Method for managing the transmission of data including semantic content

Abstract

Method for managing the transmission of data including semantic content. The invention relates to a method for managing the transmission of data (D) including semantic content through a communication channel (C), characterized in that it comprises the following steps: a step of calculating data (S) representative of the semantic content of the transmitted data (D); a step of transmitting through the channel (C) both the transmitted data (D) and the data (S) representative of the previously calculated semantic content. Figure 1

Description

Title of the invention: Method for managing the transmission of data including semantic content Technical field

[0001] The technical field is that of data transmission and error detection in computing and telecommunications.

[0002] More specifically, the invention relates to a method for managing the transmission of data including semantic content through a transmission channel. The invention also relates to a method for managing the detection of errors occurring during the transmission of data including semantic content.

[0003] We can talk about data transmission in several cases, for example when computer data is transmitted over a computer network, which is local to several computers, or over a telecommunications network which can be local or long distance, or when data is transferred from one medium to another, that is to say for example copied from a file system of a computer to a memory medium connected to the computer such as a USB key or an external hard drive, or vice versa. In all these transmission cases, transmission errors can always occur.

[0004] Computer data is conventionally defined as a sequence of bits. During the transmission of data, one or more bits constituting this data may be modified by the transmission channel. In this case, there is a transmission error, and the data is no longer transmitted in its entirety.

[0005] Furthermore, computer data can also represent information that has meaning for human users receiving this data. For example, a text encoded in the form of computer data is then transformed into a sequence of bits but also has meaning and therefore has semantic content. This is also the case for an image or videos or audio tracks. Recent developments in artificial intelligence are based on the possibility of objectifying this notion of semantic content by being able to extract from computer data another data representative of its semantic content.Two different data but whose semantic content is similar, for example two different images of cats (one black and the other white), will then be very different if we consider their representation as a sequence of bits (the pixels indicating the color of the cat will have opposite representations to each other) but will be closer if we consider the data representing the semantic content of the two images.

[0006] From the beginnings of computing and telecommunications, several techniques of Error detection has been developed, so that transmission can be restarted when a transmission error is detected. State of the art

[0007] Error detection techniques are generally based on the following principle.

[0008] The computer data considered are presented as sequences of bits. One or more redundant data are calculated from the bit sequence in question. For example, checksums or cyclic redundancy checks (CRCs) can be calculated. These redundant data are transmitted together with the original data. The entity that receives the original data, or that manages the new medium to which it has been transferred, will then perform the same redundant data calculation from the received data and compare the result with the received redundant data. A difference between the received redundant data and the data recalculated from the original data indicates that there has been a transmission error.

[0009] Once a transmission error is detected, several actions are possible, depending on the capabilities provided by the redundant data calculation formula. In the simplest version, once an error is detected, the action consists of requesting a new transmission of the original data and its associated redundant data. In a more elaborate version, the redundant data is calculated in such a way as to allow the location where the transmission error occurred to be located. The corresponding action will then consist of retransmitting this portion transmitted with an error. In an even more elaborate version, the redundant data not only allows the transmission error to be located but also allows a correction to be proposed, without having to request a retransmission. This is called an error-correcting code.

[0010] These techniques are now well established in the general field of computing and telecommunications.

[0011] However, the techniques in question only focus on the representation of data in the form of bit sequences. In this case, transmission errors will be detected as soon as at least one bit is inverted by the transmission. This results in a significant number of retransmissions and therefore a significant waste of energy or material resources.

[0012] The invention improves the situation. Statement of the invention

[0013] According to a first functional aspect, the invention relates to a method for managing the transmission of data including semantic content through a communication channel. communication characterized in that it comprises the following steps: • a step of calculating data representative of the semantic content of the transmitted data; • a step of transmission through the channel of both the transmitted data and the data representative of the semantic content calculated previously.

[0014] Thanks to the invention, the semantic content of a data item is taken into account during transmission in order to produce control data which will be used to perform error detection. This data item representative of the semantic content is calculated and transmitted, preferably at the same time as the original data item, to perform subsequent checks.

