JP2019008369A - Information processing apparatus, authentication system, authentication method and program - Google Patents

Abstract

An information processing apparatus that learns in advance features unique to an operator's input operation and can determine whether authentication is permitted or not based on the input operation performed by the operator under a task condition according to the learning state. provide. An information processing apparatus includes a learning processing unit, an authentication method determination unit, and an input operation method authentication unit. The learning processing unit collects operator operation information, learns an operator identification model for identifying an operator's input operation using a plurality of learning data based on the collected operation information, and performs authentication based on the model To determine whether or not to permit acceptable task conditions and authentication. The authentication method determination unit determines whether or not authentication based on the model is permitted for the operator when requesting authentication based on the model. Whether the input operation method authenticating unit succeeds in authenticating the operator according to the request based on a plurality of identification results based on the model of the operator according to the request with respect to the input operation responding to the task condition, when permitted. Determine whether. [Selection] Figure 2

Description

  The present invention relates to an information processing device, an authentication system, an authentication method, and a program, and more particularly, to an information processing device, an authentication system, an authentication method, and a program that have an authentication function.

  In recent years, the importance of data protection, that is, providing data that can be accessed only by a specific user has increased as the importance of data handled by computers has increased. As a method for realizing data protection, a technique called user authentication is generally known.

  As user authentication, various means are provided, for example, a password authentication method in which authentication is successful by inputting a secret password set by the user, a fingerprint authentication method, a face authentication method, an iris authentication method, Various methods such as PIN (Personal Identification Number) code authentication and picture authentication have been proposed and put into practical use.

  Among conventional authentication methods, desktop computers and the like still use password authentication most because they do not have a fingerprint reading device. However, password authentication has not been sufficient in terms of user convenience. For example, in password authentication, the input characters tend to be hidden, so that input errors tend to increase. There is also a possibility that the user himself forgets the password set by himself / herself. Furthermore, there is a vulnerability of the password itself when the user sets a weak password or uses the default password as it is, and there is a risk of being stolen by a third party.

  In addition, Patent Document 1 (Japanese Patent Laid-Open No. 2000-132514) is also known in relation to user authentication. Patent Document 1 discloses a configuration in which an input operation of a personal password is learned in addition to the password itself and is used as an authentication unit. This prevents impersonation due to password theft.

  However, the conventional technique and the technique of Patent Document 1 described above are merely techniques that use the input operation as an auxiliary to password authentication, and are not sufficient in that the user still has to remember and input a complicated password. It was.

  The present disclosure has been made in view of the above points, learns in advance features unique to the operator's input operation, and performs authentication based on the input operation by the operator under task conditions according to the learning state. It is an object of the present invention to provide an information processing apparatus capable of determining whether or not to permit.

  According to the present disclosure, in order to solve the above-described problem, an information processing apparatus that authenticates an operator having the following characteristics is provided. The information processing apparatus uses a collection means for collecting operation information by an operator via the input device and a plurality of learning data based on the collected operation information, and uses the operator's input device for the operation information. When learning based on the operator identification model, the learning means for learning the operator identification model for identifying the input operation and one or more verification data based on the collected operation information are used. And a determination means for determining whether or not to permit the authentication. The information processing apparatus further includes: a determination unit configured to determine whether authentication based on the operator identification model is permitted for an operator involved in the request when requesting authentication based on the operator identification model; Authentication decision to determine whether or not to successfully authenticate the operator for the request based on a plurality of identification results based on the operator identification model of the operator for the request for the input operation in response to the task condition Means.

  With the above configuration, it is possible to learn in advance features unique to the input operation by the operator, and to determine whether the authentication is permitted or not based on the input operation by the operator under a task condition corresponding to the learning state.

The hardware block diagram of the computer used as an information processing apparatus by this embodiment. The functional block diagram of the login authentication module for providing the authentication function based on input operation in the information processing apparatus by this embodiment. The more detailed functional block diagram of the learning process part in the login authentication module by this embodiment. The figure which shows the more detailed functional block of the learning state determination part in the learning process part by this embodiment with a periphery functional structure. The more detailed functional block diagram of the input operation authentication system authentication part in the login authentication module by this embodiment. The figure which illustrates the (A) login user selection screen, the (B) authentication screen of a password authentication system, and the (C) authentication screen of an input operation authentication system displayed in the case of login authentication in this embodiment. The flowchart which shows the flow of the authentication process based on input operation which the information processing apparatus in this embodiment performs. The figure which shows the data structure of the learning data prepared for learning in the authentication process based on input operation in this embodiment. The functional block diagram of the learning process part in the login authentication module by other embodiment. The functional block diagram of the input operation system authentication part in the login authentication module by other embodiment.

  Hereinafter, although it demonstrates with this embodiment, embodiment is not limited to embodiment mentioned later.

  The information processing apparatus, the authentication system, and the authentication method according to the present embodiment learn an operator identification model in advance in order to identify an input operation by an operator via an input device such as a keyboard, and the learned operator identification model Is used to determine whether or not the authentication of the operator relating to the authentication request is permitted based on the input operation via the input device.

  In the embodiments described below, an information processing apparatus having an authentication function based on an input operation alone and an authentication method executed by the information processing apparatus will be described as an example. However, the information processing apparatus, the authentication system, and the authentication method are not limited to these. In other embodiments, the authentication system including the information processing apparatus and the cloud or the remote computer on the remote server side and the authentication method that the information processing apparatus and the remote computer execute in cooperation with each other are not prevented. Absent.

  Hereinafter, the hardware configuration of an information processing apparatus including an authentication function based on an input operation according to the present embodiment will be described with reference to FIG. FIG. 1 is a hardware configuration diagram of a computer 10 used as the information processing apparatus according to the present embodiment. The computer 10 is typically configured as a general-purpose computer.

  A computer 10 shown in FIG. 1 includes a central processing unit (CPU) 12, a north bridge 14 that is connected to the CPU 12 and a memory, and a south bridge 16. The south bridge 16 is connected to the north bridge 14 via a dedicated bus or a PCI bus, and is connected to an I / O such as a PCI bus or USB.

  A RAM (Random Access Memory) 18 that provides a work area for the CPU 12 is connected to the north bridge 14. The north bridge 14 may further be connected to a graphic board 20 that outputs a video signal. A display 50 may be connected to the graphic board 20 via a video output interface.

  A PCI (Peripheral Component Interconnect) 22, a LAN port 24, an IEEE 1394 port 26, a USB port 28, an auxiliary storage device 30, an audio input / output 32, and a serial port 34 can be connected to the south bridge 16. The auxiliary storage device 30 is an HDD (Hard Disk Drive), an SSD (Solid State Drive), or the like, and an operating system for controlling the computer 10, programs for realizing the functional units described later, various system information, and various types Stores setting information. The LAN port 24 is an interface device that connects the computer 10 to a network by wire and wireless.

