Information processing device, parameter estimation method, and program
The information processing device converts user prompts into numerical parameters using a calculation unit with feedback mechanisms to optimize network resource allocation, addressing the challenge of dynamic user demands and ensuring efficient resource allocation.
Patent Information
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- NIPPON TELEGRAPH & TELEPHONE CORP
- Filing Date
- 2024-11-12
- Publication Date
- 2026-05-22
AI Technical Summary
Conventional network resource allocation technologies fail to accurately reflect user requirements due to their assumption that user needs are given numerically, leading to inefficient resource allocation and inability to adapt to changing user demands.
An information processing device with a calculation unit that converts user prompts into numerical parameters through a request classification model, language model, and user request parameter conversion model, adjusting parameters based on user feedback to optimize resource allocation.
Enables accurate reflection of user requests in resource allocation optimization, adaptively adjusting to changing demands, preventing unrealistic allocations, and reducing computational costs through parallel processing and feedback-based adjustments.
Smart Images
Figure 2026085000000001_ABST
Abstract
Description
[Technical Field]
[0001] This invention relates to network resource allocation technology. [Background technology]
[0002] Various network services are provided to users through ICT (Information and Communication Technology) infrastructure.
[0003] To ensure stable operation within the ICT infrastructure, such as avoiding congestion, while satisfying the diverse needs of network service users (reduced latency, increased storage and CPU capacity), it is necessary to appropriately allocate ICT resources on the ICT infrastructure. Furthermore, since user needs change constantly, it is necessary to continuously adjust the allocation of ICT resources to facilitate the fulfillment of evolving needs and to accommodate even vague requests.
[0004] To address the above-mentioned demands, several conventional technologies have been proposed that formulate ICT resource allocation as an optimization problem, given network conditions such as traffic volume and latency that meet user requirements (for example, Non-Patent Document 1). [Prior art documents] [Non-patent literature]
[0005] [Non-Patent Document 1] Wissal Attaoui etl., VNF and CNF Placement in 5G:Recent Advances and Future Trends, IEEE TRANSACTIONS ON NETWORK AND SERVICE MANAGEMENT, VOL. 20, NO. 4, DECEMBER 2023 [Overview of the Initiative] [Problems that the invention aims to solve]
[0006] However, the above methods using conventional technology assume that network conditions that satisfy the user's requirements are given, but the user's requirements cannot always be directly obtained as numerical parameters.
[0007] If user requests cannot be obtained as numerical parameters, they cannot be properly reflected in the optimization problem of ICT resource allocation, and user requests cannot be met. Furthermore, since user requests change over time, a system that efficiently learns user requests is necessary.
[0008] This invention has been made in view of the above points, and aims to provide a technology for estimating appropriate parameters from user requirements. [Means for solving the problem]
[0009] According to the disclosed technology, an information processing device for estimating parameters used to perform resource allocation, A calculation unit that estimates the parameters from prompts representing user requests and adjusts the parameters based on user feedback regarding those parameters. An information processing device equipped with [this feature] is provided. [Effects of the Invention]
[0010] According to the disclosed technology, a technique is provided for estimating appropriate parameters from user requirements. [Brief explanation of the drawing]
[0011] [Figure 1] This figure shows an example configuration of the information processing device 100. [Figure 2] This figure shows an example of the functional configuration of the arithmetic unit 120. [Figure 3] This is a diagram illustrating the general operation of the arithmetic unit 120. [Figure 4] It is a flowchart for explaining the operation of the calculation unit 120. [Figure 5] It is a diagram showing an example of the hardware configuration of the device.
Embodiments for Carrying Out the Invention
[0012] Hereinafter, embodiments of the present invention (these embodiments) will be described with reference to the drawings. The embodiments described below are merely examples, and the embodiments to which the present invention is applied are not limited to the following embodiments.
[0013] (Example of Device Configuration) FIG. 1 shows a configuration example of an information processing apparatus 100 that performs resource allocation for a network (including servers and the like) in these embodiments. The information processing apparatus 100 includes a mechanism that converts a prompt representing a user's request (including intention) into a numerical parameter, and performs resource allocation through feedback from the user in order to respond to the user's requests that change moment by moment.
