Method for generating a manufacturing data set for a cosmetic composition to be modified and associated devices

Abstract

Method for generating a manufacturing data set for a cosmetic composition to be modified and associated devices. The present invention relates to a method for generating at least one manufacturing data set for a cosmetic composition, said method comprising a step of: - obtaining a manufacturing data set and one or more manufacturing data to be modified in the set obtained, - calculating a similarity coefficient between each manufacturing data to be modified and a plurality of candidate manufacturing data, - selecting, for each manufacturing data to be modified, at least one manufacturing data according to the calculated similarity coefficients, to obtain at least one selected data, - generating a manufacturing data set, to obtain a generated set in which each manufacturing data to be modified is replaced by a selected manufacturing data.Figure for the abbreviation: figure 2.

Description

Title of the invention: Method for generating a manufacturing data set for a cosmetic composition to be modified and associated devices TECHNICAL FIELD OF THE INVENTION

[0001] The present invention relates to a method for generating a set of manufacturing data enabling the formulation, preparation, and / or manufacture of a cosmetic composition. It also relates to a method for manufacturing a cosmetic composition. It also relates to the devices involved in the preceding methods. TECHNOLOGICAL BACKGROUND OF THE INVENTION

[0002] The formulation of environmentally friendly cosmetic compositions, that is to say, whose design and development take into account environmental issues, is becoming a major concern in order to help meet global challenges.

[0003] It is therefore essential to propose more sustainable compositions and / or preparation processes and / or raw materials, thus enabling us to meet these environmental challenges.

[0004] In particular, in this context, it becomes important to replace certain materials with alternatives that have a better environmental footprint, in particular by reducing the use of raw materials from petrochemicals.

[0005] One objective is to offer eco-responsible materials, in particular those derived from green chemistry, in order to limit the environmental impact of the composition (and, in addition, of its primary and / or secondary packaging) from its manufacture to its end of life, in particular through materials with a good biodegradability profile and / or from renewable sources.

[0006] The formulation of new cosmetic compositions may also be the result of new regulations.

[0007] For example, following European directives of 2015, it was prohibited to produce lipstick or nail polish containing cobalt nitrate.

[0008] Reformulation may also be required by the fact that a raw material may be acceptable in one legislation but not in another.

[0009] More broadly, it may also be desirable to modify a cosmetic composition with a view to optimizing its manufacturing cost, optimizing its industrialization, or even depending on the availability of its constituent ingredients. It may also be desirable to modify it to improve and evolve its properties for consumers.

[0010] It can also be useful when it comes to ensuring the availability of ingredients or raw materials for certain cosmetic compositions. Summary of the invention

[0011] There is therefore a need for a process which is easy to implement and which makes it possible to generate a set of manufacturing data for a cosmetic composition that meets a specific context, in particular a reduced environmental footprint.

[0012] To this end, the description relates to a method for generating at least one set of manufacturing data, the manufacturing data set comprising manufacturing data for a cosmetic composition, said method being implemented by a computing device and comprising a generation phase, the generation phase comprising a step of:

[0013] - obtaining a set of manufacturing data and one or more data manufacturing process to be modified in the overall result,

[0014] - calculation of a similarity coefficient between each manufacturing data to be modified and a plurality of candidate manufacturing data,

[0015] - selection, for each manufacturing data to be modified, of at least one data manufacturing data is selected from among the plurality of candidate manufacturing data based on the calculated similarity coefficients, in order to obtain at least one selected data point.

[0016] - generation of at least one set of manufacturing data, to obtain at less one generated set, each generated set being the obtained set in which each manufacturing data to be modified is replaced by a respective selected manufacturing data.

[0017] According to other advantageous aspects of the invention, the generation process comprises one or more of the following features, taken individually or in all technically possible combinations:

[0018] - each manufacturing data point is associated with several values, each value being a quantification of a respective property of the manufacturing data, the similarity coefficient of a manufacturing data depending on at least one value of the manufacturing data.

[0019] - a property is a measurable property.

[0020] - a property is a property that can be assessed qualitatively.

[0021] - during the calculation step, the calculation device (12) calculates at least one coefficient similarity can be determined by applying the following formula:

[0022] r\ — i . ij] E ( ^cand,— ~ 1^^111111(4'^ ( Xllia / :. Xcand, j ), 1] j

[0023] With: • C(A, B) denotes the similarity coefficient between manufacturing data A and manufacturing data B, • xmodî denotes the i-th manufacturing data point to be modified, i being a non-zero integer. • xcand. j denotes the j-th candidate manufacturing data point, where j is a non-zero integer. • w denotes the sum of the weights applied to the p properties • wp denotes the weight applied to the p-th property (Wp = 1 by default) • n is the number of properties,

[0024] ^P ( xmodlp xcand. J)~ Rp • -^modip: value of the p-th property of xmodi, p being an integer between 1 and n, • ycmd, j : value of the p-th property of xcand, j, • Rp: this value is at least equal to the difference between the 5th and 95th percentile of the distribution of values ​​of the p-th property. - for a specific manufacturing data to be modified, the selected candidate manufacturing data is the deletion of the specific manufacturing data, each generated set then being the set obtained in which the specific manufacturing data is deleted and each other manufacturing data to be modified possibly is replaced by a respective selected manufacturing data. - the cosmetic composition is part of a cosmetic product, the manufacturing data being chosen from a list consisting of data relating to the cosmetic composition, data relating to packaging of the cosmetic product and data relating to the manufacturing technique of the cosmetic product used. - The manufacturing data are selected from a list of raw materials, a list of molecular structures, and a list of characteristics.

[0025] The description also relates to a method for generating a manufacturing data set, the manufacturing data set comprising manufacturing data for a cosmetic composition likely to meet at least one acceptability criterion, said method being implemented by a computing device and comprising an inference phase, the inference phase comprising a step of: - obtaining a set to be completed, the set including manufacturing data for a cosmetic composition, and possibly at least one constraint parameter to be respected by the chemical composition, and - application of a technique to the set to be completed and possibly at least one constraint parameter to obtain a completed set, the completed set including manufacturing data for a cosmetic composition likely to meet said at least one acceptability criterion,

[0026] the technique comprising the use of at least one machine learning model, the at least one machine learning model being suitable for completing a set to be completed.

[0027] According to other advantageous aspects of the invention, the generation process comprises one or more of the following features, taken individually or in all technically possible combinations: - the learning model was trained to complete a set to be completed with training on at least one database bringing together cosmetic compositions that meet said at least one acceptability criterion and possibly cosmetic compositions that do not meet said at least one acceptability criterion. - the process also includes a model training phase on a training database during which a merit function is used, the merit function favoring models capable of generating completed sets not part of the training database and respecting at least one acceptability criterion. - manufacturing data is a list of raw materials or ingredients, possibly accompanied by their respective proportions in the cosmetic composition. - during the obtaining stage, at least one constraint parameter is obtained, a constraint parameter being a completeness parameter defining a number of manufacturing data missing in the set to be completed. - during the application phase, the model is used only once, the completeness parameter corresponding to a number of missing data being equal to 1. - a constraint parameter is a global completeness parameter corresponding to a number of missing data greater than 1, the application phase comprising several iterations, each iteration comprising the application of the model to the set generated in the previous iteration with a completeness parameter for said iteration corresponding to a number of missing data equal to 1, the model being applied in the first iteration on the set to be completed. - the model includes an encoder and / or a decoder. - the model is a masked predictive transformer model. - the cosmetic composition is part of a cosmetic product, the manufacturing data being chosen from a list consisting of data relating to the cosmetic composition, data relating to packaging of the cosmetic product and data relating to the manufacturing technique of the cosmetic product used. - the process includes, among other things, a phase of updating the model by retraining the model using the sets used during the interference phase.

