the time series signal can be used to perform other tasks (such as driving a three-dimensional digital model). In particular, because the data processing model has good signal reconstruction performance, the time series signal reconstructed by the data processing model is better, which is conducive to achieving the purpose of reconstructing a high-quality time series signal.

[0207] In addition, the discrete representation data used in the signal generation process can be obtained by modifying the existing discrete representation data, which is conducive to improving the diversity of the discrete representation data, thereby facilitating the reconstruction of a variety of high-quality timing signals, and further conducive to improving the diversity of timing signals.

[0208] Furthermore, the embodiments of the present disclosure do not limit the execution entity of the signal generation method provided in the embodiments of the present disclosure. For example, the signal generation method provided in the embodiments of the present disclosure can be applied to a terminal device or a server. For another example, the signal generation method provided in the embodiments of the present disclosure can also be implemented through the data exchange process between the terminal device and the server.

[0209] Based on the data processing method provided in the embodiments of the present disclosure, the embodiments of the present disclosure also provide a data processing device, which is explained and illustrated below in conjunction with Figure 6. Figure 6 is a schematic diagram of the structure of a data processing device provided in the embodiments of the present disclosure. It should be noted that for the technical details of the data processing device provided in the embodiments of the present disclosure, please refer to the relevant content of the data processing method above.

[0210] As shown in FIG6 , the data processing device 600 provided in an embodiment of the present disclosure includes:

[0211] A first acquiring unit 601 is configured to acquire abstract representation data of a first time series signal;

[0212] A first searching unit 602 is configured to search for a first discrete data combination corresponding to the abstract representation data from at least two discrete data spaces;

[0213] A first updating unit 603 is configured to update the at least two discrete data spaces according to the combination of the abstract representation data and the first discrete data;

[0214] A second search unit 604 is configured to search for a second discrete data combination corresponding to the abstract representation data from the at least two updated discrete data spaces; the at least two updated discrete data spaces are used to provide a plurality of candidate discrete data combinations, the plurality of candidate discrete data combinations including the second discrete data combination, wherein a gap representation data between the discrete representation data determined using the second discrete data combination and the abstract representation data is smaller than a gap representation data between the discrete representation data and the abstract representation data determined using any other candidate discrete data combination among the plurality of candidate discrete data combinations except the second discrete data combination;

[0215] The first determining unit 605 is configured to determine discrete representation data of the first time series signal according to the second discrete data combination.

[0216] In a possible implementation manner, the second discrete data combination is determined using a light column search algorithm.

[0217] In a possible implementation manner, the number of the first timing signals is K, where K is a positive integer;

[0218] The data processing device 600 further includes:

[0219] A third determining unit is configured to determine a target search algorithm corresponding to the k-th first time series signal from at least one candidate search algorithm, where k is a positive integer and k≤K;

[0220] The second search unit 604 is specifically used to: search for a second discrete data combination corresponding to the abstract representation data of the kth first time series signal from the updated at least two discrete data spaces according to the target search algorithm corresponding to the kth first time series signal, where k is a positive integer and k≤K.

[0221] In a possible implementation manner, the third determination unit is specifically configured to determine a target search algorithm corresponding to the k-th first time series signal from at least one candidate search algorithm according to a preset selection probability corresponding to each candidate search algorithm.

[0222] In one possible implementation, the at least one candidate search algorithm includes a light column search algorithm and a greedy algorithm.

[0223] In a possible implementation manner, the first timing signal is an action sequence signal.

[0224] In one possible implementation, the at least two discrete data spaces include a first discrete data space and a second discrete data space, and there is a difference between at least one candidate discrete data provided by the first discrete data space and at least one candidate discrete data provided by the second discrete data space.

[0225] In a possible implementation manner, the at least one candidate discrete data provided by each of the at least two discrete data spaces is the same.

[0226] In one possible implementation, the abstract representation data includes at least one eigenvector; the first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each of the eigenvectors; the second discrete data combination corresponding to the abstract representation data includes the second discrete data combination corresponding to each of the eigenvectors; for any eigenvector, the gap representation data between the discrete representation data determined by using the second discrete data combination corresponding to the eigenvector and the eigenvector is smaller than the gap representation data between the discrete representation data determined by using any other candidate discrete data combination among the several candidate discrete data combinations except the second discrete data combination corresponding to the eigenvector and the eigenvector.