[0014] As shown in FIG. 1, the information processing apparatus 100 includes an input / output unit 110 and a calculation unit 120. Information necessary for the calculation by the calculation unit 120 is input from the input / output unit 110, and the input information is passed to the calculation unit 120. The calculation unit 120 performs calculation processing for resource allocation and passes the processing result to the input / output unit 110. The input / output unit 110 outputs the calculation result of the resource allocation.
[0015] More specifically, the calculation unit 120 performs calculation of resource allocation by solving a resource optimization problem for a certain user. Here, a parameter Q that reflects the user's request is used. i Let i be an index related to the type of request. Examples of the type of request include changes in resource capacity, changes in delay time, and the like. The calculation unit 120 dynamically estimates Q while solving the resource optimization problem based on feedback from multiple conversations (chats) with the user via the input / output unit 110. i
[0016] (Details of the calculation unit 120) Figure 2 shows an example of the functional configuration of the calculation unit 120. As shown in Figure 2, the calculation unit 120 includes a request classification model 121, a language model 122, a user request parameter conversion model 123, a request parameter adjustment model 124, a resource optimization problem solving unit 125, and a control unit 126.
[0017] The request classification model 121, language model 122, user request parameter conversion model 123, and request parameter adjustment model 124 may also be referred to as the classification unit 121, language processing unit 122, conversion unit 123, and adjustment unit 124, respectively. Furthermore, the resource optimization problem solving unit 125 may be provided outside the information processing device 100.
[0018] Referring to Figure 3, the operation overview of the arithmetic unit 120 having the above configuration will be explained. Figure 3 also shows a user 200. User 200 includes, for example, a terminal held by the user. User 200 may also be a concept that includes the user's network (the target of resource allocation). The control unit 126 is a functional unit that controls repetitive processing and waiting time, and the control by the control unit 126 is explained in Figure 4, which will be described later.
[0019] <S1(ステップ1)> The request classification model 121 receives user requests (prompts regarding requests) from user 200. The request classification model 121 classifies the prompts regarding requests according to the type of request. In other words, the request classification model 121 classifies requests according to the input parameters of the resource optimization problem. Figure 3 shows that processing is performed for each classified request (e.g., requests regarding resource capacity, requests regarding delay conditions). Note that a machine learning model such as a support vector machine can be used as the request classification model 121.
[0020] <s2> The language model 122 receives prompts regarding requests as input. The language model 122 converts the prompts regarding requests into embeddable vectors (called embedding vectors) and outputs these embedding vectors. The language model 122 is not limited to a specific language model, but for example, the BERT model can be used as the language model 122.
[0021] <s3> The user request parameter conversion model 123 uses the embedding vector obtained by the language model 122 as parameter Q i Convert to parameter Q. i These are, for example, numerical parameters. As a user-requested parameter conversion model 123, for example, a machine learning model of a neural network can be used.
[0022] <s4> The request parameter adjustment model 124 determines the feasibility of the resource allocation problem for the user request parameters converted in S3. In determining the feasibility of the resource allocation problem, the resource optimization problem solving unit 125 performs the solution.
[0023] The feasibility of a resource allocation problem indicates whether or not resource allocation can be performed based on the user request parameters. If feasibility is not met, the request parameter adjustment model 124 adjusts the values of the request parameters through feedback from the user 200.
[0024] <s5> The resource optimization problem solving unit 125 includes the request parameters adjusted in S4 (parameter Q reflecting the estimated user's requests). i The following is input: The resource optimization problem solving unit 125 calculates the parameter Q that reflects the estimated user's requirements. i Based on this, we will attempt to solve the resource optimization problem.
[0025] <s6> The resource allocation result from the resource optimization problem solving unit 125 is notified to the user 200, and after the resource allocation is reflected in the user 200's system, the calculation unit 120 obtains the user's request again and repeats steps S1 to S5. In this way, the calculation unit 120 calculates the resource allocation while estimating the request parameters.
