Search

KR-20260062744-A - Satellite type task-intention multi-modal Quantum federated learning system and method and computing device for executing the same

KR20260062744AKR 20260062744 AKR20260062744 AKR 20260062744AKR-20260062744-A

Abstract

The present invention relates to a mission-oriented multimodal quantum federated learning system and method for a satellite and a computing device for performing the same. The learning system of the present invention includes a central server, and the central server includes a bit-based neural network model for preprocessing data generated from a satellite and a quantum-based neural network model for determining the attitude control of the satellite based on the preprocessed data.

Inventors

  • 홍충선
  • 박유민

Assignees

  • 경희대학교 산학협력단

Dates

Publication Date
20260507
Application Date
20241029

Claims (12)

  1. In a mission-oriented multimodal quantum federated learning system for microsatellites, The above learning system is, It includes one or more processors; and a central server having memory for storing one or more programs executed by said one or more processors, and The above central server is, A bit-based neural network model for preprocessing data generated from satellites; and A mission-oriented multimodal quantum federated learning system for a satellite, characterized by including a quantum-based neural network model that determines attitude control of the satellite based on the preprocessed data.
  2. In paragraph 1, The above bit-based neural network model An image encoder that receives image-based training data generated from the above satellite; An image converter that receives vibration-based learning data generated from the above satellite; A vector encoder that receives attitude and current state data and sensing information data of the satellite generated from the satellite; A multimodal embedding synthesis network for representing the multimodal input of the above satellite as compressed information; A mission-oriented multimodal quantum federated learning system for a satellite, characterized by including a multi-head activation function hidden layer that extracts information to be used for correlation or optimization between data transformed from the above-mentioned embedding synthesis network.
  3. In paragraph 2, The above multimodal embedding synthesis network is A mission-oriented multimodal quantum federated learning system for satellites, characterized by comprising a multimodal multi-head attention module that processes multiple input data in parallel to transform them into a single vector, and an attention embedding layer that processes information to grasp the implied meaning from the data transformed by the multimodal multi-head attention module.
  4. In paragraph 2, The above quantum-based neural network model An embedding circuit that receives information output from the above multi-head activation function-based hidden layer and converts quantum-based data into satellite information signals into quantum-based data; A parameterized quantum circuit that processes quantum-based data from the above-mentioned embedding circuit, extracts features and patterns, and performs optimization of attitude control; and A measuring device that converts the above quantum-based data into actual bit-based data; A mission-oriented multimodal quantum federated learning system for a satellite characterized by including a final attitude control vector unit according to the mission of the satellite through the above measuring instrument.
  5. In paragraph 1, The above central server is, A learning network model that receives training data and learns satellite control; and It further includes a module for deriving lane values based on numerical optimization through the above training data, A mission-oriented multimodal quantum federated learning system for satellites characterized by adding the Mean Squared Error (MSE) of the output of the learning network model and the output of the lane value derivation module to the loss value to derive a final loss value, and updating the parameters of the learning network model based on the final loss value.
  6. In paragraph 1, The above learning system is, A mission-oriented multimodal quantum federated learning system for a satellite, characterized by further including an information unit representing mission information of the satellite between the bit-based neural network model and the quantum-based neural network model.
  7. In paragraph 1, A mission-oriented multimodal quantum federated learning system for satellites, characterized in that the above satellites are divided into bit-based satellites and quantum-based satellites, and exchange information with each other and perform knowledge cooperation during the learning or inference process, respectively.
  8. One or more processors, and A method performed on a computing device mounted on a bit-based satellite having a memory for storing one or more programs executed by the above one or more processors, wherein A step of training the bit-based neural network model to input data generated from the satellite into the bit-based neural network model for preprocessing; and A mission-oriented multimodal quantum federated learning method for a satellite, characterized by including the step of transmitting data preprocessed from the bit-based neural network model to a quantum-based satellite to determine the attitude control of the satellite through the quantum-based neural network model.
  9. In paragraph 8, The step of performing the final attitude control of the above satellite is, A step of receiving a multimodal input consisting of image-based learning data, vibration-based learning data, attitude and current state data of the satellite, and sensing information data generated from the satellite; and A step of representing the multimodal input of the above satellite as compressed information through a multimodal embedding synthesis network; A step of extracting information to be used for the relationship between the data transformed in the above step or for optimization; A step of receiving information output from the above step and converting the satellite's information signal into quantum-based data; A step of processing the quantum-based data converted in the above step, extracting features and patterns, and performing optimization of attitude control; A mission-oriented multimodal quantum federated learning method for a satellite, comprising the step of converting quantum-based data processed in the above step into actual bit-based data to perform final attitude control according to the mission of the satellite.
  10. One or more processors, and A method performed on a computing device mounted on a quantum-based satellite having a memory for storing one or more programs executed by the above-mentioned one or more processors, wherein A step of receiving data preprocessed by said bit-based neural network model from a satellite equipped with said bit-based neural network model; and A mission-oriented multimodal quantum federated learning method for a satellite, characterized by including a step of learning a quantum-based neural network model to determine the attitude control of the satellite based on the preprocessed data.
  11. One or more processors; Memory; and Includes one or more programs, The above one or more programs are configured to be stored in the memory and executed by the above one or more processors, and are a computing device mounted on a bit-based satellite, The above one or more programs are, Instructions for training the bit-based neural network model to input data generated from the satellite into the bit-based neural network model for preprocessing; and A computing device comprising a command to transmit data preprocessed from the bit-based neural network model to a quantum-based satellite to determine the attitude control of the satellite through the quantum-based neural network model.
  12. One or more processors; Memory; and Includes one or more programs, The above one or more programs are configured to be stored in the memory and executed by the above one or more processors, and are a computing device mounted on a quantum-based satellite, The above one or more programs are, A command for receiving data preprocessed by said bit-based neural network model from a satellite equipped with said bit-based neural network model; and A computing device comprising instructions for training a quantum-based neural network model to determine the attitude control of the satellite based on the above-mentioned preprocessed data.