[0227] In one possible implementation, the data processing model includes an encoding module, a quantization module, and a decoding module; the abstract representation data of the first time series signal is obtained by the encoding module encoding the first time series signal; the discrete representation data of the first time series signal is obtained by the quantization module performing discrete representation processing on the abstract representation data; the decoding module is used to determine the signal reconstruction result corresponding to the first time series signal based on the discrete representation data of the first time series signal.

[0228] In a possible implementation manner, the data processing device 600 further includes:

[0229] a fourth acquiring unit, configured to acquire the first time series signal from a sample data set;

[0230] A second updating unit is configured to, after determining a signal reconstruction result corresponding to the first time series signal based on the discrete representation data of the first time series signal, update the data processing model based on the first time series signal, the signal reconstruction result, the abstract representation data, and the discrete representation data, and return to continue executing the step of obtaining the first time series signal from the sample data set until a preset stop condition is reached.

[0231] Based on the relevant content of the above-mentioned data processing device 600, it can be known that for the data processing device 600 provided by the embodiment of the present disclosure, the abstract representation data of the first time series signal is first obtained; then, a first discrete data combination corresponding to the abstract representation data is searched from at least two discrete data spaces; secondly, based on the abstract representation data and the first discrete data combination, the at least two discrete data spaces are updated so that the updated at least two discrete data spaces can better provide several candidate discrete data combinations; then, a second discrete data combination corresponding to the abstract representation data is searched from the updated at least two discrete data spaces so that the gap representation data between the discrete representation data determined using the second discrete data combination and the abstract representation data is smaller than the gap representation data between the discrete representation data determined using any other candidate discrete data combination among the several candidate discrete data combinations except the second discrete data combination and the abstract representation data, so that the second discrete data combination can represent the global optimal solution determined using the updated at least two discrete data spaces; finally, based on the second discrete data combination, the discrete representation data of the first time series signal is determined, so that the purpose of discrete representation of a time series signal can be achieved.

[0232] Based on the discrete representation determination method provided in the embodiments of the present disclosure, the embodiments of the present disclosure also provide a discrete representation determination device, which is explained and illustrated below in conjunction with Figure 7. Figure 7 is a schematic structural diagram of a discrete representation determination device provided in the embodiments of the present disclosure. It should be noted that for the technical details of the discrete representation determination device provided in the embodiments of the present disclosure, please refer to the relevant content of the discrete representation determination method above.

[0233] As shown in FIG7 , the discrete representation determination apparatus 700 provided by an embodiment of the present disclosure includes:

[0234] A second acquiring unit 701 is configured to acquire a second timing signal;

[0235] The second determination unit 702 is used to determine the discrete representation data of the second time series signal using the encoding module and quantization module in the pre-built data processing model; the data processing model is determined using any implementation of the data processing method provided by the present disclosure (for example, the model training process above).

[0236] Based on the relevant content of the above-mentioned discrete representation determination device 700, it can be known that for the discrete representation determination device 700 provided by the embodiment of the present disclosure, after obtaining a time series signal, the time series signal can be directly discretely represented by means of a trained data processing model to obtain discrete representation data of the time series signal, thereby achieving the purpose of performing discrete representation processing on any time series signal. Among them, because the data processing model has good discrete representation performance, the discrete representation determined by the data processing model for a time series signal can better represent the detail information carried by the time series signal, so that the time series signal can more accurately represent the time series signal, thereby effectively reducing the loss of details caused by the discrete representation processing of the time series signal, thereby helping to improve the discrete representation effect.

[0237] Based on the signal generation method provided in the embodiments of the present disclosure, the embodiments of the present disclosure also provide a signal generation device, which will be explained and illustrated below in conjunction with Figure 8. Figure 8 is a schematic structural diagram of a signal generation device provided in the embodiments of the present disclosure. It should be noted that for the technical details of the signal generation device provided in the embodiments of the present disclosure, please refer to the relevant content of the signal generation method above.