[0015] Until now, error detection techniques have only focused on the representation of transmitted data in the form of bit sequences. However, transmitted data may have rich semantic content, which is intrinsic to the subsequent use of the data in question. For example, if a piece of data is a text, its semantic content is essential in the future use of this textual data, namely its reading. And similarly, if the data in question is an image, its semantic content, namely the image represented, its figurative content if the image is figurative or abstract otherwise, is the reason for the existence of the data. Recent progress in artificial intelligence is largely based on the possibility of having objective data representing the notions of semantic content of other data.

[0016] Such semantic content may very well withstand certain data transmission errors. Or, on the contrary, transmission errors that are quantitatively insignificant may have strong semantic importance. For example, a few changes in pixel colors in an image will not a priori fundamentally change the meaning of the image in question. But if the image is a road traffic situation and the few pixels that change color are those of a traffic light that thus changes from red to green, the meaning of the image changes completely. In the first case, even if there have been transmission errors in the data represented by a sequence of bits, it is not necessary to request a retransmission of the data, whereas it is necessary in the second case.

[0017] It is possible to request that data be retransmitted beyond a threshold number of bits that have been transmitted erroneously. But in the first case, unnecessary retransmission requests will be made. Whereas in the second case, if the threshold for triggering a retransmission is high, necessary retransmissions will not be requested.

[0018] The advantage of the invention is therefore to avoid carrying out retransmissions each time there has been transmission of erroneous bits. The invention therefore provides savings in energy and material resources. A retransmission is not requested if the semantic content has not been modified, or in a small proportion that is acceptable to the receiver. The calculated data representative of the semantic content will make it possible to evaluate this proportion of modification of the semantic content.

[0019] According to a first embodiment, the calculated data representative of the semantic content of the transmitted data is a vector of real numbers.

[0020] In the field of artificial intelligence, a technique called embedding has now been established. The embedding technique was initially applied to texts and recently to images. It consists of finding a representation of data in the form of a vector of real numbers while preserving the semantic content of the data. This representation is then used for all types of tasks, for example machine translation when the data is a text. According to the invention, the representation in the form of a vector of real numbers will be used as control data to ensure whether or not the semantic content of data is preserved during its transmission.

[0021] According to one embodiment, which may be implemented alternatively or cumulatively with the preceding embodiments, the calculated data representative of the semantic content of the transmitted data is a vector of complex numbers. It is possible to use a vector of complex numbers rather than a vector of real numbers as the target of the embedding. This choice of data makes it possible to encode more information in each point of the vector, since each point is then a complex number which will contain two real data.

[0022] According to one embodiment, which may be implemented alternatively or cumulatively with the preceding embodiments, the calculated data representative of the semantic content of the transmitted data is a computer object which takes the form of an array of floats. A real number must be represented in a finite manner to be processed in a computer manner by a computer. A float is such a representation of a real number for computer processing. To reduce the size taken by the calculated data representative of the semantic content, the number of bits used to encode each float will be limited, for example to 32 bits.

[0023] According to a second embodiment, which may be implemented alternatively or cumulatively with the previous embodiment, the calculated data representative of the semantic content of the transmitted data is obtained by applying a transformer to the transmitted data.

[0024] In the field of artificial intelligence, a transformer is a category of tool that processes semantic data for a task. A first particularity of transformers is to ensure, when they process sequential data (such as text), to preserve the semantic relationship that exists between the different elements of the sequential data (the different units lexical, or words, of the text, to continue the previous example) using a so-called attention mechanism. The use of transformers is therefore a way of guaranteeing a good representation by the calculated data of the semantic content of the transmitted data.

[0025] Another particularity of transformers is to be able to process sequential data (such as text) without having to process the data in question sequentially, that is to say without having to consult the data, with a view to its processing, starting from the beginning and going to the end of the data. In other words, a transformer will process sequential data in a single block. In the invention, the task carried out by the transformer is to calculate data representing the semantic content of the transmitted data. The advantage of using a transformer for this task is to allow processing of the data in a single block, which makes it possible to carry out calculations in parallel. The calculation is therefore more efficient and consumes less energy and hardware resources. Initially created for text processing, transformers are now also applied to images and therefore potentially to videos.

