Search

KR-20260064207-A - ELECTRONIC DEVICE AND MODEL SELECTION METHOD THEREOF

KR20260064207AKR 20260064207 AKR20260064207 AKR 20260064207AKR-20260064207-A

Abstract

A method for selecting a model of an electronic device according to the present disclosure may include: identifying at least one model learned based on a first data set, a first data set associated with a source domain, and a second data set associated with a target domain; for each of the at least one model, acquiring at least one first feature corresponding to at least one first data included in the first data set, and acquiring at least one second feature corresponding to at least one second data included in the second data set; acquiring at least one third feature having a set dimension corresponding to at least one second feature based on at least one first feature; and selecting a target model among at least one model based on a first score calculated based on at least one first feature and at least one third feature.

Inventors

  • 안세현
  • 김기경
  • 안찬호

Assignees

  • 삼성전자주식회사

Dates

Publication Date
20260507
Application Date
20241031

Claims (20)

  1. As a method for selecting a model of an electronic device, A step of identifying at least one model trained based on a first dataset, said first dataset associated with a source domain, and a second dataset associated with a target domain; For each of the above at least one model, a step of acquiring at least one first feature corresponding to at least one first data included in the first data set, and acquiring at least one second feature corresponding to at least one second data included in the second data set; Based on the at least one first feature, a step of obtaining at least one third feature having a set dimension corresponding to the at least one second feature; and A method comprising the step of selecting a target model among the at least one model based on a first score calculated based on the at least one first feature and the at least one third feature. How to select an electronic device model.
  2. In Article 1, The step of acquiring at least one third feature is, A method comprising the step of obtaining at least one third feature by performing principal component analysis using a matrix composed of at least one first feature. How to select an electronic device model.
  3. In Article 2, The step of acquiring at least one third feature is, A step of identifying at least one eigenvalue of a set number corresponding to the set dimension among the eigenvalues calculated based on the above matrix; A step of identifying at least one eigenvector corresponding to the above at least one eigenvalue; and A method comprising the step of obtaining the at least one third feature by projecting the at least one second feature onto the at least one eigenvector. How to select an electronic device model.
  4. In Article 1, The dimensions set above are, Based on the similarity between the source domain and the target domain, How to select an electronic device model.
  5. In Paragraph 4, If the similarity between the source domain and the target domain is less than or equal to a set level, the set dimension is set to be smaller than the set value. How to select an electronic device model.
  6. In Article 1, The step of acquiring at least one third feature is, A step comprising acquiring at least one third feature by using an artificial intelligence model, How to select an electronic device model.
  7. In Article 1, The step of selecting the above target model is, A step of calculating a first value representing the diversity of at least one first feature through a predetermined operation for calculating the diversity of the feature; A step of calculating a second value representing the diversity of at least one third feature through the above predetermined operation; and A method comprising the step of selecting the target model based on the first score calculated based on the first value and the second value. How to select an electronic device model.
  8. In Article 7, The above first score is, Corresponding to the value obtained by dividing the above second value by the above first value, How to select an electronic device model.
  9. In Article 7, The above first value is, The variance of a first matrix composed of at least one first feature, and The above second value is, The variance of the second matrix composed of at least one third feature, How to select an electronic device model.
  10. In Article 1, The above at least one first data and the above at least one second data are, unlabeled data, How to select an electronic device model.
  11. In Article 1, For each of the models included in the model set, a step of identifying a second score representing the performance of the model calculated based on the first data set; For each of the models included in the above model set, a step of identifying information regarding resources to be required for training the model; and The method further comprises the step of identifying at least one model among the models included in the model set based on the second score and information about the resource. How to select an electronic device model.
  12. In Article 11, The above at least one model is, Among the models included in the above model set, including a model in which the second score is greater than or equal to a set value, How to select an electronic device model.
  13. In Article 1, A step of identifying information about a task to be performed on a user's terminal and information about resources available on the terminal; and The method further comprises the step of determining at least some of the models included in the model set as at least one model based on information about the above task and information about the above resources. How to select an electronic device model.
  14. In Article 1, A method further comprising the step of tuning the parameters of the target model based on the second data set. How to select an electronic device model.
  15. In Article 10, The third dataset, which is a dataset related to the target domain, is, Includes label data, A method further comprising the step of tuning the parameters of the target model based on the third data set. How to select an electronic device model.
  16. In Article 1, The above at least one first data is, It includes a set first number of data among the data included in the first data set above, and The above at least one second data is, It includes a set second number of data among the data included in the second data set above, and How to select an electronic device model.
  17. A computer-readable, non-transient recording medium having a program for executing the model selection method of claim 1 on a computer.
  18. As an electronic device for performing a model selection method, One or more processors; and It includes memory for storing one or more instructions executed by the above one or more processors, and The above one or more processors, by executing the above one or more instructions, Identifying at least one model trained based on a first dataset, said first dataset related to a source domain and said second dataset related to a target domain, and For each of the above at least one model, at least one first feature corresponding to at least one first data included in the first data set is obtained, and at least one second feature corresponding to at least one second data included in the second data set is obtained, and Based on the above at least one first feature, at least one third feature having a set dimension corresponding to the above at least one second feature is obtained, and Selecting a target model among the at least one model based on a first score calculated based on the at least one first feature and the at least one third feature. Electronic device.
  19. In Article 18, The above processor is, By performing principal component analysis using the matrix composed of the above at least one first feature, the above at least one third feature is obtained. Electronic device.
  20. As a method for selecting a model of an electronic device, A step of identifying information about a task to be performed on a user's terminal and information about resources available on said terminal; A step of determining at least one model among models included in a set of models composed of models learned based on a first dataset related to a source domain, based on information regarding the above-mentioned task and information regarding the above-mentioned resources; A step of identifying a second data set associated with the first data set and the target domain, wherein the first data set includes first unlabeled data of the source domain and the second data set includes second unlabeled data of the target domain; and A method comprising the step of selecting a target model among at least one model based on the first unlabeled data and the second unlabeled data. How to select an electronic device model.