  Input devices such as a keyboard 52 and a mouse 54 are connected to the USB port 28. The computer reads the operating system and the program for the authentication function based on the input operation according to the present embodiment from the auxiliary storage device 30 and expands them in the work space provided by the RAM 18. Thereby, under the control of the CPU 12, each functional unit and each process described later are realized.

  In the present embodiment, the display 50 is used as a display device that displays a task for requesting input from an operator in an authentication function based on an input operation. In this embodiment, the keyboard 52 is used as an input device that receives an input operation in the authentication function.

  In the information processing apparatus having the authentication function based on the input operation according to the present embodiment, any commercially available products can be used for the display 50 and the keyboard 52, and no special apparatus is required. The information processing apparatus itself may have a general hardware configuration, and preferably has sufficient computing resources to execute or control a machine learning algorithm described later in the background. Is not required. The hardware configuration illustrated in FIG. 1 is an example, and an external device such as a camera may be provided in order to improve the recognition accuracy of the authentication function based on the input operation according to the present embodiment. Furthermore, although the keyboard 52 is described as an example of the input device, a mouse or a touch pad may be provided.

  FIG. 2 is a functional block diagram of a login authentication module for providing an authentication function based on the above-described input operation in the information processing apparatus according to the present embodiment. As shown in FIG. 2, the functional configuration of the login authentication module 100 includes an authentication method determination unit 110, a password method authentication unit 112, a learning processing unit 120, and an input operation method authentication unit 150.

  Note that the login authentication module 100 according to the present embodiment performs this predetermined authentication when an authentication method based on an input operation (hereinafter referred to as an input operation authentication method) is selected as an authentication method for a predetermined user account. Called in response to the start of login authentication processing for a user account. In other words, in the following explanation, authentication by the input operation authentication method is specified for the user account corresponding to the operator who operates the information processing apparatus, or the input operation authentication method is enabled as an option. In the following description, it is assumed that the authentication request has been made by designating the input operation authentication method.

  In the embodiment described below, it is assumed that the authentication function of the input operation authentication method according to this embodiment is provided in the login or sign-in processing for verifying the credential information for accessing the desktop on the computer. . However, the application range of the authentication function of the input operation authentication method is not limited to login or sign-in processing. In another embodiment, the authentication of the input operation authentication method is used for an authentication procedure required when accessing a resource such as a network shared folder or when promoting to administrator authority when installing an application or changing a setting. Function can be applied. In yet another embodiment, the authentication function of the input operation authentication method can be applied to login to a service or application provided on the Internet or a network in an organization.

  When the input operation authentication method is designated, the authentication method determination unit 110 determines whether or not authentication of the input operation authentication method is permitted for the operator who requests the request when authentication is requested. To do. As will be described later, for an operator who requests authentication, if the preparation for performing the authentication of the input operation authentication method is not yet completed, the authentication method determination unit 110 permits the authentication of the input operation authentication method. The password method authentication unit 112 is called for alternative authentication. In this case, after login authentication, the learning processing unit 120 is called to prepare for authentication of the input operation authentication method. When the preparation for performing the authentication of the input operation authentication method is completed, the authentication method determination unit 110 determines that the authentication of the input operation authentication method is permitted, and calls the input operation method authentication unit 150. The authentication method determination unit 110 constitutes a determination unit in the present embodiment.

  The password method authentication unit 112 executes password authentication using a normal user name and password. The password method authentication unit 112 is called when the authentication of the input operation authentication method according to the present embodiment is designated and the preparation is not completed, and alternatively performs login authentication by password authentication. . It should be noted that various settings according to alternative methods, for example, password settings, are performed in advance. In the embodiment to be described, password authentication is exemplified as an alternative method. However, the authentication is not particularly limited, and PIN code authentication or picture authentication may be performed in other embodiments.

  The learning processing unit 120 collects input operation information via an input device by an operator corresponding to the currently logged-in user account, prepares a plurality of learning data based on the collected operation information, and a valid operator The operator identification model for identifying the input operation is learned. Here, the operator identification model is configured by giving a parameter generated by learning to authenticate a specific operator to a predictor described later, and whether the input is by a specific operator. A model such as a neural network for identifying Here, since a predetermined alternative authentication process is passed, it may be considered that a legitimate operator corresponding to the currently logged-in user account is performing various operations. The learning processing unit 120 is called when authentication of the input operation authentication method is designated, and is called at least when the preparation is not completed.After the login, learning data is collected, and when sufficient learning is completed, Gives permission for authentication of the input operation authentication method.

  In the embodiment to be described, when the preparation is not completed, the learning processing unit 120 is called to collect and learn learning data, and on the other hand, the authorization of the input operation authentication method is once given. The following description will be made assuming that the learning processing unit 120 is not used. However, it is not limited to such an aspect. In another embodiment, after permission for authentication of the input operation authentication method is once given, learning data may be collected after login, learning may be continued, and the accuracy of authentication may be improved.

  The input operation method authentication unit 150 is called when the authentication of the input operation authentication method is designated and the authentication of the input operation authentication method is permitted, and performs authentication based on the input operation of the operator concerning the request. A process for determining whether or not to succeed is performed. Here, the input operation authentication method is a method in which the operator is asked to input an arbitrary non-secret character string during authentication, and the input operation actually performed by the operator at this time is determined in advance. This is a method for verifying the legitimacy of the operator based on the judgment as to whether or not it matches the characteristics of the input operation of the operator that has been learned.

  Hereinafter, a more detailed functional configuration of the learning processing unit 120 configuring the login authentication module 100 will be described with reference to FIGS. 3 and 4. FIG. 3 is a more detailed functional block diagram of the learning processing unit 120 in the login authentication module according to the present embodiment. As illustrated in FIG. 3, the functional configuration of the learning processing unit 120 includes a feature vector generation unit 122, a dummy input generation unit 124, a machine learning unit 128, a predictor 132, and a learning state determination unit 138. It is.

  The feature vector generation unit 122 collects operation information by the operator via the keyboard 52, and generates a feature vector constituting each of the learning data and the verification data from the operation information by the operator. Here, the feature vector is data converted from the original raw information so that it can be handled by machine learning, and is input to the predictor 132 constituting the operator identification model 130 described later. is there. As a feature vector, for example, in addition to a pressed key, a representation of one or both of a key time interval and a key pressing strength can be used. The feature vector generation unit 122 constitutes a collection unit in the present embodiment. Details of the feature vector will be described later.

  Based on the input operation statistical information 126 prepared by collecting input operations by many operators in advance, the dummy input generation unit 124 includes virtual pseudo operation information (hereinafter referred to as “virtual pseudo operation information”) by other operators than the above-described operators. , Also referred to as dummy data). The pseudo operation information is generated in the form of the feature vector described above. The pseudo operation information can be used for both learning and testing of the operator identification model 130. The dummy input generation unit 124 constitutes pseudo information generation means in the present embodiment.