[0028] The description also relates to a method for manufacturing a cosmetic composition, the method being implemented by a manufacturing system, the manufacturing method comprising:

[0029] - an implementation phase of a method for generating at least one set of manufacturing data, the generation process being as previously described, and

[0030] - a manufacturing phase of a cosmetic composition from each set generated during the implementation phase.

[0031] The description also describes a computing device configured to generate at least one manufacturing data set, the manufacturing data set comprising manufacturing data for a cosmetic composition and:

[0032] - obtain a set of manufacturing data and one or more data points manufacturing process to be modified in the overall result.

[0033] - calculate a similarity coefficient between each manufacturing data to be modified and a plurality of candidate manufacturing data,

[0034] - select, for each manufacturing data to be modified, at least one data manufacturing data from among the plurality of candidate manufacturing data based on calculated similarity coefficients, to obtain at least one selected data point,

[0035] - generate at least one manufacturing data set, to obtain at least one generated set, each generated set being the obtained set in which each manufacturing data to be modified.

[0036] The description also relates to a system for manufacturing a cosmetic composition, the manufacturing system comprising:

[0037] - a computing device as previously described, and

[0038] - a manufacturing device suitable for manufacturing a cosmetic composition from of the manufacturing data set obtained by the calculation device.

[0039] The description also describes a method for predicting a set of characteristics associated with a list of raw materials, said method being put into

[0040]

[0041]

[0042]

[0043]

[0044]

[0045]

[0046]

[0047]

[0048]

[0049]

[0050] operated by a prediction device and comprising an inference phase, the inference phase comprising a step of: - obtaining a list of properties associated with the list of raw materials, and - application of a technique on the list of properties to obtain the characteristics of the raw material and / or the list of raw materials, the technique involving the use of at least one Bayesian network, said Bayesian network being determined to link characteristics of a raw material to properties. According to other advantageous aspects of the invention, the prediction method comprises one or more of the following features, taken individually or in all technically possible combinations: - the process also includes the application of a function f linking the value of a characteristic to a distribution of a node and to the list of properties according to the following equation: X, = 4 P ( -■ P^ With : • ; value of characteristic i, P( iïp.^ p.*) : probability of node i given the n parent nodes of i. - Arcs and Bayesian network distributions are determined by a group of experts or using known data. - the list of characteristics includes all characteristics whose associated probability distribution is greater than an associated prediction threshold. - the process also includes an update phase during which the user can update the Bayesian network based on expert opinion and data known. - the process involves applying an unsupervised machine learning algorithm to expert opinions and known data to determine the Bayesian network. - a characteristic is a measurable property. - a characteristic is a property that can be evaluated qualitatively. The description also outlines a process for selecting one or more lists of candidate raw materials to replace an initial list of raw materials, comprising the following steps: - determination of a list of properties associated with the first list of raw materials and an associated sensory attribute, - calculation of cosmetic composition characteristics by implementing a prediction process as previously described, - application of a machine learning model to determine a plurality of candidate raw material lists based on the sensory attribute whose characteristics are close to the calculated characteristics, and - selection of one or more lists of raw materials from the plurality of candidate lists of raw materials according to a selection criterion associated with a threshold assigned to each of the characteristics.

[0051] The description also describes a device for predicting a set of characteristics associated with a list of raw materials, said predictive device being configured to - obtain a list of properties associated with the list of raw materials, and - applying a technique to the list of properties to obtain the characteristics of the raw material and / or the list of raw materials,

[0052] the technique comprising the use of at least one Bayesian network, said Bayesian network being determined to link characteristics of a raw material to properties. BRIEF DESCRIPTION OF THE FIGURES

[0053] The invention will become clearer upon reading the following description, given solely by way of non-limiting example, and made with reference to the drawings in which: • [Fig. 1] [Fig. 1] is a schematic representation of a manufacturing system comprising a computing device and a manufacturing device, • [Fig.2] [Fig.2] is a flowchart of an example of the implementation of a process for generating a set of manufacturing data, • [Fig.3] [Fig.3] is a block representation of an example model used by the generation process of [Fig.2], • [Fig.4] [Fig.4] is a block representation of another example of model used by the generation process of [Fig.2], • [Fig. 5] [Fig. 5] is a block representation of yet another example of the model used by the generation process of [Fig.2], • [Fig. 6] [Fig. 6] is a flowchart of an example of the implementation of a another method for generating a set of manufacturing data, • [Fig.7] [Fig.7] is a schematic representation of an example space latent allowing for a better understanding of the process of [Fig.6], • [Fig.8] [Fig.8] is a flowchart of an example of the implementation of a process for manufacturing a cosmetic composition, and • [Fig.9] [Fig.9] is a flowchart of an example of the implementation of a prediction method.

[0054] DETAILED DESCRIPTION OF PREFERRED EMBODIMENTS Description of the manufacturing system

[0055] Figure 1 shows schematically a manufacturing system 10 for a cosmetic product.

[0056] The manufacturing system 10 is suitable for manufacturing the cosmetic product.

[0057] The concept of "cosmetic product" is defined in the section relating to the manufacturing data of a cosmetic product.

[0058] The manufacturing system 10 includes a computing device 12 and a manufacturing device 14.

[0059] The calculation device 12 is suitable for implementing a method for generating a set of manufacturing data enabling the manufacture of a cosmetic product, or at least a part of it, namely at least one cosmetic composition.

[0060] Several examples of generation methods will be described later. These methods are computer-implemented methods.

[0061] Further information on the manufacturing data forming part of the assembly is present in the corresponding section of the description.

[0062] The computing device 12 is a computer comprising one or more electronic components such as: one or more single-core or multi-core processors collectively represented by a processor, one or more graphics processing units (GPUs), a hard drive, random access memory (RAM), a display interface, or an input / output interface. It is desirable that the computer be implemented, for example, in the form of a desktop computer, an in-vehicle computer, a tablet, or a smartphone.

[0063] In the example of [Fig.1], the computing device 12 includes a processing unit 16 comprising and a memory 18 associated with the processing unit 16.

[0064] The processing unit 16 is an electronic circuit designed to manipulate and / or transform data represented by electronic or physical quantities in computer registers and / or memories into other similar data corresponding to physical data in register memories or other types of display devices, transmission devices or storage devices.

[0065] As specific examples, the processing unit 16 is implemented as a programmable logic component, such as an FPGA (Field Programmable Gate Array), or as an integrated circuit, such as an ASIC (Application-Specific Integrated Circuit).

[0066] Alternatively, when the method is implemented in the form of one or more software programs, i.e., in the form of a computer program, also called a computer program product, it is further capable of being stored on a computer-readable medium (not shown). The computer-readable medium is, for example, a medium capable of storing electronic instructions and being connected to a bus of a computer system. By way of example, the readable medium is an optical disc, a magneto-optical disc, ROM, RAM, any type of non-volatile memory (e.g., FLASH or NVRAM), or a magnetic card. A computer program comprising software instructions is then stored on the readable medium.

[0067] The manufacturing device 14 is suitable for manufacturing the cosmetic product from the manufacturing data set generated by the processing unit 16.

[0068] Description of a generation method starting from a set of data to be completed

[0069] The operation of the calculation device 12 is now illustrated with reference to [Fig.2], which is a flowchart illustrating an example of the implementation of a generation process.

[0070] The generation process is a process aimed at generating a set of manufacturing data for a cosmetic composition likely to meet at least one acceptability criterion from a set of manufacturing data to be completed.