Description

Satellite-type task-intention multi-modal quantum federated learning system and method and computing device for executing the same The present invention relates to a mission-oriented multimodal quantum federated learning system and method for satellites. Federated learning, performed in conventional computing, is a machine learning technique in which multiple terminals and a single server cooperate to learn a global model. Here, the terminals can be, for example, Internet of Things (IoT) devices or smartphones. This conventional federated learning has the advantage of overcoming the shortage of training samples required to learn from a limited amount of local data. Furthermore, the efficiency of conventional federated learning increases as the number of terminals—which are heterogeneous devices connected wirelessly—increases. An on-board model mounted on a satellite is, for example, an AI-based object detection model for images of objects (e.g., airplanes) captured by a satellite optical camera installed on the satellite. On-board model training on satellites is required for purposes such as data transmission constraints, real-time data processing, and security, but there are limitations to training existing deep learning models in the current computing-constrained environment on satellites. In addition, since the data generated from the satellite is various types of multimodal data, a preprocessing module was required to process it. Furthermore, existing satellite attitude control systems have the disadvantage of being optimized for specific missions, making it impossible to perform various missions simultaneously. When switching attitude control based on mission changes, firmware updates must be performed directly via ground stations; the resulting communication delays or overhead pose a risk of causing significant problems during satellite mission execution. On-board model training on satellites is required for purposes such as data transmission constraints, real-time data processing, and security, but existing deep learning model training has limitations in the current computing-constrained environment on satellites. Furthermore, there is still a lack of data preprocessing algorithms capable of effectively performing low-resource fusion processing on satellites for multimodal data generated from satellites, and due to unstable communication environments such as sensor errors caused by solar wind, there is an urgent need for a cooperative learning system between satellites that can use autonomously collected data for training. FIG. 1 is a schematic diagram illustrating a quantum federated learning system according to an embodiment of the present invention. FIG. 2 is a schematic diagram of a mission-oriented multimodal quantum federated learning system for a microsatellite in one embodiment of the present invention. FIG. 3 is a block diagram showing the configuration of a central server according to one embodiment of the present invention. Figure 4 is a photograph of a deep learning network structure for preprocessing multimodal data. FIG. 5 is a block diagram showing the sequence of learning attitude control through training data. FIG. 6 is a flowchart schematically illustrating a quantum association learning method according to an embodiment of the present invention. Figure 7 is a flowchart for the steps to obtain final attitude control of the satellite. FIG. 8 is a flowchart showing the steps for updating to control the attitude of a satellite. Figure 9 is a diagram illustrating cooperative learning between satellites for updating. Figure 10 is a photograph showing the knowledge cooperation process of a bit-based satellite and a quantum-based satellite. FIG. 11 is a block diagram illustrating a computing environment including a computing device suitable for use in exemplary embodiments. Hereinafter, specific embodiments of the present invention will be described with reference to the drawings. The following detailed description is provided to facilitate a comprehensive understanding of the methods, apparatuses, and/or systems described herein. However, this is merely illustrative and the present invention is not limited thereto. In describing the embodiments of the present invention, detailed descriptions of known technologies related to the present invention are omitted if it is determined that such detailed descriptions may unnecessarily obscure the essence of the invention. Furthermore, the terms described below are defined in consideration of their functions within the present invention, and these definitions may vary depending on the intentions or practices of the user or operator. Therefore, such definitions should be based on the content throughout this specification. Terms used in the detailed description are intended merely to describe the embodiments of the present invention and should not be limiting in any way. Unless explicitly stated otherwise, expressions in the singular form include the