[0026] According to one embodiment, which may be implemented alternatively or cumulatively with the preceding embodiments, the calculated data representative of the semantic content of the transmitted data is obtained using an artificial neural network.

[0027] Using deep learning techniques, it is possible to configure an artificial neural network to perform a dedicated task. The task targeted here is to produce data representative of the semantic content of the transmitted data. Deep learning will ensure that a neural network produces such data, in a robust manner, with a reasonable size of data representative of the semantic content.

[0028] According to one embodiment, which may be implemented alternatively or cumulatively with the preceding modes, the calculated data representative of the semantic content of the transmitted data is obtained using a variational auto-encoder.

[0029] A variational autoencoder is a particular type of artificial neural network that can be used for this type of task.

[0030] According to one embodiment, which may be implemented alternatively or cumulatively with the preceding modes, the calculated data representative of the semantic content of the transmitted data is obtained by extracting internal data from a recurrent neural network.

[0031] A recurrent neural network is a type of artificial neural network which has the particularity of presenting cycles in the influence relationships between the different neurons present in the network. In this case, the application of the neural network to a piece of data, here the transmitted data, will trigger modifications to the weights of the neurons present in the neural network, these modifications being propagated by the cycles present in the architecture of the neural network. The weights of the neurons thus modified can then be extracted and form data representative of the semantic content of the transmitted data.

[0032] In particular, it will be possible to use a recurrent neural network with short and long term memory (translation of the English Long short-term memory), also called LSTM recurrent neural network (for the acronym of Long Short-Term Memory).

[0033] According to a third embodiment, which may be implemented alternatively or cumulatively with the preceding modes, the channel is adapted to the direct transmission of the semantic content of the transmitted data and the calculated data representative of the semantic content of the transmitted data is obtained by an operation of summarizing the semantic content of the transmitted data.

[0034] One possible way to improve the transmission of data corresponding to semantic contents is to set up transmission channels adapted to the direct transmission of the semantic content of a piece of data. Rather than aiming for the transmission of a sequence of bits, the transmission channel thus adapted will aim for the transmission of a piece of data such as an embedding in the form of a vector of real numbers for example. We can then speak of a semantic channel to name the transmission channel.

[0035] In this case of direct transmission of the semantic content of a data item, the control data item representing the semantic content of the transmitted data item must be adapted. The use of an automatic summarization process makes it possible to obtain this result. The advantage of choosing a summary operation is to reduce the size of the control data item relative to the transmitted data item. This reduction in size is important to ensure that the transmission process does not add too large a quantity of data to be transmitted by adding control data item.

[0036] According to a second functional aspect, the invention relates to a method for managing the detection of errors occurring during the transmission of data including semantic content through a communication channel, taking into account the semantic content of the transmitted data, characterized in that it comprises the following steps: • a step of receiving at the same time two data corresponding on the one hand to the data including a semantic content transmitted through the channel and on the other hand to a data calculated from the data including a content semantic, representative of this semantic content; • a step of calculating data representative of the semantic content of the received data including semantic content; • an error detection step based on a comparison between the received data calculated before transmission and the data calculated after transmission, the comparison taking into account their respective semantic contents.

[0037] Thanks to the second functional aspect of the invention, the detection of transmission errors itself is carried out. Two data are received, the transmitted data itself on the one hand and a calculated data representing the semantic content of the transmitted data on the other hand. From the received transmitted data, a similar calculation is made, then the two calculated data, that which was calculated before transmission then transmitted, and that which was calculated after transmission, are compared. The calculation method before and after transmission will advantageously be the same in order to improve the consistency of the results.