[0071] The incomplete nature of the manufacturing data set to be completed is reflected by a completeness parameter forming a contextual input parameter which, as will be seen below, may be implicit or explicit.

[0072] The completeness parameter, reflecting the incomplete nature of the manufacturing data set, can be expressed, in particular, as the number of missing raw materials and / or the total number of raw materials to be included in the completed composition. It is understood that it can alternatively or complementaryly be expressed as a constraint parameter, such as a price, performance, or environmental compatibility constraint, leading to the search for additional raw materials to complete the set. This constraint parameter may be different from the acceptability criterion or may correspond to the acceptability criterion.

[0073] In this description, the term “manufacturing data” refers to data enabling the formulation and / or preparation of the cosmetic product and could also be translated as “preparation data”.

[0074] In the remainder of this description, the set that is generated will be called the "completed set" (satisfying the completeness parameter), and the set that is in entry will be named "set to be completed" (not satisfying the completeness parameter).

[0075] More precisely, in the example that will be described, it is assumed that the set to be completed is a list of raw materials constituting a cosmetic composition which, a priori, does not meet the acceptability criterion. The set to be completed is a list that is implicitly or explicitly missing one or more raw materials. In particular, the set to be completed is a list whose number of constituent raw materials is less than the number of desired raw materials after application of the completeness parameter.

[0076] For example, when the completeness parameter is the total number of expected raw materials, then the total number of raw materials in the composition to be completed is less than the completeness parameter. When the completeness parameter is the number of missing or to-be-added raw materials, then said completeness parameter added to the number of raw materials in the composition to be completed corresponds to the number of raw materials in the complete composition sought and to be generated by the model described below. The completeness parameter can also be a range of values, allowing the model to add a number of raw materials between a lower and an upper bound. It is also possible to specify only a lower or upper bound. If necessary, a default value can also be provided.

[0077] Furthermore, it is assumed that the completed set is a completed list of raw materials, capable of meeting at least one acceptability criterion, it being understood that this example can be transposed to any type of manufacturing data as described in the corresponding section, namely the section entitled "Description of the different conceivable manufacturing data sets".

[0078] For example, the set to be completed may be an empty set (without raw material).

[0079] The process therefore aims to generate the data enabling the manufacture of a cosmetic composition comprising, in a physiologically acceptable environment, a plurality of raw materials.

[0080] A raw material is a set of one or more ingredients used in the manufacture of cosmetic products. These raw materials can be of natural or synthetic origin and play various roles in cosmetic formulations, ranging from active agents to texturizing agents. For example, a raw material could be glycerin, elastin, or hyaluronic acid. In the remainder of this description, a raw material is defined by the set of ingredients that compose it, a set of molecular structures of the ingredients that compose it, a set of concentrations of the ingredients that compose it, a set of properties of the ingredients that compose it or a set of characteristics.

[0081] By "physiologically acceptable" is meant a medium compatible with keratinous materials. By "keratinous materials" is meant skin, mucous membranes, and / or hair and nails. Preferably, the keratinous materials are skin, particularly facial skin, mucous membranes such as the lips, and / or hair and nails such as eyelashes.

[0082] By way of non-limiting example, such a cosmetic composition is a hair treatment or care product (e.g. a shampoo), a skin treatment or care product (e.g. a moisturizer, a sunscreen, an anti-aging cream), a makeup product (e.g. a lipstick, a mascara, a foundation, or a gloss).

[0083] In the example described, the generation process includes a learning phase P20, an inference phase P30 and an update phase P40.

[0084] This is only a simple example, it being understood that in general the inference phase P30 and update phase P40 are implemented successively to continually improve the model that is used.

[0085] It can also be noted that the implementation of the P40 update phase is not mandatory.

[0086] The learning phase P20 can be carried out offline, i.e. by a computer different from the computing device 12, and preferably prior to the use of the computing device 12.

[0087] During the learning phase P20, a machine learning model M is trained via a machine learning algorithm A to complete a set to be completed according to a completeness parameter defining, directly or indirectly, a number of manufacturing data missing in the set to be completed.

[0088] A machine learning algorithm A (commonly called a machine learning algorithm) is an algorithm that teaches / trains a model M to automatically find patterns and / or general input-output relationships from training (or learning) data. After training, the model M could make similar inferences and / or predictions. In a mathematical sense, the model M could be considered as an equation that maps the inputs to the output. Machine learning algorithms can be classified into several categories according to the types of tasks to be performed (classification, regression, generation), the learning approach used (supervised, unsupervised, semi-supervised, self-supervised, and reinforcement learning), and the algorithmic theory (neural network, tree model, probabilistic model, polynomial model, Bayesian model, SVM, etc.).

[0089] The learning algorithm applied in this process relates to self-supervised and reinforced learning for a generation task. It is based on neural network theory, or more precisely on the Transformer for training and inference / generation theory.

[0090] The model M at the output of the learning phase is a Transformer-type neural network.

[0091] A neural network comprises an ordered succession of layers of neurons, each of which takes its inputs from the outputs of the previous layer.

[0092] More precisely, each layer comprises neurons taking their inputs from the outputs of the neurons of the previous layer, or from the input variables for the first layer.

[0093] Alternatively, more complex neural network structures can be envisaged with a layer that can be linked to a layer further away than the immediately preceding layer.

[0094] Each neuron is also associated with an operation, that is to say a type of processing, to be carried out by said neuron within the corresponding processing layer.

[0095] Each layer is connected to the other layers by a plurality of synapses. A synaptic weight is associated with each synapse, and each synapse forms a link between two neurons. It is often a real number, which takes on both positive and negative values. In some cases, the synaptic weight is a complex number.

[0096] Each neuron is designed to perform a weighted sum of the value(s) received from the neurons of the preceding layer, each value being multiplied by the respective synaptic weight of each synapse, or connection, between said neuron and the neurons of the preceding layer, and then to apply an activation function, typically a non-linear function, to said weighted sum, and to deliver at the output of said neuron, in particular to the neurons of the next layer connected to it, the value resulting from the application of the activation function. The activation function introduces non-linearity into the processing performed by each neuron. The sigmoid function, the hyperbolic tangent function, and the Heaviside function are examples of activation functions.

[0097] As an optional complement, each neuron is also capable of applying, in addition, a multiplicative factor, also called bias, to the output of the activation function, and the value delivered at the output of said neuron is then the product of the bias value and the value from the activation function.

[0098] A convolutional neural network is also sometimes called a convolutional neural network or by the acronym CNN, which refers to the English term "#Convolutional Neural Networks#".

[0099] In a convolutional neural network, each neuron in the same layer exhibits exactly the same connection pattern as its neighboring neurons, but at different input positions. The connection pattern is called the convolutional kernel or, more often, "kemel" in reference to the corresponding English term.

[0100] A fully connected layer of neurons is a layer in which the neurons of said layer are each connected to all the neurons of the preceding layer.

[0101] Such a type of layer is more often referred to by the English term "#fully connected#", and sometimes designated by the name "#dense layer#".

[0102] Alternatively, model M applies linear regression, logistic regression, a decision tree, a random forest, principal component analysis (PCA), a Naive Bayes algorithm or a K-nearest neighbors (KNN) algorithm

[0103] Alternatively, the M model is a support vector machine or a K-Means Clustering model.

[0104] Herein, "#machine learning#" means that the model M is learned to perform a task using a learning algorithm A run on a computer and applied to training data. The model M learns to perform a task through learning algorithms A, which are an optimization process with respect to a well-defined metric. For example, optimization could be minimizing the least-squares error between predicted and actually measured values, or maximizing the reward in the case of reinforcement learning, on the training data.