[0026] <s7> The calculation unit 120 checks for any changes in user requests after a predetermined (or dynamically changed) waiting time has elapsed. If there are any changes in user requests, it repeats steps S1 to S6. If there are no changes in user requests, it terminates the process.
[0027] (Examples) Next, as an example, a more specific example of the operation of the calculation unit 120 will be explained with reference to the flowchart in Figure 4.
[0028] In this embodiment, we assume a situation where user 200 prompts us with the question, "Could you increase the amount of my storage and reduce the latency of my network?" regarding the amount of storage and acceptable latency. At this time, the arithmetic unit 120 estimates the parameters "requested storage capacity Q1" and "acceptable latency Q2," which reflect the user's request, while solving the resource optimization problem.
[0029] In this embodiment, we have optimized prompts related to storage volume and latency, but optimization is not limited to these. Similar optimization is possible for prompts related to other elements (e.g., CPU, memory, network bandwidth, GPU).
[0030] First, the calculation unit 120 sets the initial values of the user-requested parameters with time k=0.
[0031] In S11, the request classification model 121 classifies the input prompts into two types: "increase the amount of my storage," which is a request prompt regarding the amount of storage, and "reduce the latency of my network," which is a request prompt regarding the acceptable latency.
[0032] In S12, for each requirement prompt classified in S11, the language model 122 converts the requirement prompt into an embedding vector e i , i ∈ {1, 2}.
[0033] In S13, based on the embedding vector e i obtained in S12, the user requirement parameter conversion model 123 calculates the change amount ΔQ i , i ∈ {1, 2} of Q i , i ∈ {1, 2}.
[0034] In S14, the requirement parameter adjustment model 124 uses the change amount ΔQ i to update the parameter as "Q i (k + 1) ← Q i (k) + ΔQ i ". When proceeding from S16 to S14 described later, the change amount ΔQ i is the adjusted (executable) change amount.
[0035] Given Q i (k + 1), the resource optimization problem solver 125 solves the resource allocation problem of the network. In solving the resource allocation problem, for example, for a network topology composed of a plurality of nodes and links, parameters x indicating the resource allocation destination and parameters y representing the communication path between the user 200 and the resource allocation node are obtained. This problem is given as an integer programming problem and can be solved by an existing solver.
[0036] In S15, the requirement parameter adjustment model 124 determines the feasibility of the resource allocation problem. If the requirement parameter adjustment model 124 determines that the optimization problem does not meet the feasibility with the current Q i (k + 1), in S16, it adjusts Q i (k + 1) through feedback with the user 200. The adjustment here is the change amount ΔQ i This may involve adjustments to the parameters. Specifically, for example, multiple parameter candidates that result in a Pareto optimal solution are presented to the user 200, and the value selected by the user 200 is obtained. Steps S14 to S16 may be performed multiple times until the feasibility is satisfied.
[0037] If S15 is answered with Yes (feasibility is met), the process proceeds to S17. In S17, the arithmetic unit 120 notifies the user 200 of the resource allocation calculation result obtained through the solution via the input / output unit 110, and the user 200 executes the resource allocation for the network. The arithmetic unit 120 receives feedback from the user 200 regarding the execution result.
[0038] In S18, the control unit 126 in the calculation unit 120 determines the user's satisfaction level based on the feedback. If the determination result is No (for example, if a new request prompt requiring a parameter change is received), the process proceeds to S20. In S20, k ← k+1 is set, and the process proceeds to S21. If there is no timeout, in S22, the calculation unit 22 receives a new request prompt from the user 200. Note that if a new request prompt has already been received in S18, S22 may be omitted. After that, the process from S11 is repeated.
[0039] In S18, if the control unit 126 determines that the user 200's request has been satisfied by the user 200, that is, if it determines ΔQ in response to the request prompt, i If it is estimated that = 0, the process proceeds to S19. In S19, the control unit 126 waits for the specified waiting time, then proceeds to S20. In other words, the process for updating the parameter is stopped until the predetermined waiting time has elapsed. After S20, if it is determined in S21 that a timeout has occurred, the process terminates.
[0040] The waiting time may be a fixed time predetermined, or the control unit 126 may set the waiting time to increase with each repetition of the cycle.