  The machine learning unit 128 uses a plurality of learning data based on the collected operation information to execute a predetermined machine learning algorithm, thereby identifying a parameter 134 of the predictor 132 for identifying an input operation of a valid operator. To learn. Preferably, the machine learning unit 128 further uses a plurality of learning data based on the pseudo operation information generated by the dummy input generation unit 124 in addition to the plurality of learning data based on the actual operation information of the operator, The parameters 134 of the predictor 132 can be learned. The accuracy of machine learning can be improved by giving, as learning data, data having the characteristics of an input operation by a person other than the operator. The machine learning unit 128 provides the optimized parameter 134 of the predictor 132. The machine learning unit 128 constitutes learning means in the present embodiment.

  Here, the predictor 132 refers to a mathematical model for identifying an input operation of a predetermined operator. The predictor 132 identifies whether the input operation representing the given input operation is performed by a predetermined operator or another operator. The identification result is output. As described above, the input data representing the operator's input operation is given as a feature vector. For this reason, the predictor 132 is configured to output an identification result for identifying whether or not the input feature vector is an input operation of the operator.

  The identification result is typically represented by a probability. For example, the probability that the input data is from the target operator (VALID) or the probability that the input data is from a person other than the target operator (INVALID) is output. Alternatively, it is output as a probability that the input data is from each operator (USER1, USER2).

  A parameter 134 adjusted by a machine learning algorithm is calculated and set in the predictor 132 so that a feature vector included in each learning data is input and a correct identification result is output. The predictor 132 and its parameters 134 constitute the operator identification model 130 in this embodiment.

  Any machine learning algorithm known so far can be used as the machine learning algorithm executed by the machine learning unit 128, and an appropriate algorithm may be used according to the mode of the predictor 132 to which machine learning is applied. . As a machine learning algorithm, supervised learning can be preferably performed. In this case, a valid (VALID) correct answer label indicating that the input has been made by the operator is attached to the learning data generated from the actual operation information of the operator. On the other hand, the learning data generated from the pseudo operation information by the dummy input generation unit 124 is attached with an invalid (INVALID) correct answer label indicating that the input was made by a person other than the operator.

  In addition, the predictor 132 to which the machine learning algorithm is applied is not particularly limited, but a neural network (ANN: Artificial Neural Network) and a support vector machine (SVM) or a combination thereof may be used. Can be used. For example, the neural network can adjust the parameters by the error back-propagation method so that the input can be correctly identified as the learning data. The neural network includes an input layer, a hidden layer, and an output layer, and parameters to be learned include a weight between layers, a bias of each layer, and the like. For example, for SVM, an algorithm well known as a solution method for a quadratic programming problem such as a sequential minimum problem optimization method can be used. The parameters to be learned by the SVM include parameters of linear threshold elements.

  The specific calculation in the machine learning algorithm of the machine learning unit 128 may be performed by the CPU of the computer 10 or may be performed by a GPU connected to the computer as necessary. Alternatively, it is also possible to give learning data to another computer system connected via a network and perform a specific calculation in a machine learning algorithm, so that the information processing apparatus only obtains a learning product. Good. The same applies to specific calculations in the predictor 132.

  Part of the collected operation information is used not only for learning data but also for verification data for verifying the accuracy of the predictor. Similarly, part of the pseudo operation information is also used for verification data. The predictor 132 receives the input of the feature vector constituting the verification data under the given parameter 134, and whether the input operation of each verification data has been performed by the operator or another operator Is identified, and the identification result is output. Since the correct answer label is given in the verification data, the identification accuracy of the predictor 132 can be evaluated from the identification result output by the predictor 132. The identification result prepared for evaluating the identification accuracy is referred to as a post-learning inspection result 136.

  The learning state determination unit 138 uses one or more verification data based on the collected operation information and the post-learning inspection result 136 output from the predictor 132 for the verification data to determine the learning state of the operator identification model 130. evaluate. The learning state determination unit 138 determines, according to the result of evaluation of the learning state, an acceptable task condition for authentication based on the operator identification model 130 in the current learning state and whether to permit the authentication. Here, the acceptable task condition can include designation of the number of characters of an arbitrary character string that is requested to be input upon authentication.

  When the authentication based on the input operation is permitted, the fact that the authentication of the input operation authentication method is permitted for the operator's account is recorded. Then, the determined task condition and the learned parameter 134 of the predictor 132 are passed from the learning processing unit 120 to the input operation method authentication unit 150.

  FIG. 4 shows more detailed functional blocks of the learning state determination unit 138 in the learning processing unit 120 according to the present embodiment, together with the peripheral functional configuration. As shown in FIG. 4, the functional configuration of the learning state determination unit 138 includes a validity verification unit 140 and an authentication error accuracy calculation unit 142.

  The validity verification unit 140 determines whether or not the post-learning inspection result for each verification data is correct. Whether the input data for the post-learning inspection result 136 is generated by the feature vector generation unit 122 from the collected actual operation information or is based on the pseudo operation information generated by the dummy input generation unit 124 Are known, and each verification data can be given a correct answer label. For this reason, it is possible to determine whether the value output as the identification result is correct. However, since the identification result is represented by a probability, it is possible to make a mistake, and a measurement record of the post-learning inspection result 136 is recorded, and the calculated probability is a set of input values of the verification data. It is preferable to evaluate that there is no significant difference.

  Here, whether or not the verification data generated from the actual input operation has yielded a result that the probability of being a valid operator input is 50% or more, Thus, whether or not a result that the probability that the input is a valid operator is less than 50% is obtained is a criterion. The validity verification unit 140 determines that the identification result is not valid when the condition is not satisfied.

  The authentication error accuracy calculation unit 142 is a subject sufficient to authenticate from the probability obtained as the post-learning inspection result 136 based on the post-learning inspection result 136 by the current operator identification model 130 for each of the one or more verification data. Determine the conditions. More specifically, the authentication error accuracy calculation unit 142 calculates the number of characters that can obtain a predetermined authentication accuracy by calculating the number of character strings that can be used as an input and calculating the number of characters that are sufficient for authentication. To do. If the number of characters does not satisfy a predetermined condition, the authentication error accuracy calculation unit 142 can determine that authentication based on the operator identification model 130 is not permitted.

  Here, the predetermined standard is a standard for determining the number of characters at a level that does not cause inconvenience for human input. For example, for login authentication, an input of 11 characters or less is allowed, but if it is considered inconvenient to require an input of more than 11 characters, it is given on the basis of 11 characters or less. That's fine. When the setting is made so that the authentication accuracy is sufficiently high such as 99% or more, the character string of the subject is a character number of a level that does not cause inconvenience for human input, such as 11 characters. In this case, authentication of the input operation authentication method is permitted.