[0105] The task to be performed here is the generation of the completed set of raw materials from a set to be completed or a subset of the completed set.

[0106] Model M can thus be seen as a generative or predictive model.

[0107] It is also assumed that a training database has been established for this purpose. This training database contains examples of data that the M model can read repeatedly to understand the implicit patterns or relationships present in the data.

[0108] The generation of such a database varies depending on the type of Transformer and the learning approach adopted.

[0109] For a Transformer that contains only the encoder and is learned through a learning approach called "MLM" (Masking Language Modeling), each training example consists of a list to be completed in which some raw materials are masked and another completed list in which all raw materials are exposed. Each data pair in the database can be generated from a real cosmetic composition whose list of raw materials The list is known and one or more raw materials are masked. A real cosmetic composition is implicitly considered chemically valid and minimally complete. When training a Transformer of the encoder type, the list to be completed is submitted as input to model M. The information from the input data is propagated through the different layers of the model for processing until it reaches the output layer. The output layer predicts the raw materials masked at the input. The predicted raw materials are compared with the corresponding raw materials in the completed list, calculating an error. This error is used to readjust the parameters in model M via the backpropagation mechanism to reduce the error. This mechanism is performed repeatedly to progressively reduce the overall error in the prediction of the masked raw materials.

[0110] For a Transformer containing only a decoder, each training example consists of a completed list of raw materials in a cosmetic composition with a 'START' token added to the beginning of the list and the same list completed with an 'END' token added to the end. Each data pair in the database can be generated from a real cosmetic composition whose list of raw materials is known and by adding the 'START' token to the beginning of the list or the 'END' token to the end of the list, respectively. A real cosmetic composition is considered chemically valid and minimally completed. When training such a model containing only a decoder, the list containing the 'START' token will be injected into a self-attention layer via the 'VALUE', 'KEY', and 'QUERY' inputs.The self-attention layers encode the information in the list and propagate it to the output layer to predict the list containing the token 'END'. The weights in the self-attention layers, which are recalculated for each token in the list containing the token 'START', will be masked by a so-called 'CAUSAL' mask. This 'CAUSAL' masking could mimic an autoregressive generation mechanism. The prediction error is calculated by comparing the sequence of raw materials generated in this autoregressive way with the subset list containing the token 'END'. The error will be used to readjust the parameters in the M model via the backpropagation mechanism to reduce this error. This mechanism is performed repeatedly to progressively reduce the overall error in the prediction of the masked raw materials.

[0111] For a Transformer containing an encoder and a decoder, each training example consists of a list to be completed from a subset of raw materials removed from a completed list of cosmetic composition, and two lists containing a complementary subset of raw materials in the The same list is completed with cosmetic compositions. One of the supplementary lists has an additional 'START' token added to the beginning of the list, and the other has an additional 'END' token added to the end of the list. Each triplet of training data can be generated by splitting a completed list of real cosmetic compositions and adding the 'START' token to the beginning of the supplementary list or the 'END' token to the end of the supplementary list, respectively. A real cosmetic composition is implicitly considered chemically valid and minimally completed. When training a Transformer containing an encoder and a decoder, the list to be completed for a subset of raw materials is fed to an encoder, which encodes it into a vector representing all the essential information for the prediction sequence. This encoded vector is then fed into the decoder via the 'KEY' and 'VALUE' inputs of the second self-attention layer.The supplementary list containing the token 'START' will be injected into the same self-attention layer via the 'QUERY' input, after being encoded by a previous self-attention layer. The information encoded by the second self-attention layer will be propagated through the remaining layers to the output layer to predict the supplementary list containing the token 'END'. The weights in the self-attention layers, which are recalculated for each token in the list containing the token 'START', will be masked by a so-called 'CAUSAL' mask. This 'CAUSAL' masking could mimic an autoregressive generation mechanism. The prediction error is calculated by comparing the sequence of raw materials generated in this autoregressive manner with the subset list containing the token 'END'. The error will be used to readjust the parameters in the M model via the backpropagation mechanism to reduce this error.This mechanism is performed repeatedly in order to progressively reduce the overall error in predicting hidden raw materials.

[0112] Through this simple example, it is clear that it is possible to generate a large database since a single composition can generate a plurality of pairs for training.

[0113] Any training technique can then be used to obtain the model.

[0114] For example, learning is supervised learning or unsupervised learning or self-supervised learning that has been applied in the process.

[0115] Other techniques are conceivable such as reinforcement learning techniques used in the process to encourage the model to propose manufacturing datasets enabling the manufacture of a cosmetic product not part of the training database and respecting at least one acceptability criterion.

[0116] According to one embodiment, a merit function (also called a loss function) is used, the merit function favoring models capable of generating manufacturing datasets deemed "complete" allowing the manufacture of a cosmetic product not part of the database and meeting at least one acceptability criterion.

[0117] More specifically, learning is achieved by estimating weights or scores associated with a reward function.

[0118] In this example, the generation of known data is neutral with respect to the reward function.

[0119] On the other hand, the generation of data that is part of a list of data to be excluded results in a negative weight or score, while the generation of unknown data results in a positive weight or score.

[0120] By taking an average, the model M determines an associated acceptability criterion allowing to validate or not a future generation.

[0121] For other types of M models such as those based on the K-nearest neighbors (KNN) algorithm, training includes storing training data, calculating Euclidean distances between test points and all training points, selecting nearest neighbors and predicting by classification or regression (see especially [Fig.6]).

[0122] During the inference phase P30, the computing device 12 applies the learned model to input data comprising a composition to be completed and a completeness parameter.

[0123] The completeness parameter can be explicit, for example by directly providing the total expected number of raw materials or the number of missing raw materials to be added. The completeness parameter can also be defined by default or implicitly and determined from the data of the composition to be completed by using a masking marker or token in the data of said composition to be completed. According to a particular embodiment, the completeness parameter is inferred from an additional constraint parameter, such as a price constraint, for example.

[0124] According to the example described, the inference phase P30 comprises a getting step E301 and an applying step E302.

[0125] During the E301 obtaining step, the calculation device 12 obtains a set to be completed of manufacturing data and the completeness parameter.

[0126] For example, the calculation device 12 reads this list and the completeness parameter from memory 18.

[0127] Furthermore, in certain embodiments, the user can enter a completed list and select at least one raw material to exclude, which will associated with a MASK token subsequently. In such a case, the number of MASK tokens determines the completeness parameter (exact, minimum or maximum number of raw materials to add e.g.).

[0128] During the application step E302, the computing device 12 applies a technique to the set to be completed to obtain a completed set. Optionally, this is associated with a probability that the completed set satisfies the model's acceptability criterion. Completed sets whose probability of satisfying the criterion is greater than a predetermined threshold (>0.8, for example) will be presented to the user.

[0129] The technique here consists of applying the model M to the set to be completed in order to obtain a set completed according to the completeness parameter and the acceptability criterion.

[0130] Several topologies for the model M are conceivable.

[0131] According to a first example, the M model is applied only once.

[0132] In this example, the model M includes at least part of a transformer, more precisely the encoder.

[0133] As a particular example, the M model uses a BERT transformer model (which refers to the corresponding English name of "Bidirectional Encoder Representations from Transformers" literally meaning "bidirectional representations of encoders from transformers").

[0134] The MPT model is a neural network comprising several layers represented in [Fig.3].

[0135] In general, the M model has several layers, among which two sets of layers play an interesting role.

[0136] The first set is a multi-head attention set. This first set allows the model to estimate the connection between the input elements and to extract relationships between nearby and distant elements with the same efficiency.