[0238] As shown in FIG8 , the signal generating device 800 provided in an embodiment of the present disclosure includes:

[0239] The third acquisition unit 801 is used to acquire target discrete representation data;

[0240] The signal generating unit 802 is used to generate a timing signal corresponding to the target discrete representation data using a decoding module in a pre-built data processing model; the data processing model is determined using the data processing method described in any one of claims 8-9.

[0241] In a possible implementation manner, the target discrete representation data is determined based on the discrete representation data to be used stored in a preset storage space.

[0242] In a possible implementation manner, the target discrete representation data is the discrete representation data to be used.

[0243] In a possible implementation manner, the target discrete representation data is determined by modifying the discrete representation data to be used.

[0244] Based on the relevant content of the above-mentioned signal generating device 800, it can be known that for the signal generating device 800 provided by the embodiment of the present disclosure, after obtaining a discrete representation data, it is possible to directly use the trained data processing model to perform signal reconstruction processing on the discrete representation data to obtain a new time series signal, so that the time series signal can be used to perform other tasks (for example, driving three-dimensional digital models and other tasks) in the future. Among them, because the data processing model has good signal reconstruction performance, the time series signal reconstructed by means of the data processing model is better, which is conducive to achieving the purpose of reconstructing high-quality time series signals. In addition, the discrete representation data can be obtained by modifying the existing discrete representation data, which is conducive to improving the diversity of discrete representation data, thereby facilitating the reconstruction of a variety of high-quality time series signals, and further facilitating the improvement of the diversity of time series signals.

[0245] In addition, an embodiment of the present disclosure also provides an electronic device, which includes a processor and a memory: the memory is used to store instructions or computer programs; the processor is used to execute the instructions or computer programs in the memory, so that the electronic device executes any implementation of the data processing method provided by the embodiment of the present disclosure, or executes any implementation of the discrete representation determination method provided by the embodiment of the present disclosure, or executes any implementation of the signal generation method provided by the embodiment of the present disclosure.

[0246] Referring to FIG9 , a schematic diagram of the structure of an electronic device 900 suitable for implementing embodiments of the present disclosure is shown. Terminal devices in embodiments of the present disclosure may include, but are not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. The electronic device shown in FIG9 is merely an example and should not limit the functionality or scope of use of embodiments of the present disclosure.

[0247] As shown in Figure 9, the electronic device 900 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage device 908 into a random access memory (RAM) 903. Various programs and data required for the operation of the electronic device 900 are also stored in the RAM 903. The processing device 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0248] Typically, the following devices may be connected to the I / O interface 905: an input device 906 including, for example, a touch screen, a touchpad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 907 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 908 including, for example, a magnetic tape, a hard disk, etc.; and a communication device 909. The communication device 909 may allow the electronic device 900 to communicate with other devices wirelessly or by wire to exchange data. Although FIG9 shows the electronic device 900 with various devices, it should be understood that it is not required to implement or have all of the devices shown. More or fewer devices may be implemented or have alternatively.

[0249] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present disclosure includes a computer program product, which includes a computer program carried on a non-transitory computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication device 909, or installed from the storage device 908, or installed from the ROM 902. When the computer program is executed by the processing device 901, the above-mentioned functions defined in the method of the embodiment of the present disclosure are performed.

[0250] The electronic device provided by the embodiment of the present disclosure and the method provided by the above embodiment belong to the same inventive concept. For technical details not fully described in this embodiment, please refer to the above embodiment, and this embodiment has the same beneficial effects as the above embodiment.

[0251] An embodiment of the present disclosure also provides a computer-readable medium, which stores instructions or computer programs. When the instructions or computer programs are executed on a device, the device executes any implementation of the data processing method provided by the embodiment of the present disclosure, or executes any implementation of the discrete representation determination method provided by the embodiment of the present disclosure, or executes any implementation of the signal generation method provided by the embodiment of the present disclosure.

[0252] It should be noted that the computer-readable medium mentioned above in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or component, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium that contains or stores a program that can be used by or in conjunction with an instruction execution system, device, or component. In the present disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium may be transmitted using any suitable medium, including but not limited to wires, optical cables, RF (radio frequency), etc., or any suitable combination thereof.