  In this way, if the learning state is good and the state can be identified with high accuracy, authentication can be determined by imposing the task of inputting a character string having a shorter number of characters on the operator. become. On the other hand, if the learning state is not good and identification is not possible with very high accuracy, a problem of inputting a character string having a longer number of characters is imposed on the operator. Alternatively, if the learning state is poor, the learning is incomplete, and authentication itself based on the input operation is not permitted.

  It should be noted that it is possible to statistically determine how many characters are required to maintain a predetermined authentication accuracy. In addition, if there is a difference between the probability that the verification data can be determined to be a correct result and the erroneous determination probability, the probability that a correct result is returned for a plurality of input data is more than 50% has a large number of characters in the character string. It improves so much. Details of how to calculate how many characters are required to maintain the authentication accuracy will be described later.

  Hereinafter, a more detailed functional configuration of the input operation method authenticating unit 150 configuring the login authentication module 100 will be described with reference to FIG. FIG. 5 is a more detailed functional block diagram of the input operation method authenticating unit 150 in the login authentication module 100 according to the present embodiment. As illustrated in FIG. 5, the functional configuration of the input operation method authentication unit 150 includes an authentication character string generation unit 152, a feature vector generation unit 154, a predictor 156, and an authentication determination unit 162.

  The authentication character string generation unit 152 generates a character string corresponding to the task condition passed from the learning processing unit 120 and displayed on a display device such as the display 50 to prompt input. The authentication character string generation unit 152 holds, for example, a word dictionary, and selects a character string that matches the task condition from character strings registered in the word dictionary. For example, when the task condition includes designation of 11 characters, it is possible to select a word having 11 characters or more, or to select a plurality of words having 11 characters or more in total. The authentication character string generation unit 152 constitutes an authentication input generation unit in the present embodiment.

  The feature vector generation unit 154 is similar to the feature vector generation unit 122 of the learning processing unit 120 described above, and converts input operation information input in response to the task condition into one or more authentication feature vectors. . The feature vector generation unit 154 constitutes conversion means in the present embodiment.

  The predictor 156 is the same as the predictor 132 of the learning processing unit 120 described above, and the parameter 158 is also output from the learning processing unit 120 in response to the determination of permitting authentication based on the input operation. Yes, it is the same as the parameter 134 at the completion of learning. The predictor 156 and its parameters 158 constitute the operator identification model 160 in this embodiment.

  When authentication based on an input operation is permitted, the authentication determination unit 162 is based on a plurality of identification results by the operator identification model 160 corresponding to the operator who is requested to authenticate the input operation in response to the task condition. It is determined whether or not the authentication of the operator related to the authentication request is successful. The authentication determination unit 162 constitutes an authentication determination unit in the present embodiment. Details of the process for determining whether or not the authentication is successful will be described later.

  In the embodiment to be described, the authentication character string generation unit 152 typically randomly selects one or more words that match the task condition. On the other hand, the identification result by the operator identification model 160 for each verification data includes, for example, probability information that the probability that the input is a valid operator is 80%. By using such certain reliability information, it is possible to extract which feature vector has high recognition accuracy.

  Therefore, the learning state determination unit 138 of the learning processing unit 120 described above can extract one or more pieces of verification data whose reliability satisfies a predetermined condition and notify the input operation method authentication unit 150 of this. The learning state determination unit 138 also constitutes extraction means in the present embodiment. Then, the authentication character string generation unit 152 generates a character string by preferentially using characters related to the extracted one or more pieces of verification data, thereby generating a character string that generates a feature vector with high authentication accuracy. Can be specified. Thereby, authentication accuracy can be improved. If this is not performed, for example, a character string input by the operator is determined to be a legitimate operator, but even a probability of about 51% may be adopted. On the other hand, the occurrence of such an event can be prevented by preferentially selecting and using a character string with high recognition accuracy. That is, the authentication determination unit 162 (to be described later) can more accurately determine whether authentication is successful.

  In the above-described embodiment, the learning of the recognition model is not particularly performed after the learning is once completed. However, it is not particularly limited to such an embodiment. In another embodiment, even after learning is completed, after logging in with authentication based on the input operation authentication method or other authentication, collection of key operations is continued, and a certain amount of learning data is obtained each time. It is possible to re-learn and update the parameters of the recognition model.

  FIG. 6 exemplifies (A) a user / account selection screen, (B) an authentication screen for a password authentication method, and (C) an authentication screen for an input operation authentication method displayed at the time of login authentication in the present embodiment. First, a predetermined user account is selected from the pull-down menu 202 on the user account selection screen 200 shown in FIG. If the selected user account is designated for the input operation authentication method and learning of the operator identification model for performing the input operation authentication method has not yet been completed, FIG. An authentication screen 210 of the password authentication method shown in (B) is displayed. On the other hand, if learning of the operator identification model has already been completed for the selected user account, an authentication screen 220 of the input operation authentication method shown in FIG. 6C is displayed.

  The authentication screen 210 of the password authentication method shown in FIG. 6B is used in many situations because the authentication procedure is possible if the display 50, the keyboard 52, and the computer 10 are provided. However, on the other hand, there are the following inadequate points.

  That is, first, there is a point that a password input error is likely to occur. This is because the input character string in the text box 214 for inputting the password is displayed with being masked with “*” or a black circle, and it is difficult to confirm the input content. Second, the user does not always set a password without vulnerability. There are not a few cases where the default password is used as it is, and there are many users who use a trivial character string such as “password” or “1234”. Thirdly, there is a risk that a third party can view the entered situation. In principle, the password must be kept secret, but it may not be seen by a third party, and the password is required to be updated at a predetermined frequency, resulting in an increase in the load on the user. In addition, due to the strength of the password, “It is necessary to use one or more special characters such as underbars”, “Must include uppercase letters in the alphabet”, “Must contain at least one number and one alphabet”, “ There are restrictions on each service to be used, such as “Password must be 8 characters or longer”, and you may be forced to use a password that is difficult to remember.

  On the other hand, in the authentication of the input operation authentication method shown in FIG. 6C, the operator inputs a character string in the text box 224 for user authentication after selecting a user account. However, the character string 226 that is required to be input by the operator at this time is not a password but a specific character string as a task that requires the computer to input for authentication. Therefore, since it is not necessary to keep it secret, it can be displayed in the text box 224 without being masked. In the example of FIG. 6C, the character string 226 of the assignment is “authentication”, and the operator inputs the same character string “authentication” as it is.

  Here, the success or failure of the authentication in response to the input of the character string is performed using the operator identification model 160 that has learned a pattern that characterizes the input operation of the operator in advance by machine learning. The content of the input character string itself is not used for authentication, and the input content may be disclosed, and there is little problem even if the input state is stolen.