[0038] The comparison makes it possible to detect possible transmission errors that have an impact on the semantic content of the transmitted data. If the calculated semantic contents are identical, or very close, it will be considered that there is no transmission error. On the contrary, if the calculated semantic contents are very different, the method detects a transmission error, which could for example trigger a retransmission. This difference between semantic contents is partly independent of the transmission errors that have occurred on the sequence of bits transmitted in the communication channel.

[0039] Only transmission errors that have an impact on the semantic content are therefore corrected, which reduces the requirements for hardware and energy resources. Error detection can use a threshold mechanism, or any other mechanism, to evaluate the difference between the two calculated data and estimate the need for retransmission.

[0040] According to a first embodiment of this second functional aspect, the two calculated data are vectors of real numbers of identical size and the comparison of the two calculated data is done by calculating their cosine similarity.

[0041] In this embodiment, the calculated data, representative of the semantic content, are embeddings into vectors of real numbers. An effective way to compare this type of data is to use the cosine similarity between the two vectors. The value thus calculated will be between -1 and 1; it is 1 when the two vectors are opposite, 0 when they are orthogonal; and 1 when they are collinear. A value sufficiently close to 1 will therefore indicate that the transmission has not modified the semantic content of the transmitted data, in any case not in significant proportions, and a retransmission will not be necessary.

[0042] According to one embodiment, which may be implemented alternatively or cumulatively with the preceding embodiments, the two calculated data are vectors of real numbers of identical size and the comparison of the two calculated data is done by calculating their Euclidean distance. This embodiment provides a possible alternative to the comparison of the two calculated data.

[0043] According to a second embodiment of this second functional aspect, which may be implemented alternatively or cumulatively with the previous mode, the transmission channel is adapted to the direct transmission of the semantic content of a transmitted data item and the two calculated data items are obtained by an operation of summarizing the semantic content of the transmitted data item before and after transmission.

[0044] In this mode, as seen previously, the semantic content is already transmitted as is. The data for checking the correct transmission can then be obtained by a semantic summary method adapted to the nature of the transmitted data which has just been received.

[0045] According to a first material aspect, the invention relates to an entity for managing the transmission of data including semantic content through a communication channel, characterized in that it comprises the following modules: • a module for calculating data representative of the semantic content of the transmitted data; • a transmission module through the channel of both the transmitted data and the data representing the semantic content calculated by the calculation module.

[0046] According to a second material aspect, the invention relates to an entity for managing the detection of errors occurring during the transmission of data including semantic content through a communication channel, the error detection taking into account the semantic content of the transmitted data, characterized in that it comprises the following modules: • a module for receiving two data items at the same time, corresponding on the one hand to data items including semantic content transmitted through the channel and on the other hand to data items calculated from the data items including semantic content; • a module for calculating data representative of the semantic content of the received data including semantic content; • an error detection module based on a comparison between the received data calculated before transmission and the data calculated after transmission, the comparison taking into account their respective semantic contents.

[0047] According to a third material aspect, the invention relates to network equipment comprising a management entity, according to the invention, for the transmission of data including semantic content through a communication channel.

[0048] According to a fourth material aspect, the invention relates to network equipment comprising a management entity, according to the invention, for the detection of errors occurring during the transmission of data including semantic content through a communication channel.

[0049] The network equipment in question here may be any type of network equipment. It may be, for example, routers, or gateways, or terminals. Any type of network equipment may be supplemented by management entities which will make it possible to manage the transmission and error detection taking into account the semantic content of the data according to the invention.

[0050] According to another material aspect, the invention relates to a computer program capable of being implemented by equipment, the program comprising code instructions which, when executed by a processor, carries out the steps of the method for managing the transmission of data defined above.

[0051] According to another material aspect, the invention relates to a computer program capable of being implemented by equipment, the program comprising code instructions which, when executed by a processor, carries out the steps of the error detection management method defined above.

[0052] Finally, according to another material aspect, the invention relates to data media on which are recorded computer programs comprising sequences of instructions for implementing the transmission management and error detection methods defined above.