[0253] In some embodiments, the client and server can communicate using any currently known or later developed network protocol, such as HTTP (Hypertext Transfer Protocol), and can be interconnected with any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network ("LAN"), a wide area network ("WAN"), an internet (e.g., the Internet), and a peer-to-peer network (e.g., an ad hoc peer-to-peer network), as well as any currently known or later developed network.

[0254] The computer-readable medium may be included in the electronic device, or may exist independently without being incorporated into the electronic device.

[0255] The computer-readable medium carries one or more programs. When the one or more programs are executed by the electronic device, the electronic device can perform the method.

[0256] Computer program code for performing the operations of the present disclosure may be written in one or more programming languages, or a combination thereof, including, but not limited to, object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a stand-alone software package, partially on the user's computer and partially on a remote computer, or entirely on the remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., through the Internet using an Internet service provider).

[0257] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, program segment, or a part of code, and the module, program segment, or a part of code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order than that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram and / or flowchart, and the combination of the boxes in the block diagram and / or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0258] The units involved in the embodiments described in this disclosure may be implemented in software or hardware, wherein the name of a unit / module does not, in some cases, limit the unit itself.

[0259] The functions described above herein may be performed, at least in part, by one or more hardware logic components. For example, and without limitation, exemplary types of hardware logic components that may be used include: field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chip (SOCs), complex programmable logic devices (CPLDs), and the like.

[0260] In the context of the present disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in conjunction with an instruction execution system, device or equipment. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or equipment, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium can include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0261] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Reference can be made to the common and similar parts between the various embodiments. For the systems or devices disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method description.

[0262] It should be understood that in the present disclosure, "at least one (item)" refers to one or more, and "plurality" refers to two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the previous and next associated objects are in an "or" relationship. "At least one of the following items" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0263] It should also be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or device comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article, or device. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of additional identical elements in the process, method, article, or device comprising the element.

[0264] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0265] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present disclosure. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present disclosure. Therefore, the present disclosure is not limited to the embodiments shown herein, but is intended to be construed in the widest manner consistent with the principles and novel features disclosed herein.

Claims

1. A data processing method, comprising: Acquire abstract representation data of the first time series signal; Searching for a first discrete data combination corresponding to the abstract representation data from at least two discrete data spaces; updating the at least two discrete data spaces according to the combination of the abstract representation data and the first discrete data; searching for a second discrete data combination corresponding to the abstract representation data from at least two updated discrete data spaces; the at least two updated discrete data spaces are used to provide a plurality of candidate discrete data combinations, the plurality of candidate discrete data combinations including the second discrete data combination, the gap characterization data between the discrete representation data determined by using the second discrete data combination and the abstract representation data is smaller than the gap characterization data between the discrete representation data and the abstract representation data determined by using any other candidate discrete data combination among the plurality of candidate discrete data combinations except the second discrete data combination; According to the second discrete data combination, discrete representation data of the first time series signal is determined.

2. The method of claim 1, wherein the second discrete data combination is determined using a beam search algorithm.

3. The method according to claim 1, wherein the number of the first timing signals is K, and K is a positive integer; The method further comprises: Determine a target search algorithm corresponding to the kth first time series signal from at least one candidate search algorithm, where k is a positive integer, k≤K; The step of searching for a second discrete data combination corresponding to the abstract representation data from the updated at least two discrete data spaces comprises: According to the target search algorithm corresponding to the kth first time series signal, a second discrete data combination corresponding to the abstract representation data of the kth first time series signal is searched from the at least two updated discrete data spaces, where k is a positive integer and k≤K.

4. The method according to claim 3, wherein determining the target search algorithm corresponding to the kth first timing signal from at least one candidate search algorithm comprises: According to the preset selection probabilities corresponding to the candidate search algorithms, a target search algorithm corresponding to the kth first timing signal is determined from at least one candidate search algorithm. The method according to claim 3 , wherein the at least one candidate search algorithm comprises a beam search algorithm and a greedy algorithm. The method according to claim 1 , wherein the first timing signal is an action sequence signal.

7. The method according to claim 1, wherein the at least two discrete data spaces include a first discrete data space and a second discrete data space, and there is a difference between at least one candidate discrete data provided by the first discrete data space and at least one candidate discrete data provided by the second discrete data space; or, At least one candidate discrete data provided by each of the at least two discrete data spaces are the same.