  Hereinafter, the flow of the authentication process of the input operation authentication method executed by the information processing apparatus according to the present embodiment will be described in more detail with reference to FIG. FIG. 7 is a flowchart illustrating a flow of authentication processing based on an input operation executed by the information processing apparatus according to the present embodiment.

  The process shown in FIG. 7 is started from step S100 in response to a login authentication request from the operator. In the following description, it is assumed that the input operation authentication method is designated as the login authentication method for the information processing apparatus for the operator. In step S101, the information processing apparatus acquires the selected user name. When a predetermined user name is selected from the pull-down menu 202 on the user selection screen shown in FIG. 6A and the OK button 204 is pressed, the selected user name is passed to the login authentication module 100.

  In step S102, the information processing apparatus determines whether the operator identification model of the operator corresponding to the selected user name has been learned by the authentication method determination unit 110, and authentication of the input operation authentication method is permitted. Determine. If it is determined in step S102 that it has not been learned and is not permitted (NO), the process branches to step S103, and the process in the learning stage is performed thereafter.

  In step S103, the information processing apparatus performs password authentication as an alternative authentication method for the operator using the password method authentication unit 112. Note that the password is set in advance as a normal authentication method for the operator. In step S104, the information processing apparatus determines whether login authentication has succeeded. If it is determined in step S104 that login authentication has failed (NO), the process returns to step S103, and password authentication attempts are repeated. On the other hand, if it is determined in step S104 that the login authentication is successful (YES), the process proceeds to step S105.

  After successful login, the operator can freely operate the information processing apparatus within the scope of his / her authentication authority. The work includes operations involving keyboard operations in daily work such as document creation and coding work. The logged-in operator only performs daily work as usual. On the other hand, in step S105, the information processing apparatus collects key operation information performed through the keyboard 52 in the background by the learning processing unit 120 in the work by the logged-in user, and uses the information as input data for machine learning. Convert.

  In step S106, the information processing apparatus determines whether a certain amount of input data has been obtained. Here, the fixed amount means a reasonable data amount when applying the machine learning algorithm. Instead of a certain amount, a period may be set every day. If it is determined in step S106 that a certain amount of input data has not yet been obtained (NO), the process loops to step S105, and input data collection is continued. On the other hand, if it is determined in step S106 that a certain amount of input data has been obtained (YES), the process proceeds to step S107.

  In step S <b> 107, the information processing apparatus uses the feature vector generation unit 122 to generate learning data from the collected input data. The information processing apparatus can also generate dummy learning data from the pseudo operation information. The collected input data is converted into a feature vector of a predetermined format and is taken into a machine learning algorithm. Since the input data is taken in as new input data while the operator works, the learning data is accumulated as the operator operates the information processing apparatus.

  FIG. 8 shows a data structure of learning data prepared for learning in the authentication process based on the input operation in the present embodiment. When machine learning is performed, raw input data is shaped into a specific format that can be easily handled by a machine learning algorithm. This formatted data is referred to as a feature vector. Hereinafter, an example of a specific method for creating a feature vector from a keyboard operation will be described.

  In generating the feature vector, it is desirable that the operator's features are appropriately expressed. First, as a first method, a method for learning the characteristics of the operator based on the key operation time will be described.

  For example, when the operator presses the keys in the order of “a”, “b”, and “c”, “b” is 0.22 seconds between the “a” and “b” key operations, respectively. If it is assumed that the key operation of “c” takes 0.31 seconds, information on this operation can be obtained, for example, by using the input key and its time interval, for example, (a, b, 0.22) and (b, c , 0.31) can be represented by two feature vectors. Here, the time interval may be the interval between the timing when the first key is pressed and the timing when the second key is pressed, or the timing when the first key is released and the second key It may be an interval from the timing when the key is pressed, may be an interval between a timing when the first key is pressed and a timing when the second key is released, or a timing when the first key is released. It may be an interval from the timing when the second key is released.

  Furthermore, in the above example, one feature vector is generated for a set of two key operations, but one feature vector can be generated for a set of three or four key operations. For example, in the case of a set of three key operations, it can be expressed by one feature vector such as (a, b, c, 0.22, 0.31). If a feature vector is configured by combining more key operations, key operation information having a strong dependency between keys can be obtained, and the input operation of the operator can be easily identified. On the other hand, the dimension of the feature vector increases, the number of data required for learning increases, and the calculation time also increases, so there is a trade-off relationship. Further, in the above example, a real value such as 0.22 seconds or 0.31 seconds is used, but it may be expressed by an integer value indicating the magnitude of the time interval.

  Also, in supervised machine learning, when learning is performed, learning data that is a set of a feature vector and a correct answer label in the feature vector is generated and used as input. Since the feature vector generated by the feature vector generation unit 122 in the key operation is the content actually typed by the operator who logged in properly, it can be determined that this is an input by the operator. That is, each learning data based on the feature vector generation unit 122 is configured as a set of a feature vector and a valid (VALID) correct label, as illustrated in FIG.

  On the other hand, in order to improve the accuracy of machine learning, it is preferable to provide feature data input by a person other than the operator as learning data. The dummy input generation unit 124 holds a list of statistical information on general operator behavior obtained beforehand from statistics of key input data. The dummy input generation unit 124 receives the statistical information list as an input, and applies a bias such as a normal distribution to the time of the list so that a person other than the operator performs a pseudo key input. Generate feature vectors. The learning data based on the dummy input generation unit 124 is configured as a set of a feature vector and an invalid (INVALID) correct label, as illustrated in FIG.

  Note that there may be a plurality of lists of statistical information, or they may be generated separately. For example, learning data for a group with fast key input and learning data for a group with slow key input may be generated. Furthermore, in addition to the data generated from the statistical information, learning data based on input operations of other operators actually measured may be added.

  In the above description, the feature vector representing the key and the time interval of the key has been explained. However, if the feature vector can be obtained from what the operator uses as an input device, the feature vector is generated. Can be used for For example, if the keyboard has a function capable of detecting the pressure with which a key is pressed, a feature vector can be created by using the pressed key and the pressed pressure (strength of key pressing). In human interface devices such as keyboards, data transmission methods to computers are standardized, and pressure data notification is known as a value from 0 to 1023. Here, The feature vector can be obtained by using the pressed key and the pressure value detected by the key. The dummy information generation algorithm of the dummy input generation unit 124 may be an algorithm corresponding to the structure of the feature vector, and no special change is required for other portions. In addition, as long as the input operation is performed by the operator using the input device, the operation is not limited to the keyboard operation, and may be a mouse operation or a touch pad operation. Furthermore, when an external device such as the camera described above is provided, for example, the camera is configured to take which key to press with a finger, etc., identified by image processing, and information of the pressed finger is used as unique information in the feature vector. By including, recognition accuracy can be improved. For example, the recognition accuracy can be improved by using an operator's heel to press the “a” key with a ring finger. Hereinafter, for the sake of simplicity, description will be made assuming that all key operations are feature extraction by a combination of two key operations.