[0137] The second set is a direct-action point-to-point neural network.

[0138] Such a neural network is more often referred to by the corresponding English name "Position-Wise Feed-Forward Network".

[0139] The second set here serves to help change the representation and to capture more information about the context. A particular example of such a model example is now described with reference to [Fig.3].

[0140] In the example shown, the model M has 4 encoders. Each encoder has an "#input embedding#" layer, a "#multi-head attention#" layer which includes two "#addition / normalisation#" sub-layers, a "#feed forward#" layer and a "#classification layer#" layer.

[0141] The “#input embedding#” layer or “#entry vectors#” in French is a layer responsible for converting the presented tokens into integer identifiers in a continuous vector format.

[0142] The "#multi-head attention#" layer continues to encode the various previously generated tokens with weighting factors (also called attention weights). These weights are calculated for each token and represent implicit relational information between the tokens with respect to the final prediction.

[0143] The "#addition / normalization#" layers consist of concatenating each original token (before passing through the "#multi-head attention#" layer) at the output of each multi-head attention sublayer, thereby enriching the number of informative features used for the final prediction. Then, normalization is performed to stabilize the learning process.

[0144] The feed-forward or direct-action layer is composed of several sublayers of neurons, each of which is interconnected. The neurons in each of these sublayers receive weighted inputs from the neurons in the preceding layer and transmit their outputs to the following layers. At each activated layer, a dot product is applied between the weights associated with each neuron and their inputs before the application of an activation function.

[0145] The entire training of such a neural network consists of making the values ​​of said weights converge in order to match the theoretical outputs of the training data with the actual outputs.

[0146] The outputs of the “#feed forward#” layer are “#contextualized embedding#” or “#contextualized vectors#”.

[0147] The "#classification layer#" transforms the normalized contextualized vectors concatenated with the tokens entering the feedforward layer into task-specific predictions. Typically, such predictions are text classification, named entity recognition, or answering questions. In our example, these are raw materials for manufacturing a cosmetic composition.

[0148] According to a second example, the model M is applied several times with a reinjection into input of all or part of the completed sets obtained at the output of the model M. The reinjection is possibly accompanied by an update of the completeness parameter (addition of a hidden raw material in the reinjected data for an implicit determination of the updated completeness parameter or explicit update of the completeness parameter).

[0149] This means that in a first calculation iteration, the model M is applied to the set to be completed and gives as output a first intermediate set then, at a second iteration, the model is applied to the first intermediate set to give as output a second intermediate set and so on until a final iteration during which the application of model M makes it possible to obtain the completed set.

[0150] According to one embodiment, the number of iterations of the application corresponds to the number of manufacturing data missing in the set to be completed.

[0151] In such a case, at each iteration, the model M adds a manufacturing data point to the set given as input to the model M at each iteration. According to one embodiment, the overall completeness parameter is greater than 1 missing raw material, for example, 3 missing raw materials, with each iteration using an intermediate completeness parameter corresponding to 1 missing raw material. Thus, after 3 iterations, the complete composition is obtained.

[0152] In such a case, the model M is a masked predictive transformer model.

[0153] Such a model is often referred to by the abbreviation MPT, which refers to the corresponding English name "#Masked Predictive Transformer#".

[0154] Model M is a neural network comprising several layers represented in [Fig.4].

[0155] In the example shown, the model M includes one “#input embedding#” layer, two “#multi-head attention#” layers, three “#addition / normalisation#” layers, one “#feed forward#” layer, one “#linear#” layer, one “#softmax#” layer and one “#argmax#” layer.

[0156] The linear layer is a layer used in neural networks that applies a linear transformation (or matrix multiplication with a transformation matrix) to the input data using weights and biases. Applying this linear layer transforms the received input data into either a lower dimensionality for dimensionality reduction or a higher dimensionality for more complex tasks.

[0157] The softmax layer is a layer used in neural networks that applies a softmax function (or a normalized exponential function) to real-number input data in order to convert it into a format that could represent the probability distribution over a number of choices. It is useful for assigning a probability to each possible answer in prediction.

[0158] The argmax layer always follows the softmax layer. The application of this layer is simply to extract the element with the highest value in the probability vector generated by the softmax layer. This element will be taken as the final answer for the prediction.

[0159] According to a third example, the model M is applied several times with a reinjection at the input of the second sub-model of the output of the model M.

[0160] This means that in a first calculation iteration, the model M is applied to the set to be completed and gives at least a first intermediate set as output, then, in a second iteration, the model is applied to one of the first intermediate sets to give at least a second intermediate set as output, and so on until a final iteration during which the application of the model M makes it possible to obtain the completed set.

[0161] According to one embodiment, the number of iterations of the application corresponds to the number of manufacturing data missing in the set to be completed.

[0162] In such a case, at each iteration, the model M adds a manufacturing data to the set given as input to the model M at each iteration.

[0163] In such a case, the M model is a masked predictive transformer model.

[0164] Such a model is often referred to by the abbreviation MPT, which refers to the corresponding English name "#Masked Predictive Transformer#".

[0165] Model M is a neural network comprising several layers represented in [Fig.5].

[0166] Model M comprises a first sub-model and a second sub-model.

[0167] The first submodel includes an "input embedding#" layer, a layer “#multi-head attention#”, two layers “#addition / normalization#” and one layer of “#feed forward#”.

[0168] The output of the first sub-model is then applied to the input of a second layer "#multi head attention#" of the second sub-model comprising a layer "#input embedding#", a first layer "#multi-head attention#", three layers "#addition / normalisation#", the second layer "#multi head attention#", a layer of "#feed forward#", a layer "#linear#", a layer "#softmax#" and a layer "#argmax#".

[0169] During the P40 update phase, the completed set obtained is, where appropriate after verification that it meets or does not meet the acceptability criterion, integrated into training data and the M model is retrained from the training data obtained.

[0170] For example, the weights associated with each sub-layer of the "#feedforward#" layer are determined by the computing device 12 in order to match the model outputs to the training data.

[0171] The generation process thus makes it possible to obtain a complete set.

[0172] Through these examples, it appears that the machine learning model can also take into account one or more additional input parameters to determine the acceptability criterion or criteria.

[0173] These input parameters are constraint parameters to be respected by the cosmetic composition (and therefore indirectly by the completed assembly).

[0174] As an example of additional input data, the number of missing raw materials may be cited, in which case the constraint parameter is a completeness parameter or a quantification of a performance that the cosmetic product should achieve.

[0175] In the case where the constraint parameter is a completeness parameter, it may be envisaged to implement the previous embodiments as follows.

[0176] During the application phase, the model is used only once, the completeness parameter corresponding to a number of missing data equal to 1.

[0177] In the iterative embodiment, a constraint parameter is a global completeness parameter corresponding to a number of missing data greater than 1 and the application phase comprises several iterations, each iteration comprising the application of the model to the set generated in the previous iteration with a completeness parameter for said iteration corresponding to a number of missing data equal to 1, the model being applied in the first iteration on the set to be completed.

[0178] In the case where the constraint parameter is a quantification of a performance that the cosmetic product should achieve, this quantification makes it possible to define an acceptability criterion.

[0179] More generally, the acceptability criterion is deduced from the available information, either solely from the constraint parameter(s), particularly when the assembly to be completed is empty (without raw material), or from the assembly to be completed, particularly in the absence of constraint parameters, or from both, i.e. from the raw materials of the assembly to be completed and the parameters.

[0180] In an extreme case, the acceptability criterion can thus be reduced to a formula that is biologically acceptable for the part of the body intended to receive it.