8. The method of claim 1, wherein the abstract representation data comprises at least one feature vector; The first discrete data combination corresponding to the abstract representation data includes the first discrete data combination corresponding to each of the feature vectors; The second discrete data combination corresponding to the abstract representation data includes the second discrete data combination corresponding to each of the feature vectors; For any feature vector, the gap characterization data between the discrete representation data determined by the second discrete data combination corresponding to the feature vector and the feature vector is smaller than the gap characterization data between the discrete representation data determined by any other candidate discrete data combination among the several candidate discrete data combinations except the second discrete data combination corresponding to the feature vector and the feature vector.

9. The method according to any one of claims 1 to 8, wherein the data processing model comprises an encoding module, a quantization module and a decoding module; The abstract representation data of the first time series signal is obtained by encoding the first time series signal by the encoding module; The discrete representation data of the first time series signal is obtained by the quantization module performing discrete representation processing on the abstract representation data; The decoding module is used to determine a signal reconstruction result corresponding to the first time series signal according to the discrete representation data of the first time series signal.

10. The method according to claim 9, wherein before acquiring the abstract representation data of the first timing signal, the method further comprises: Acquire the first time series signal from a sample data set; After determining the signal reconstruction result corresponding to the first time series signal according to the discrete representation data of the first time series signal, the method further includes: The data processing model is updated according to the first timing signal, the signal reconstruction result, the abstract representation data and the discrete representation data, and the step of obtaining the first timing signal from the sample data set is returned to continue until a preset stop condition is reached.

11. A method for determining a discrete representation, comprising: Acquire a second timing signal; The discrete representation data of the second timing signal is determined by using the encoding module and the quantization module in the pre-constructed data processing model; the data processing model is determined by using the data processing method described in any one of claims 9-10.

12. A signal generation method, comprising: Obtain target discrete representation data; Using a decoding module in a pre-built data processing model, a timing signal corresponding to the target discrete representation data is generated; The data processing model is determined using the data processing method described in any one of claims 9-10.

13. The method according to claim 12, wherein the target discrete representation data is determined according to the discrete representation data to be used stored in a preset storage space.

14. The method according to claim 13, wherein the target discrete representation data is the discrete representation data to be used; or, The target discrete representation data is determined by modifying the discrete representation data to be used.

15. A data processing device, comprising: A first acquisition unit, used to acquire abstract representation data of a first time series signal; A first searching unit, configured to search for a first discrete data combination corresponding to the abstract representation data from at least two discrete data spaces; A first updating unit, configured to update the at least two discrete data spaces according to the abstract representation data and the first discrete data combination; A second search unit is configured to search for a second discrete data combination corresponding to the abstract representation data from at least two updated discrete data spaces; the at least two updated discrete data spaces are configured to provide a plurality of candidate discrete data combinations, the plurality of candidate discrete data combinations including the second discrete data combination, and the gap characterization data between the discrete representation data determined by using the second discrete data combination and the abstract representation data is smaller than the gap characterization data between the discrete representation data and the abstract representation data determined by using any other candidate discrete data combination among the plurality of candidate discrete data combinations except the second discrete data combination; The first determining unit is used to determine the discrete representation data of the first time series signal according to the second discrete data combination.

16. A discrete representation determination device, comprising: A second acquisition unit, used for acquiring a second timing signal; The second determination unit is used to determine the discrete representation data of the second timing signal using the encoding module and the quantization module in the pre-built data processing model; the data processing model is determined using the data processing method described in any one of claims 8-9.

17. A signal generating device, comprising: A third acquisition unit is used to acquire target discrete representation data; A signal generating unit is used to generate a timing signal corresponding to the target discrete representation data by using a decoding module in a pre-built data processing model; the data processing model is determined by using the data processing method described in any one of claims 8-9.

18. An electronic device, comprising: Processor and memory; The memory is used to store instructions or computer programs; The processor is used to execute the instructions or computer programs in the memory so that the electronic device executes the method according to any one of claims 1 to 14.

19. A computer-readable medium, wherein the computer-readable medium stores instructions or computer programs, and when the instructions or computer programs are executed on a device, the device executes the method according to any one of claims 1 to 14.

Images