  In step S108, the information processing apparatus inputs the machine learning algorithm 128 using the learning data obtained at the present time to the machine learning algorithm, and obtains the optimized parameter 134 of the predictor 132. Here, prediction data for receiving learning data from the feature vector generation unit 122 and the dummy input generation unit 124 and determining whether or not the feature vector of each learning data is input by the operator is not used. An optimized parameter 134 for the generator 132 is generated. When the key operation is input, even the key operation of the same operator is not always performed at the same time interval, so the determination is made by a probabilistic algorithm, and the determination result is also output probabilistically.

  In step 109, the information processing apparatus performs a test of the operator identification model including the obtained predictor 132 and parameter 134, and more specifically, a task condition to be imposed in order to obtain a predetermined accuracy during authentication, more specifically Determine the number of characters required. The predictor 132 receives the parameter 134 from the machine learning unit 128, determines the probability that the verification data is input by the operator, and determines whether the input is from the operator as a post-learning inspection result 136. Output.

  As described above, in order to maintain the calculation accuracy of the authentication accuracy, it is possible to statistically determine how many characters are required to be input. When there is a difference between the identification probability that each verification data (for example, representing two key operations) can be correctly identified and the erroneous identification probability, a plurality of input data (character string input operations that require a plurality of key operations) The probability that the correct result is over 50% increases as the number of characters increases.

  For example, when the probability that each verification data from the operator can be correctly identified is p, the probability of incorrect identification is represented by (1-p). In this case, the probability that k or more input values can be correctly determined out of n pieces of input data can be calculated by the following equation in order to follow the binomial distribution.

In the above formula, parentheses (n k) ^T represent the number of combinations of n to k, that is, binomial coefficients. Actually, the probability that the verification data can be correctly determined also depends on the input character set and the time interval at that time. For this reason, it is difficult to obtain the probability as a fixed value, and it may be required to perform rigorous calculation. Can be obtained.

  Since the binomial distribution can be approximated by a standard normal distribution when the parameter n of the data is large, if it can be approximated for simplicity, the number of samples n is equal to the identification probability p that can correctly identify each verification data of the operator. The binomial distribution for the data can be approximated by the following equation.

  Here, X is a standard normal distribution, and Y is a binomial distribution to be obtained. At this time, assuming that the number of samples of n is 10 and p = 60%, the value can be calculated by the following formula when the probability that input data of more than half can be correctly identified is estimated.

  However, since this is not consistent with the approximate condition that n is sufficiently large, it is different from the actual estimation result and is an approximation. When actually performing the calculation, it would be possible to obtain a correct result by simply calculating the probability of the binomial distribution without approximation.

  This probability varies depending on the number of samples of n and also varies depending on the value of p. For example, if the probability is calculated in the same manner, assuming that p = 0.6, the number n of input data necessary to obtain the probability of determining that the majority data is correct with a probability of 99% is estimated to be about 130. The On the other hand, when p = 0.7, it is estimated that n = 30 is sufficient, and when p = 0.8, it is estimated that n is about 10, and if the measurement accuracy increases, it is sufficient. It is thought that it will settle to a practical value.

  That is, in step S108, machine learning is repeatedly performed for each piece of verification data so as to increase the identification probability p whether the value is input by the operator or not, and the parameters are adjusted. In step S109, on the basis of the identification probability p, the probability that the determination is erroneous under the condition that more than half of the plurality of feature vectors is input by the operator is calculated, and the determination is made as an error. If the probability of performing is less than a certain standard (for example, 1% or less), the required number of feature vectors that can be determined at a sufficient level is obtained.

  In step S110, the information processing apparatus determines whether the test result is good. If it is determined in step S110 that the test result is not good (NO), the process returns to step S105, and further learning data is collected for the operator, and learning is attempted. This improves the accuracy of the operator identification model. On the other hand, if it is determined in step S110 that the test result is good (YES), the process branches to step S111 to shift to the learned state for the user account of the operator. In the above case, it is determined that authentication is possible when the required number of feature vectors is within a number (for example, 11 characters) within a range that can be input by a human, and otherwise, it is determined that authentication is impossible.

  On the other hand, if it is determined in step S102 that learning has been performed and authentication of the input operation authentication method is permitted (YES), the process branches to step S112, and an authentication stage process is performed.

  In step S112, the information processing apparatus displays the character string of the task to be input and instructs the input of the task character string. At this time, the authentication character string generation unit 152 receives the specified number of characters given as the task condition and creates an input character string therefrom. For example, as illustrated in the description of the authentication error accuracy calculation unit 142 with reference to FIG. 4, when 10 feature vectors are required for authentication, the number of characters necessary to create 10 feature vectors is: Since two characters are used as one feature vector, there are 11 characters. Therefore, the authentication character string generation unit 152 generates a character string of 11 characters or more. At this time, the character string for authentication generation unit 152 generates a character string by preferentially using characters related to the verification data with high reliability, thereby generating a character string that generates a feature vector with high authentication accuracy. May be generated.

  The character string generated by the authentication character string generation unit 152 is transmitted to the operator via the display 50 on the login authentication screen, and the operator inputs characters according to the instruction. In step S113, the information processing apparatus accepts an input in response to the task character string from the operator. In step S <b> 114, the information processing apparatus generates one or more feature vectors for authentication based on the input operation in response to the task character string by the feature vector generation unit 154. As described above, when a character string of 11 characters is generated, 10 feature vectors are generated. If the input character string is not the same as the task character string and a predetermined number of feature vectors cannot be obtained, typically, the input of the task character string is required again.

  In step S115, the information processing apparatus inputs the authentication feature vector to the predictor 156 of the operator identification model 160, and the predictor 156 inputs the feature vector input by the operator based on the parameter 158. The identification result of whether or not is output. Identification results are obtained for several feature vectors.

  In step S <b> 116, the information processing apparatus uses the authentication determination unit 162 to determine whether or not the user is a valid operator based on a plurality of identification results. The authentication determination unit 162 counts the number of feature vectors determined to be input from a valid operator. As described above, the probability that an event occurs when the number of feature vectors determined to be valid exceeds half of the input feature vectors can be estimated. In the above example, if it is determined that more than half of the feature vectors are input by the operator, the number of characters in the task condition is set so that the input of the operator is 99% of the entire character string. ing. That is, if more than half of the input data is determined to be input by the operator, it can be determined that the input is made by the user with a probability of 99%. With this result, the authentication determination unit 162 can determine whether the authentication has succeeded or failed.