Description

Electronic Device and Model Selection Method Thereof The present disclosure relates to an electronic device and a method for selecting the model thereof. Models trained on large amounts of data may be publicly available for use in tasks across various domains. In this regard, benchmark scores indicating the performance of these models may also be publicly available. Models with high benchmark scores generally demonstrate high performance in tasks across diverse domains. FIG. 1 is a drawing for explaining an electronic device according to one embodiment. FIG. 2 is a flowchart illustrating a method for selecting a model of an electronic device according to one embodiment. FIG. 3 is a flowchart for explaining in more detail a method for selecting a model of an electronic device according to one embodiment. FIG. 4 is a diagram illustrating a method for further training a target model into a model suitable for a specific domain by tuning the target model according to one embodiment. FIG. 5 is a diagram illustrating a method for determining a set of candidate models based on information about tasks to be performed on a user's terminal and information about resources available on the terminal. FIG. 6 is a diagram illustrating a graph showing the time required to select a model based on a model selection method according to one embodiment. The terms used in the embodiments have been selected to be as widely used as possible, taking into account their functions in the present disclosure; however, these may vary depending on the intent of those skilled in the art, case law, the emergence of new technologies, etc. Additionally, in specific cases, terms have been arbitrarily selected by the applicant, and in such cases, their meanings will be described in detail in the relevant explanatory section. Therefore, terms used in the present disclosure should be defined not merely by their names, but based on their meanings and the overall content of the present disclosure. When a part of a specification is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components. Furthermore, terms such as "...part" or "...module" as used in the specification refer to a unit that processes at least one function or operation, and this may be implemented in hardware or software, or as a combination of hardware and software. Embodiments of the present disclosure are described below with reference to the attached drawings so that those skilled in the art can easily implement them. However, the present disclosure may be embodied in various different forms and is not limited to the embodiments described herein. Embodiments of the present disclosure will be described in detail below with reference to the drawings. FIG. 1 is a drawing for explaining an electronic device according to one embodiment. Referring to FIG. 1, the electronic device (100) may include one or more processors (101) and memory (102). The electronic device (100) illustrated in FIG. 1 is illustrated only with components related to the present embodiment. Therefore, it can be understood by those skilled in the art related to the present embodiment that other general-purpose components may be included in addition to the components illustrated in FIG. 1. One or more processors (101) can control the overall operation of the electronic device (100). One or more processors (101) may be composed of at least one hardware unit. Additionally, one or more processors (101) may operate by one or more software modules generated by executing one or more instructions stored in memory (102). One or more processors (101) can control embodiments performed by the electronic device (100) through interaction with memory (102) and further components that the electronic device (100) may further include. According to one embodiment, one or more processors (101) identify at least one model learned based on a first data set, a first data set associated with a source domain, and a second data set associated with a target domain, and for each of the at least one model, acquire at least one first feature corresponding to at least one first data included in the first data set, acquire at least one second feature corresponding to at least one second data included in the second data set, acquire at least one third feature having a set dimension corresponding to at least one second feature based on at least one first feature, and select a target model among at least one model based on a first score calculated based on at least one first feature and at least one third feature. In the present specification, each of at least one model may be a pre-trained model based on a large amount of data. More specifically, each of at least one model may be a model trained by using a large amount of data as training data and by inputting a large amount of resources (e.g., resources of 10,000 GPU hours or more).