[0181] Where appropriate, the term “raw material” may include packaging or an application device for the cosmetic composition, the packaging and the device then being intended to form, together, a cosmetic product.

[0182] Thus, based on the same principle described above, it is possible to provide a model with the data of a cosmetic composition considered incomplete, not in relation to a missing chemical raw material, but in relation to a missing application organ or packaging to constitute a final cosmetic product, this final cosmetic product constituting the completed cosmetic composition in the sense of the embodiments described above.

[0183] It is of course possible to combine these acceptability criteria to generate the most satisfactory complete set possible, the model being trained accordingly in accordance with the acceptability criteria to be targeted.

[0184] Description of a method for generating from a set of data to be modified

[0185] The operation of the calculation device 12 is now illustrated with reference to [Fig.6], which is a flowchart illustrating another example of the implementation of another generation process.

[0186] The generation process according to [Fig.6] comprises a determination phase P50 and a generation phase P60.

[0187] During the P50 determination phase, a list of properties and / or molecular structures associated with each of the raw materials is determined.

[0188] During the P60 generation phase, the computing device 10 obtains one or more sets to complete.

[0189] The generation phase P60 comprises a obtaining step E601, a calculation step E602, a selection step E603 and a generation step E604.

[0190] During the obtaining step E601, the calculation device 12 obtains a list of raw materials and one or more raw materials to be modified in the resulting set.

[0191] For example, the user enters the list of raw materials to be modified via the input interface which the calculation device 12 reads from memory 18.

[0192] During the calculation step E602, the calculation device 12 calculates a similarity coefficient between each raw material to be modified and a plurality of candidate raw materials.

[0193] According to a particular example, each raw material is associated with several property values.

[0194] The term property is to be understood here in a broad sense as including physically measurable properties but also qualitative properties that can be assessed on an appropriate rating scale.

[0195] More specifically, the term property refers to physical, chemical, functional or biological properties.

[0196] For example, a physical property defines the structure of the raw material as its state, solubility or viscosity.

[0197] For example, a chemical property chemically defines the raw material through its pH, its associated carbon chain length or its oxidation.

[0198] For example, a functional property defines different practical capabilities of the raw material such as its moisturizing, anti-aging or emollient capacity.

[0199] For example, a biological property defines a biological capacity of the raw material such as its antibacterial, healing or soothing capacity.

[0200] Each value is a quantification of a respective property of the raw material, so that the supply of all the values ​​of the properties of a

[0201]

[0202]

[0203]

[0204]

[0205] raw material can be seen as the provision of a signature of the raw material. The calculated similarity coefficient then depends on at least one value of a property of the raw material. Several examples of such a calculation can be given. In an embodiment with quantifiable properties, calculator 12 applies the following equation: C(X^^j, Xcan(j 1 - p—^pV^sp(x,mdi, xcand, j ) ? J With : • C(A, B) denotes the similarity coefficient between material A and material B, • xmodj denotes the i-th raw material to be modified, where i is a non-zero integer. • Xcand j denotes the j-th candidate raw material, j being a non-zero integer, • w denotes the sum of the weights applied to the p properties • wp denotes the weight applied to the p-th property (Wp = 1 by default) • n is the number of properties,

[0206] ^P ( xnwdlp xcaiul, j ) — RP

[0207]

[0208]

[0209]

[0210]

[0211]

[0212] • ^modip: value of the p-th property of xmodj, where p is an integer between 1 and n, • ycand, j : value of the p-th property of xetmd, j, • Rp: this value is at least equal to the difference between the 5th and 95th percentile of the distribution of values ​​of the p-th property. • If the value of the property is qualitatively measurable: SP (Xmodj~ xcand, j) 0 SI xmodlp xcand, j •> SÙtOn According to an example, the number of properties n is equal to the number of different properties of the raw materials of the given set during the obtaining step E601. According to one variant, the calculation step E602 is implemented differently. In this variant, the calculation step E602 includes a search step and a calculation step. During the search step, the computing device 12 searches for a plurality of reference lists. Typically, a reference list consists of sets that may or may not have undergone changes in their associated raw material lists.

[0213] Reference lists are kept in a database via a distributed and scalable data structure (SDDS, from the English Shared Domain Data Set) allowing a large amount of data to be stored by defining associated characteristics such as the source of said data.

[0214] Typically, such a structure allows more than one million reference lists to be stored in the database.

[0215] The reference lists concern cosmetic compositions that may have been modified in the past. Thus, each list is associated with a plurality of lists of raw materials.

[0216] In the example described, the reference list has N iterations and therefore N associated lists of distinct raw materials.

[0217] In addition, the reference list includes the raw material to be modified, but all separate raw material lists generated after the J-th iteration no longer include it (J being smaller than N).

[0218] For example, the search step is performed from an SQL query applied to the database.

[0219] Thus, the calculation device 12 identifies the NJ lists of separate associated materials not including the raw material to be modified.

[0220] During calculation step E602, the calculation device 12 calculates a similarity coefficient between the NJ lists of separate raw materials associated with and the J lists of raw materials containing the raw material to be modified.

[0221] For example, the calculation device 12 applies a function f such that: [° 222 1

[0223] With: • C( A, B) denotes the similarity coefficient between the set to be completed and the B-th list of associated distinct materials (j < g < N), • B; number of iterations, • X: number of raw materials not shared between the two lists.

[0224] Alternatively, calculation step E602 is implemented differently.

[0225] In this variant, the calculation step E602 comprises a conversion substep, a first calculation substep, a projection substep and a second calculation substep.

[0226] During the conversion substep, the molecular structure of the raw material is converted into molecular identifiers.

[0227] For example, such identifiers take the form of SMILES, InchiKey or Molecular fingerprints structures.

[0228] During the first calculation substep, the calculation device 12 calculates a plurality of molecular properties from the molecular identifiers.

[0229] For example, molecular properties include molar mass, partition coefficient or number of chemical bonds.

[0230] During the projection substep, the computing device 12 identifies the molecular properties linked together in order to represent them in a reduced-dimensional projection space (see [Fig.7]).

[0231] For example, the computing device 12 reduces the dimensions of the projection space to a minimum by identifying the dimensions related to molecular properties as linear compositions of other properties.

[0232] Typically, the computing device 12 can apply a principal component analysis PCA of such a step.

[0233] Principal component analysis is a statistical analysis method that allows the information contained in a set of functional data to be summarized, thus allowing the production of so-called principal component functions of minimum dimension which reproduce a maximum of information on the data studied.

[0234] In other words, functional principal component analysis is a projection of input information into an optimal projection basis.

[0235] Finally, in the second calculation step, the calculation device 12 calculates the similarity coefficient by applying a clustering algorithm in the M-dimensional projection space and calculating the similarity coefficient as a function of a distance criterion. For example, the similarity coefficient is zero for two raw materials that are not in the same cluster.

[0236] Advantageously, such a characteristic makes it possible to replace at least one raw material to be excluded with a new raw material providing one or more priority properties for the operator.

[0237] Alternatively, a method may be used to predict the characteristics of the raw material based on said properties. This device is described in the final section of this application.

[0238] During the selection step E603, the calculation device 12 selects, for each raw material to be modified, at least one raw material from among the plurality of candidate raw materials.

[0239] Each raw material thus selected is referred to as the selected raw material in the following.

[0240] The selection is carried out according to the similarity coefficients calculated during the calculation step E602.

[0241] According to a simple example, the selected raw materials are the raw materials from the plurality of candidate raw materials having the highest similarity coefficients.

[0242] In other examples, raw materials are selected based on similarity coefficients and a set of criteria associated with each of the candidate raw materials.