  If it is determined in step S116 that the operator is a valid operator (YES), login authentication is thereby successful, and in step S117, the login state is entered. On the other hand, if it is determined in step S116 that the operator is not a valid operator (NO), the process returns to step S112 to input the designated character string again.

  The authentication of the input operation authentication method according to the embodiment described above has the following advantages over the normal password authentication. That is, there is an advantage that a password input error is unlikely to occur. This is because the character string used for authentication is displayed on the display 50 and does not need to be kept secret, and the input content is also displayed on the display. Furthermore, it is not assumed that the user sets a weak password. This is because a password is not set and a weak password cannot be assumed in the first place. Furthermore, even if the input part is stolen by a third party, no major problem occurs. This is because the content of the input character string itself is not a secret, but is for confirming the consistency of the characteristics of the operator. Furthermore, there is an advantage that the operator himself / herself never forgets the password set by the operator. This is because the character string to be input is designated from the information processing apparatus side each time, so that the operator does not need to remember the password.

  Hereinafter, a plurality of other embodiments will be further described. FIG. 9 is a functional block diagram of a learning processing unit in a login authentication module according to another embodiment. The learning processing unit 120 according to the embodiment illustrated in FIG. 9 further includes a filter unit 146 before the feature vector generation unit 122. The filter unit 146 constitutes a filter unit that excludes the collected operation information by the operator via the keyboard 52 from operations related to a predetermined exclusion condition.

  In the above-described embodiments, the keyboard input data is given to the feature vector generation unit 122 and used as learning data. However, when an actual typing operation is considered, a person always performs password authentication. Typing is not done in this way. Usually, an operator may be troubled during work or take an action such as leaving his seat during work. Keyboard operations performed during such actions may include key operations that are not suitable as learning data because key operations may occur at time intervals that deviate significantly from the standard, which would not be spent when typing a password. Will be. As a predetermined exclusion condition, it is possible to perform efficient learning by limiting the range of time intervals for key operations to be acquired and not generating feature vectors for locations where the time intervals for key operations are extremely long. Become.

  FIG. 10 is a functional block diagram of an input operation method authentication unit in a login authentication module according to still another embodiment. The input operation method authenticating unit 150 according to the embodiment illustrated in FIG. 10 further includes a character string history recording unit 164. The character string history recording unit 164 is a recording unit that records the history of the character string displayed on the display 50. The authentication character string generation unit 152 manages the generation frequency of the same character string based on the recorded history. Thereby, the tolerance to the snooping of the input content can be improved.

  In recent years, surveillance cameras have been increasing in towns and stores, and smartphones are also equipped with cameras, so that it is becoming difficult for operators to know when they are being shot. There may be a need to take out a notebook PC at a destination cafe or the like to perform business, but the keyboard operation itself at the time of login is not necessarily recorded.

  In the configuration described above, a log of the character string generated by the authentication character string generation unit 152 is recorded, and the log is used with a sufficient time interval so that the same input character designation is not frequently generated. ing. Therefore, even if the login operation is recorded in the camera and the input speed is determined, the same input character designation will not come for the time being, so that the resistance to attacks can be improved.

  Further, in the above-described embodiments, the problem to be asked of the operator is that the character string having a predetermined number of characters is input only once. However, in the embodiment described above, since the identification probability of each feature vector is left to machine learning, it is difficult to guarantee how much recognition accuracy is finally obtained. For example, when the identification probability p is about 80%, 99% recognition accuracy can be maintained when more than half of the 10 feature vectors are recognized as belonging to the operator. On the other hand, when the identification probability p is about 60%, the recognition accuracy is reduced to about 74% with 10 feature vectors. However, as a result of repeated learning, there is a possibility that 60% may be the upper limit or 80%, which is difficult to know in advance. On the other hand, setting the number of characters to, for example, 100 characters is an operation that places a very large load on the user.

  Therefore, in the preferred embodiment, the number of challenges for inputting a character string as a challenge in the authentication process in the authentication determination unit 162 can be set to a plurality of times. When determining an acceptable task condition, the learning state determination unit 138 can determine the number of challenges for a task that requires input of the designated number of characters in addition to the designation of the number of characters. Based on the identification result of the current operator identification model, the learning state determination unit 138 determines the number of challenge times of two or more when a predetermined authentication accuracy cannot be obtained by a single challenge of a designated number of characters. be able to.

  In this case, when the above-described identification probability p is 60% and 10 feature vectors are used, the recognition accuracy is about 74% in one challenge, but this is repeated twice or more. And the probability changes. For example, when three challenges are made, the probability of successful authentication more than once increases to about 84%, and the probability can be improved by further increasing the number of challenges.

  As a result of machine learning, when there is not much advantage in distinguishing between the operator and others, and when the operator does not want to force the input of a long number of characters, the number of challenges is determined multiple times By doing so, the user recognition probability can be improved. Thereby, an alternative means when the result of machine learning does not converge to a good result can be provided.

  According to the embodiment described above, it is possible to learn in advance features unique to the operator's input operation, and to determine whether authentication is permitted or not based on the input operation by the operator under the task conditions according to the learning state Information processing apparatus, authentication system, and authentication method can be provided. This eliminates the need for the operator to manage passwords, so that the operator can be prevented from entering the password and at the same time prevent obvious attacks such as security brute force attacks.

  In the embodiments described above, the information processing apparatus provided with an authentication function based on an input operation alone and the authentication method executed by the information processing apparatus have been described as examples. However, the information processing apparatus, the authentication system, and the authentication method are not limited to these. In another embodiment, an authentication system including an information processing device and a cloud or a remote computer on the remote server side, and an authentication method executed in cooperation between the information processing device and the remote computer may be configured. In such an embodiment, for example, a function of the machine learning unit 128 that calculates parameters of the predictor 132 based on given learning data can be provided as a cloud service from a remote server. Furthermore, a function of the predictor 132 that outputs an identification result with respect to a given input can be provided as a cloud service from a remote server.

  The functional unit can be realized by a computer-executable program written in a legacy programming language such as assembler, C, C ++, C #, Java (registered trademark), an object-oriented programming language, or the like. ROM, EEPROM, EPROM , Stored in a device-readable recording medium such as a flash memory, a flexible disk, a CD-ROM, a CD-RW, a DVD-ROM, a DVD-RAM, a DVD-RW, a Blu-ray disc, an SD card, an MO, or through an electric communication line Can be distributed.

  Although the embodiments of the present invention have been described so far, the embodiments of the present invention are not limited to the above-described embodiments, and those skilled in the art may conceive other embodiments, additions, modifications, deletions, and the like. It can be changed within the range that can be done, and any embodiment is included in the scope of the present invention as long as the effects of the present invention are exhibited.