[0053] The data carriers may be any entity or device capable of storing the programs. For example, the carriers may comprise a storage means, such as a ROM, for example a CD ROM or a microelectronic circuit ROM, or a magnetic recording means such as a hard disk. Furthermore, the carriers may be transmissible media such as an electrical or optical signal, which may be conveyed via an electrical or optical cable, by radio or by other means. The programs according to the invention may in particular be downloaded from a network such as the Internet. Alternatively, the information carrier may be an integrated circuit in which the program is incorporated, the circuit being adapted to execute or to be used in the execution of the method in question. Brief description of the figures

[0054] The invention will be better understood upon reading the following description, given as a example, and made with reference to the attached drawings in which:

[0055] [Fig. 1] represents an entity for managing the transmission of data and an entity for managing error detection according to the invention.

[0056] [Fig.2] represents a succession of steps corresponding to a possible execution of the methods according to the invention.

[0057] [Fig.3] represents an example of data transmitted according to the invention. Detailed description

[0058] [Fig.l] represents a management entity 100 for the transmission of data D through a communication channel C and a management entity 200 for the detection of errors occurring during the transmission of data D through the channel C.

[0059] The communication channel C is represented by two dotted lines. The C channel is for example a WAN (acronym for Wide Area Network) which connects two network devices across a long distance. The C channel can also be a LAN (acronym for Local Area Network) which connects nearby devices. The C channel can also be formed by the chaining of several networks: LAN networks collecting traffic at a given location, then one or more WAN networks connecting LAN networks separated by a long distance. The communication channel C can be formed by several network devices such as routers, gateways, multiplexers or any other equipment.The physical medium of transmission can use electrical transmission, for example along copper wires, or optical transmission along an optical fiber, or the hertzian transmission of an electromagnetic wave through space, or the combination of these different media obtained by chaining separate networks. The data transmitted along the communication channel C can be transported using any type of communication protocol, for example the TCP protocol (acronym for Transmission Control Protocol) or the ATM protocol (acronym for Asynchronous Transfer Mode) or the Ethernet or GigaEthernet protocol, or any suitable protocol in general.

[0060] The management entity 100, as well as the management entity 200, are hosted in equipment or devices which have the hardware architecture of a conventional computer. The management entities 100 and 200 can in particular access one or more processors, RAM type random access memories and read-only memories such as Flash, ROM type memory (not shown in the figure). The management entities 100 and 200 can also in certain embodiments access input-output devices such as keyboards and / or screens (not shown in the figure) but other embodiments will not have them. The management entities 100 and 200 can also access specialized input-output devices allowing data to be exchanged, such as network ports, USB ports (acronym for Universal Serial Bus), Ethernet ports, antennas such as Bluetooth or WiFi or 5G antennas or any other similar device.

[0061] According to the invention, the management entity 100 implements the following steps of a method for managing the transmission of data D including semantic content through a communication channel C: • a step of calculating data S representing the semantic content of the transmitted data D; • a transmission step through channel C of both the transmitted data D and the data S representing the semantic content calculated previously.

[0062] The management entity 100 comprises the following modules for implementing the method: • a module 101 for calculating data S representing the semantic content of the transmitted data; • a module 102 for transmitting through the channel C both the transmitted data D and the data S representing the semantic content calculated by the calculation module 101.

[0063] According to the invention, the management entity 200 implements the following steps of a method for managing the detection of errors occurring during the transmission of a data item D including a semantic content through a communication channel C: • a step of receiving both two data items D' and S' corresponding on the one hand to the data item D including a semantic content transmitted through the channel C and on the other hand to a data item S calculated from the data item D including a semantic content, the data item S being representative of this semantic content; • a step of calculating data S” representative of the semantic content of the data D' received including semantic content; • an error detection step based on a comparison between the received data S' calculated before transmission and the data S” calculated after transmission, the comparison taking into account their respective semantic contents.