[0243] Typically, the set of criteria includes an economic indicator or an indicator of environmental friendliness such as biodegradability.

[0244] During generation step E604, the computing device 12 generates all the raw material sets in which each raw material to be modified is replaced by a respective selected raw material.

[0245] The generation process thus makes it possible to obtain one or more sets of completed raw materials.

[0246] This makes it possible in particular to exclude a raw material by replacing it with one or more others or even to remove it in order to obtain a new cosmetic composition with the desired properties.

[0247] Such a process then makes it possible to generate manufacturing data useful for the preparation and manufacture of innovative cosmetic compositions adapted to specific constraints.

[0248] Advantageously, the implementation of such a process is easy, in particular relatively quick.

[0249] In addition, this process also allows access to formulation, improvement or replacement levers for any person manufacturing cosmetic products from beginner to experienced or even expert.

[0250] Applications of the previous generation processes

[0251] The assembly obtained by either of the preceding generation processes is useful for many applications.

[0252] According to the example described, it can be used in a process for preparing and manufacturing a cosmetic product by manufacturing system 14.

[0253] The manufacturing process includes an implementation phase P70 and a manufacturing phase P80.

[0254] During the implementation phase P70, the computing device 12 implements one or the other of the preceding generation processes.

[0255] For example, the computing device 12 implements the inference phase P40 of the generation process according to the embodiment of [Fig.2] or the generation phase P60 of the generation process according to the embodiment of [Fig.6].

[0256] The manufacturing phase P80 aims to manufacture the cosmetic product corresponding to the assembly obtained after implementation during the implementation phase P70.

[0257] For example, the manufacturing device 14 takes as input the generated set and controls the elements that compose it so that these elements carry out the manufacture of the cosmetic product corresponding to the set generated by the generation process.

[0258] Other manufacturing processes are conceivable.

[0259] For example, during a configuration step, the operator can select packaging and an embodiment to be respected in the form of data and include them in the manufacturing data.

[0260] The manufacturing technique refers to a set of criteria to be met during the manufacture of the cosmetic composition, such as the order of incorporation, temperature, mixing elements (speed, time, turbines and / or blades), phase separation, or premixing. The manufacturing technique can notably be expressed in terms of unit process operations.

[0261] Description of the different possible manufacturing data sets

[0262] The processes for generating Figures 2 and 6 have so far been described for a particular example of a list of raw materials for illustrative purposes.

[0263] However, in a general case, manufacturing data can be of any type.

[0264] Depending on the case, manufacturing data are chosen from a list consisting of data relating to composition, data relating to packaging, in particular to an application organ of the cosmetic composition, and data relating to the manufacturing technique used.

[0265] The manufacturing data allows, depending on the case, the obtaining of a cosmetic composition or a cosmetic product.

[0266] A cosmetic composition is composed of a list of raw materials attributing a list of properties to said composition.

[0267] The term "cosmetic product" means the entirety of a cosmetic composition, packaging, or conditioning unit, including in particular an application device.

[0268] The manufacturing process refers to a set of unit operations to be respected during the manufacture of the cosmetic composition such as an order of incorporation, temperature, humidity or applied force.

[0269] Packaging means the means of preserving the cosmetic composition. Alternatively, the packaging includes an application tool or element for the associated cosmetic composition, such as an eyelash brush for mascara or a brush for foundation.

[0270] A data relating to composition is a data representative of the chemical composition of the formula.

[0271] The list of raw materials or ingredients is a particular example of data relating to composition.

[0272] Another example of data relating to composition may be the concentration of the raw material in said composition.

[0273] The representation of the raw material may differ depending on the embodiment. For example, the raw material may be represented by a chemical formula, an identifier, a trade name or a type (emulsifier or fat) and its proportion in the composition may be expressed as a mass or volume percentage, detailed linearly or with preparations.

[0274] Instead of a raw material, it is possible to consider ingredients, which are part of the raw materials, the ingredients being similarly able to be represented in different ways (formula, identifier or others).

[0275] The raw material or ingredient content is another example of data relating to composition.

[0276] Data relating to the manufacturing technique used is data enabling the characterization of the steps of a manufacturing process.

[0277] An example of data relating to the manufacturing technique is a sequence of unit process operations.

[0278] An example of data relating to the manufacturing technique is the order in which the ingredients or raw material are inserted into the manufacturing process, the temperature at which the ingredients are placed in the vat, the times of introduction and mixing, and the speed and mechanical forces applied.

[0279] Description of a method for predicting characteristics

[0280] As explained previously, the preceding generation processes use knowledge about lists of raw materials. It is therefore useful to be able to estimate or predict the characteristics of a raw material and / or a list of raw materials.

[0281] For this purpose, a prediction process can be implemented via a prediction device.

[0282] The same remarks as for the calculation device apply here to the prediction device and are therefore not repeated.

[0283] An example of the implementation of the method for predicting the characteristics of a list of raw materials is now described

[0284] The prediction process is a process aimed at estimating a set of characteristics associated with a raw material.

[0285] Alternatively, the process estimates a set of characteristics associated with a plurality of raw materials forming a cosmetic composition. Thus, in the following description, a raw material also refers to a set of raw materials associated with a list of properties previously determined from the property lists of the set of raw materials.

[0286] The term characteristic is to be understood here in a broad sense as including measurable characteristics but also qualitative characteristics that can be assessed appropriately, for example on a rating scale according to defined evaluation criteria.

[0287] Characteristics are information associated with the raw material such as functions provided, technical characteristics, physicochemical properties or contributions to sensory benefits.

[0288] The characteristics of a raw material are defined by all the technical effects induced by the properties of the latter.

[0289] For example, a characteristic of a raw material is its stickiness, its resistance to acid or its ability to stabilize the oily phase.

[0290] Such a list of characteristics also includes a value associated with each of the characteristics which may be quantitative or qualitative.

[0291] The prediction process according to [Fig.9] comprises a determination phase P90 and a prediction phase P100.

[0292] During the P90 determination phase, a list of properties associated with the raw material is determined.

[0293] The prediction phase P100 comprises a obtaining step E1001, a calculation step E1002 and a prediction step E1003.

[0294] During the E1001 retrieval step, the computing device 12 obtains a list of properties.

[0295] For example, the user manually enters the list of properties of the raw material via the input interface which the computing device 12 reads from memory 18. The list of properties of the raw material can also be retrieved automatically by the system, in particular from a suitable storage database.

[0296] In another example, the computing device 12 receives the list of properties of the raw material by an electronic device via radio communication or wired connection.

[0297] In addition, the user enters values ​​associated with each of these properties. As with the list of properties, the associated values ​​can also be automatically retrieved from a database.

[0298] During the calculation step E1002, the calculation device 12 calculates the feature list using the property list and a causal Bayesian network.

[0299] A Bayesian network is a model that allows determining an output value or a distribution of output values ​​based on a plurality of input data that influence (or not) said output value.

[0300] A Bayesian network is a probabilistic graphical model that represents a set of random variables and their conditional dependencies via a directed acyclic graph. It is used to model uncertain knowledge in various fields such as the relationships between characteristics and properties of a cosmetic composition.

[0301] To create such a Bayesian network, it is necessary to determine, prior to use, a set of variables representing the nodes of the graph, a set of arcs linking the nodes representing conditional dependencies and probabilities for each of the nodes.

[0302] Each node corresponds to a characteristic whose value then depends on the network inputs (values ​​of the properties entered by the user). Each node is associated with a parent that influences its probability distribution.

[0303] The inputs (properties) are also represented by nodes without a parent.