DESCRIPTION OF SYMBOLS 10 ... Computer, 12 ... CPU, 14 ... North bridge, 16 ... South bridge, 18 ... RAM, 20 ... Graphic board, 24 ... LAN port, 26 ... IEEE1394 port, 28 ... USB port, 30 ... Auxiliary storage device, 32 ... Audio input / output, 34 ... serial port, 50 ... display, 52 ... keyboard, 54 ... mouse, 100 ... login authentication module, 110 ... authentication method determination unit, 112 ... password method authentication unit, 120 ... learning processing unit, 122, 154 ... feature vector generation unit, 124 ... dummy input generation unit, 126 ... input operation statistical information, 128 ... machine learning unit, 130, 160 ... operator identification model, 132, 156 ... predictor, 134, 158 ... parameter, 136 ... Test result after learning, 138 ... learning state determination unit, 140 ... appropriate Verification unit 142 ... Accuracy calculation unit, 146 ... Filter unit, 150 ... Input operation method authentication unit, 152 ... Authentication character string generation unit, 162 ... Authentication judgment unit, 164 ... Character string history recording unit, 200 ... User account Selection screen 202 ... Pull-down menu 204 ... OK button 210 ... User authentication method authentication screen 214, 224 ... Text box 226 ... Input operation authentication method authentication screen 220 ... Character string

JP 2000-132514 A

Claims (13)

An information processing apparatus for authenticating an operator,
A collection means for collecting operation information by an operator via an input device;
Learning means for learning an operator identification model for identifying an input operation of the operator related to the operation information via the input device using a plurality of learning data based on the collected operation information;
Using one or more verification data based on the collected operation information, a task condition acceptable for authentication based on the operator identification model and whether to permit the authentication are determined for an operator involved in learning. A determination means;
When requesting authentication based on the operator identification model, a determination unit that determines whether authentication based on the operator identification model is permitted for the operator involved in the request;
Whether or not to successfully authenticate the operator according to the request based on a plurality of identification results by the operator identification model of the operator according to the request with respect to the input operation in response to the task condition, if permitted An information processing apparatus comprising: an authentication determination unit that determines whether or not.   The information processing apparatus further includes pseudo information generation means for generating virtual pseudo operation information by an operator other than the operator involved in the learning based on the statistical information of the operation, and the learning means The information processing apparatus according to claim 1, wherein the operator identification model is further learned using a plurality of learning data based on pseudo operation information.   The permissible task condition includes designation of the number of characters to be input upon authentication, and the determination means has a predetermined authentication accuracy based on a current identification result by the operator identification model for each of the one or more verification data. The number of characters to be obtained is determined, and if the number of characters does not satisfy a predetermined condition, it is determined that authentication based on the operator identification model is not permitted. The information processing apparatus described.   The permissible task condition further includes the number of challenges for a task that requires input of a specified number of characters, and the determining means is based on a current identification result by the operator identification model for each of the one or more verification data. 4. The information processing apparatus according to claim 3, wherein a predetermined number of challenges is determined when a predetermined authentication accuracy cannot be obtained by one challenge of the designated number of characters.   The information processing apparatus according to any one of claims 1 to 4, further comprising an authentication input generation unit configured to generate a character string for requesting input in accordance with the task condition displayed on a display device to prompt input. The information processing apparatus described.   And further comprising an extracting means for extracting one or more verification data whose reliability satisfies a predetermined condition based on a reliability included in an identification result by the operator identification model for each of the one or more verification data. The information processing apparatus according to claim 5, wherein the input generation unit generates a character string by preferentially using characters related to the extracted one or more pieces of verification data.   The recording apparatus according to claim 1, further comprising a recording unit that records a history of the character string displayed on the display device, wherein the authentication input generation unit manages the generation frequency of the character string based on the history. The information processing apparatus according to 5 or 6.   The information according to any one of claims 1 to 7, further comprising filtering means for excluding those related to an operation corresponding to a predetermined exclusion condition from operation information collected by the operator via the input device. Processing equipment.   The collection means generates a feature vector constituting each of the learning data and the verification data from operation information by an operator via the input device, and the information processing device responds to the task condition And a conversion means for converting the input operation into one or more authentication feature vectors, wherein the operator identification model receives a feature vector included in each of the plurality of learning data as an input, and a correct identification result The parameter adjusted by the machine learning algorithm is calculated so as to be output, and the input operation of the corresponding operator using the authentication feature vector or the verification feature vector as an input. The information according to any one of claims 1 to 8, characterized in that the result of identifying whether or not is output. Processing apparatus. An authentication system for authenticating an operator in an information processing device,
Collection means for collecting operation information by an operator via an input device of the information processing apparatus;
Learning means for learning an operator identification model for identifying an input operation of the operator related to the operation information via the input device using a plurality of learning data based on the collected operation information;
Using one or more verification data based on the collected operation information, a task condition acceptable for authentication based on the operator identification model and whether to permit the authentication are determined for an operator involved in learning. A determination means;
A determination unit that determines whether or not authentication based on the operator identification model is permitted for an operator involved in the request when requesting authentication based on the operator identification model in the information processing apparatus;
Whether or not to successfully authenticate the operator according to the request based on a plurality of identification results by the operator identification model of the operator according to the request with respect to the input operation in response to the task condition, if permitted An authentication system including an authentication determination means for determining whether or not. An authentication method for authenticating an operator,
A computer collecting operation information by an operator via an input device;
A step of learning an operator identification model for identifying an input operation of the operator related to the operation information via the input device using a plurality of learning data based on the collected operation information;
Whether a computer permits one or more task conditions that can be accepted in authentication based on the operator identification model for an operator involved in learning using one or more verification data based on the collected operation information, and whether or not to permit the authentication A step of determining
Determining whether or not authentication based on the operator identification model is permitted for an operator involved in the request when the computer requests authentication based on the operator identification model;
If permitted, the computer successfully authenticates the operator according to the request based on a plurality of identification results based on the operator identification model of the operator according to the request for the input operation in response to the task condition. And a step of determining whether or not to perform authentication.   The computer further includes a step of generating virtual pseudo operation information by an operator other than the operator involved in the learning based on the statistical information of the operation, and the learning step includes a plurality of steps based on the pseudo operation information. The authentication method according to claim 11, wherein the learning data is further used to learn the operator identification model. On the computer,
Collecting operation information by an operator via an input device;
Learning an operator identification model for identifying an input operation of the operator related to the operation information through the input device using a plurality of learning data based on the collected operation information;
Using one or more verification data based on the collected operation information, a task condition acceptable for authentication based on the operator identification model and whether to permit the authentication are determined for an operator involved in learning. Steps,
When requesting authentication based on the operator identification model, determining whether authentication based on the operator identification model is permitted for the operator involved in the request;
Whether or not to successfully authenticate the operator according to the request based on a plurality of identification results by the operator identification model of the operator according to the request with respect to the input operation in response to the task condition, if permitted A program for executing the step of determining whether or not.

Images