[0064] The management entity 200 comprises the following modules for implementing the steps referred to above: • a module 201 for receiving two data D' and S' corresponding on the one hand to a data D including a semantic content transmitted through the channel C and on the other hand to a data S calculated from the data D including a semantic content, representative of this semantic content; • a module 202 for calculating data S” representative of the semantic content of the data D' received including semantic content; • an error detection module 203 based on a comparison between the received data S' calculated before transmission and the data S” calculated after transmission, the comparison taking into account their respective semantic contents.

[0065] In other words, the transmitted data D and the data S calculated by the calculation module 101 belonging to the transmission management entity 100 are provided at the input of the channel C, represented by two dotted lines, by the transmission module 102 belonging to the transmission management entity 100. After transmission, these data D and S are transformed into data D' and S' respectively following possible transmission errors. The data D and S are transmitted at the same time, which is represented by the symbol D+S at the input of the channel C; and they are also received at the same time, which is represented by the symbol D'+S' at the output of the channel C by the reception module 201 belonging to the error detection management entity 200. The calculation module 202 belonging to the error detection management entity 200 will carry out the calculation of the data S” from the received data D'.Then the error detection module 203 belonging to the error detection management entity 200 will carry out the comparison between the two data S' and S”. The comparison carried out by the module 203 is represented, in [Fig.l], by the symbol S'?S” at the output of the module 203. It is this comparison result which makes it possible to detect or not a transmission error, by taking into account the semantic content of the data D transmitted and the data D' received.

[0066] [Fig. 2] illustrates a sequence of steps according to one embodiment of the invention. These steps are referenced S1 to S8.

[0067] During a first step S1, a data item D to be transmitted is submitted to the management entity 100 in charge of its transmission. The data item D has a semantic content and the methods according to the invention will exploit this semantic content to detect possible errors in the transmission through the channel C. The management entity 100 then calculates, during a second step S2, from the data item D, a data item S which is representative of the semantic content of the data item D. It is the calculation module 101 which carries out this operation.

[0068] The data D can be for example a text, and the calculation carried out by the module 101 can be an embedding which will produce a vector of reals. For this, the module 101 can use the word2vec algorithm known to those skilled in the art. Another possibility is to use a transformer type algorithm, also called a self-attentive model, which is a deep learning model. The advantage of using a transformer is that these models begin to apply to data that are images.

[0069] The data D can also be an image and the calculation carried out by the module 101 can be carried out by a visual transformer (translation of the English visitai transformer). The data D can also be an audio track and the module 101 will produce data from the spectrogram of the track in question.

[0070] Once the data S has been obtained by the calculation module 101, the transmission module 102 will proceed during a step S3 to the actual transmission of the data D through the channel C. According to the invention, the transmission of the data D must also include the transmission of the data S. The transmission of the data D and S can be done jointly, or sequentially, or even with a certain delay depending on the transmission protocols used. During step S3, the transmission module 102 submits the data D and S to the channel C which will transmit them to their destination. The transmission of the data D and S can be done by transforming them into bit sequences as is conventionally done in digital telecommunications.For example, if the data S is a vector of real numbers, an encoder such as that defined by the IEEE (acronym for the English Institute of Electrical and Electronic Engineers) standard number 754 which allows the representation of floating point real numbers can be used.

[0071] This data is then received during a step S4 by the module 201 of the management entity 200 which will manage the detection of errors in the transmission.

[0072] The data received, and therefore produced following step S4, are called D' and S' and correspond respectively to the data D and S. The transmission of the data D and S may have undergone transmission errors along the channel C, which explains why the data received are called D' and S' in order to take into account possible modifications following the transmission errors.

[0073] The data D' is provided, during a step S5, to the module 202 of the management entity 200.

[0074] The module 202 will then perform, during a step S6, a calculation of a data item S” representative of the semantic content of the data item D'. This calculation will, in general, be identical to that performed by the calculation module 101 of the management entity 100 of the transmission of the data item D. However, instead of performing the same calculation, it is possible to imagine differences in the calculation performed by the modules 101 and 202, for example to optimize the calculation time on reception, by performing a simpler calculation but which still extracts the semantic content of the data item D' sufficiently to be able to detect a transmission error.