[0304] The arcs and the probabilities associated with each of the nodes enable this dependence. Indeed, the arcs and the probabilities are determined by the expertise of a group of experts and represent the dependencies between characteristics and properties; these are called elicited probabilities.

[0305] Alternatively, the arcs and probabilities are determined from known data. Typically, known data are data linking a property to a historically known or recently discovered feature.

[0306] The values ​​assigned to each of the probabilities are modifiable over time, for example if new knowledge allows the probabilities determined by the group of experts to be refined. Thus, an arc linking a property to a characteristic means that the property influences that characteristic.

[0307] An intermediate node is present on each of the arcs allowing the weight (influence) of the property to be represented on said characteristic.

[0308] In addition, intermediate nodes can be used to represent a cumulative effect of several properties on a characteristic linking several weights together.

[0309] For example, the feature list includes all features whose associated probability distribution is greater than an associated prediction threshold.

[0310] During the prediction step E1003, the prediction device predicts the list of features and their associated values.

[0311] For example, the prediction device applies a function f linking the distribution to a value of the characteristic and to the properties such that: [03121 = ÜPie -

[0313] With: • Xj; value of characteristic i, p{ i\pu .... PM): probability distribution of node i given the n parent nodes of i,

[0314] In one embodiment, during an optional update phase, it is possible to update the Bayesian network via the prediction device.

[0315] The term "update" means the addition or modification of associated nodes, probabilities and arcs.

[0316] For example, the user enters new nodes, new arcs and new associated probabilities via the prediction device and from the expertise of expert groups.

[0317] Alternatively, the Bayesian network, the new nodes, the new arcs and the new probabilities are determined by applying one or more unsupervised learning algorithms to the known data.

[0318] The unsupervised learning algorithm(s) are part of the group: clustering algorithms, dimensionality reduction algorithm, density modeling algorithm or neural network algorithm.

[0319] Such unsupervised learning algorithms are presented in the article entitled "Learning Bayesian Networks with the bnleam R Package" published on July 10, 2010.

[0320] Advantageously, this "hybrid" approach between expert opinion and the result of one or more learning algorithms makes it possible to increase the accuracy and the sustainability over time of the results of the causal Bayesian model.

[0321] Advantageously, it is possible to obtain the property list of a raw material from these characteristics by tracing back the graph of the Bayesian network.

[0322] The term “remonter” means “determine” and is carried out by applying an optimization algorithm or Bayes’ theorem.

[0323] Typically, the optimization algorithm is a genetic algorithm.

[0324] The prediction process can be used in a process selecting one or more lists of raw materials to replace a first list of raw materials implemented by the calculation device.

[0325] Such a selection process comprises a determination step, a calculation step and a selection step.

[0326] During the determination stage, a list of properties associated with the first list of raw materials and an associated sensory attribute are determined.

[0327] A sensory attribute refers to a known and intrinsic characteristic of a cosmetic product, perceptible to a human user through at least one of their five senses: sight, smell, taste, touch, or hearing. These attributes, quantifiable or qualifyable, contribute to the overall sensory experience perceived during the use of or interaction with said product.

[0328] For example, the user manually enters the list of properties and the sensory attribute via the input interface which the computing device 12 reads from memory 18.

[0329] During the calculation step, the calculation device 12 calculates the characteristics of the cosmetic composition using a prediction process as presented above.

[0330] During the selection step, the user selects a list of raw materials from the plurality of candidate raw material lists via the calculation device 12.

[0331] Alternatively, the calculation device 12 selects one or more lists respecting a set of selection criteria dependent on the sensory attribute.

[0332] For example, a selection criterion is a threshold assigned to each of the characteristics.

[0333] Advantageously, such a process makes it possible to select one or more lists of raw materials having characteristics close to the input list.

Claims

Demands

1. A method for generating at least one manufacturing data set, the manufacturing data set comprising manufacturing data for a cosmetic composition, said method being implemented by a computing device (12) and comprising a generation phase, the generation phase comprising a step of: - obtaining a manufacturing data set and one or more manufacturing data to be modified in the set obtained, - calculating a similarity coefficient between each manufacturing data to be modified and a plurality of candidate manufacturing data, - selecting, for each manufacturing data to be modified, at least one manufacturing data from among the plurality of candidate manufacturing data according to the calculated similarity coefficients, to obtain at least one selected data, - generating at least one manufacturing data set, to obtain at least one generated set,Each generated set is the set in which each manufacturing data to be modified is replaced by a respective selected manufacturing data.

2. A generation method according to claim 1, wherein each manufacturing data is associated with several values, each value being a quantification of a respective property of the manufacturing data, the similarity coefficient of a manufacturing data depending on at least one value of the manufacturing data.

3. A generation method according to claim 2, wherein a property is a measurable property.

4. A generation method according to claim 2 or 3, wherein a property is a qualitatively assessable property.

5. A generation method according to any one of claims 2 to 4, wherein, during the calculation step, the calculation device (12) calculates at least one similarity coefficient by applying the following formula: t ■ ri \i X&md. J ) — ^ [ ^ j)- IJ J With: • C(A, B) denotes the similarity coefficient between manufacturing data A and manufacturing data B, • xnwdj denotes the i-th manufacturing data to be modified, i being a non-zero integer, • xcand, J denotes the j-th candidate manufacturing data, j being a non-zero integer, • w denotes the sum of the weights applied to the p properties, • denotes the weight applied to the p-th property (Wp = 1 by default), • n is the number of properties.

6. ^P ( xmodlp xcand> j ) ~ R? • -^modip : value of the p-th property of xmodi, p being an integer between 1 and n, • yeand. j : value of the p-th property of xamd, j, • Rp : this value is at least equal to the difference between the 5th and 95th percentile of the distribution of values ​​of the p-th property.

7. A generation method according to any one of claims 1 to 5, wherein, for a specific manufacturing data to be modified, the selected candidate manufacturing data is the deletion of the specific manufacturing data, each generated set then being the resulting set in which the specific manufacturing data is deleted and each other manufacturing data to be modified possibly is replaced by a respective selected manufacturing data.

8. A generation method according to any one of claims 1 to 6, wherein the cosmetic composition is part of a cosmetic product, the manufacturing data being chosen from a list consisting of data relating to the cosmetic composition, data relating to packaging of the cosmetic product and data relating to the manufacturing technique of the cosmetic product used.

9. A generation method according to any one of claims 1 to 7, wherein the manufacturing data are chosen from a list of raw materials, a list of molecular structures and a list of characteristics.

10. A method for manufacturing a cosmetic composition, the method being implemented by a manufacturing system (10), the manufacturing method comprising: - a phase of implementing a method for generating at least one set of manufacturing data, the generation method being according to any one of claims 1 to 8, and - a phase of manufacturing a cosmetic composition from each set generated during the implementation phase.

11. A computing device (12) configured to generate at least one manufacturing data set, the manufacturing data set comprising manufacturing data for a cosmetic composition and: - obtain a manufacturing data set and one or more manufacturing data to be modified in the obtained set, - calculate a similarity coefficient between each manufacturing data to be modified and a plurality of candidate manufacturing data, - select, for each manufacturing data to be modified, at least one manufacturing data from among the plurality of candidate manufacturing data according to the calculated similarity coefficients, to obtain at least one selected data, - generate at least one manufacturing data set, to obtain at least one generated set, each generated set being the obtained set in which each manufacturing data to be modified.

12. Manufacturing system (10) for a cosmetic composition, the manufacturing system (10) comprising: - a calculation device (12) according to claim 10, and - a manufacturing device (14) suitable for manufacturing a cosmetic composition from the manufacturing data set obtained by the calculation device (12).

Images