[0075] The module 203 of the error detection management entity 200 will then carry out, during a step S7, a comparison between the two calculated data S' and S”. One way of carrying out this comparison may be, for example, to calculate the similarity cosine if the two data S' and S” are two vectors of real numbers of the same dimension. In general, suitable distances can be defined to determine the semantic proximity of the data S' and S”. The comparison between the data S' and S” can take into account the differences in calculation methods between these two data.

[0076] Any other type of distance can be used, such as a Euclidean distance, also called an L2 distance. A Euclidean distance is calculated by the formula , / vü ct is called an L2 distance because it is formed by the square root of sum of squares. Using a power of 1, we obtain a Manhattan distance, calculated by the formula - - y], also called the Ll distance. It is possible to use a Minkowski distance where we use powers greater than 2.

[0077] It is also possible to use a discrete distance, also called LO distance. In the case of a discrete distance, the distance between the two calculated data S' and S” is 0 if the two calculated data S' and S” are identical and is 1 from the moment that the two data S' and S” are different.

[0078] This comparison then makes it possible to detect, during a step S8 carried out by the module 203, a possible transmission error. If the two data S' and S” are very far apart, a transmission error has occurred which impacts the semantic content of the transmitted data D, received as data D', and it must therefore be corrected. If the two data S' and S” are very close, it is possible that there have been one or more transmission errors at the level of the bit sequence, but these do not impact, or only slightly, the semantic content of the transmitted data D, received as data D'. The result of the comparison carried out in step S7, denoted S'?S”, makes it possible to decide in step S8 whether the possible transmission errors justify carrying out a new transmission of the data D.

[0079] [Fig. 3] represents an example of data D transmitted according to the invention. Data D is an English text, namely the sentence "The dog has corne off its leash." For computer processing, data D is represented as a sequence of bits, present above the sentence. Artificial intelligence techniques make it possible, independently of the representation of data D as a sequence of bits, to obtain a representation thereof which preserves the semantic content of data D and makes it possible to carry out processing of this semantic content. Module 101 performs such an operation and performs an embedding of data D to represent it as a vector of real numbers, which here have the value [0.134, 3.14728, ... , 0.98]. The data thus obtained is data S, representative of the semantic content of data D. Data S will be encoded by an encoder ENC into a sequence of bits, as indicated in the figure.The ENC encoder in question can be, for example, the encoder defined by the IEEE standard (acronym for the English Institute of Electrical and . Electronic Engine ers) number 754 which allows real numbers to be represented in floating point arithmetic in the form of a sequence of bits. The representation of the data S in the form of a sequence of bits will then allow the data D and S to be combined and transmitted at the same time in the transmission channel C (not shown in [Fig.3])•

[0080] Furthermore, the transmission channel C may be a conventional transmission channel as used in digital telecommunications and the data D and S transmitted are then sequences of bits. However, it is possible that the channel C is adapted to transmit data by directly taking into account their semantic content, namely here, for example, the embedding of the data D in the form of a vector of real numbers. The channel C is then called a semantic channel. In this embodiment, the semantic channel C ensures communication between two remote computer applications which transmit data D comprising semantic content, for example data D from artificial intelligence applications.The semantic channel C is then an abstraction accessible to computer applications to transmit such data, without having to take into account the encodings of data in bit sequences or other formats used in low-level protocol layers. Since the transmitted data D, in this case, is directly data representing semantic content, the control data S, calculated from the transmitted data D to allow error detection, the data D being received as data D', and the data S” calculated from the data D', can be obtained by semantic summary operations since the semantic content is transmitted directly by the transmission semantic channel C.

[0081] Finally, let us point out here that, in the present text, the term “module” can correspond to a software component as well as to a hardware component or a set of hardware and software components, a software component itself corresponding to one or more computer programs or sub-programs or, more generally, to any element of a program capable of implementing a function or a set of functions as described for the modules concerned. In the same way, a hardware component corresponds to any element of a hardware assembly capable of implementing a function or a set of functions for the module concerned (integrated circuit, smart card, memory card, etc.).

Claims

Claims

1. Method for managing the transmission of data (D) including semantic content through a communication channel (C) characterized in that it comprises the following steps: • a step of calculating data (S) representative of the semantic content of the data (D) transmitted; • a step of transmitting through the channel (C) both the data (D) transmitted and the data (S) representative of the semantic content calculated previously.

2. Method for managing the transmission of data (D) according to claim 1 characterized in that the calculated data (S) representative of the semantic content of the transmitted data (D) is a vector of real numbers.

3. Method for managing the transmission of data (D) according to one of claims 1 or 2, characterized in that the calculated data (S) representative of the semantic content of the transmitted data (D) is obtained by applying a transformer to the transmitted data (D).

4. Method for managing the transmission of data (D) according to one of claims 1 to 3, characterized in that the channel (C) is adapted to the direct transmission of the semantic content of the data (D) transmitted and the calculated data (S) is obtained by an operation of summarizing the semantic content of the data (D) transmitted.

5. Method for managing the detection of errors occurring during the transmission of data (D) including semantic content through a communication channel (C), taking into account the semantic content of the data (D) transmitted, characterized in that it comprises the following steps: • a step of receiving both two data (D', S') corresponding on the one hand to the data (D) transmitted through the channel (C) and on the other hand to a data (S) calculated from the data (D), representative of the semantic content of the data (D); • a step of calculating a data (S”) representative of the semantic content of the data (D') received; • an error detection step based on a com- comparison between the received data (S') calculated before transmission and the data (S”) calculated after transmission, the comparison taking into account their respective semantic contents.

6. Error detection management method according to claim 5, characterized in that the two calculated data (S', S”) are vectors of real numbers of identical size and the comparison of the two calculated data (S', S”) is done by calculating their cosine similarity.

7. Error detection management method according to one of claims 5 or 6, characterized in that the transmission channel (C) is adapted to the direct transmission of the semantic content of a transmitted data item (D) and the two calculated data items (S', S”) are obtained by an operation of summarizing the semantic content of the transmitted data item (D) before and after transmission.

8. Management entity (100) for the transmission of data (D) including semantic content through a communication channel (C) characterized in that it comprises the following modules: • a module (101) for calculating data (S) representative of the semantic content of the data (D) transmitted; • a module (102) for transmitting through the channel (C) both the data (D) transmitted and the data (S) representative of the semantic content calculated by the calculation module (101).

9. Management entity (200) for the detection of errors occurring during the transmission of data (D) including semantic content through a communication channel (C), the error detection taking into account the semantic content of the data (D) transmitted, characterized in that it comprises the following modules: • a module (201) for receiving two data (D', S') corresponding on the one hand to a data (D) transmitted through the channel (C) and on the other hand to a data (S) calculated from the data (D), representative of the semantic content of the data (D), the two data (D, S) being transmitted at the same time; • a module (202) for calculating a data (S”) representative of the semantic content of the data (D') received; • an error detection module (203) based on a comparison between the received data (S') calculated before transmission and the data (S”) calculated after transmission, the comparison taking into account their respective semantic contents.

10. Network equipment comprising a management entity (100) for the transmission of data (D) including semantic content through a communication channel (C) according to claim 8.

11. Network equipment comprising a management entity (200) for the detection of errors occurring during the transmission of data (D) including semantic content through a communication channel (C) according to claim 9.

12. A computer program capable of being implemented by equipment, the program comprising code instructions which, when executed by a processor, performs the steps of the transmission management method defined in claim 1.

13. A computer program capable of being implemented by equipment, the program comprising code instructions which, when executed by a processor, performs the steps of the error detection management method defined in claim 5.

14. Data carrier, on which is recorded a computer program comprising a sequence of instructions for implementing the transmission management method according to claim 1 when it is loaded into and executed by a processor.

15. Data medium, on which is recorded a computer program comprising a sequence of instructions for implementing the error detection management method according to claim 5 when it is loaded into and executed by a processor.

Images