[0041] (Example hardware configuration) The information processing device 100 described in this embodiment can be realized, for example, by having a computer execute a program. This computer may be a physical computer or a virtual machine on the cloud.
[0042] In other words, the information processing device 100 can be realized by using hardware resources such as the CPU and memory built into the computer to execute a program corresponding to the processing performed by the information processing device 100. The above program can be recorded on a computer-readable recording medium (such as portable memory), saved, and distributed. It is also possible to provide the above program via a network such as the Internet or email.
[0043] Figure 5 shows an example of the hardware configuration of the computer described above. The computer in Figure 5 has a drive device 1000, an auxiliary storage device 1002, a memory device 1003, a CPU 1004, an interface device 1005, a display device 1006, an input device 1007, an output device 1008, etc., all of which are interconnected by bus B. The computer may also be equipped with a GPU.
[0044] The program that enables processing on the computer is provided, for example, on a recording medium 1001 such as a CD-ROM or memory card. When the recording medium 1001 containing the program is set in the drive device 1000, the program is installed from the recording medium 1001 to the auxiliary storage device 1002 via the drive device 1000. However, the program does not necessarily have to be installed from the recording medium 1001; it may also be downloaded from another computer via a network. The auxiliary storage device 1002 stores the installed program as well as necessary files and data.
[0045] The memory device 1003 reads and stores a program from the auxiliary storage device 1002 when a program startup command is received. The CPU 1004 implements the functions related to the information processing device 100 according to the program stored in the memory device 1003. The interface device 1005 is used as an interface for connecting to a network, etc. The display device 1006 displays a GUI (Graphical User Interface) etc., generated by a program. The input device 1007 consists of a keyboard and mouse, buttons, or a touch panel, etc., and is used to input various operation commands. The output device 1008 outputs the calculation results.
[0046] (Summary of the embodiments, effects, etc.) As described above, the technology described in this embodiment makes it possible to appropriately reflect user requests in resource allocation optimization problems by introducing a mechanism that converts prompts representing user requests into numerical parameters.
[0047] Furthermore, by incorporating a function to process user feedback, it becomes possible to efficiently estimate parameters even if user requests change over time.
[0048] Furthermore, by subdividing user prompts by type, it becomes possible to train and process the model in parallel for each subdivided user request, significantly reducing computational and implementation costs.
[0049] Furthermore, by introducing a request parameter adjustment model, it becomes possible to prevent resource allocation based on unreasonable requests, including those resulting from user slip-ups, through feedback from users.
[0050] Furthermore, it is conceivable to introduce large-scale language models (LLMs), which are expected to develop even further in the future, as language models. It is impossible to prevent uncertainty in the output of language models, such as hallucination, which is well known in generative AI. However, even in such cases, separating the "resource optimization" function can guarantee that unrealistic resource allocations will not occur.
[0051] The following additional information is disclosed regarding the embodiments described above.
[0052] <Note> (Additional note 1) An information processing device for estimating parameters used for resource allocation, A calculation unit that estimates the parameters from prompts representing user requests and adjusts the parameters based on user feedback regarding those parameters. An information processing device equipped with the following features. (Additional note 2) The calculation unit classifies the prompt into a plurality of prompts and performs parameter estimation for each classified prompt. The information processing device described in Appendix 1. (Additional note 3) The calculation unit adjusts the parameters based on the user's feedback if the parameters do not satisfy the feasibility of resource allocation. The information processing device described in Appendix 1. (Additional note 4) The calculation unit adjusts the parameters based on the user's feedback regarding the results of resource allocation. The information processing device described in Appendix 1. (Additional note 5) The calculation unit, when it is presumed that the user's request has been satisfied, stops processing for updating parameters until a predetermined waiting time has elapsed. The information processing device described in Appendix 1. (Additional note 6) A parameter estimation method performed by an information processing device that estimates parameters used for resource allocation, The parameters are estimated from prompts representing user requests, and the parameters are adjusted based on user feedback regarding those parameters. Parameter estimation method. (Additional note 7) A non-temporary storage medium that stores a program for causing a computer to function as the arithmetic unit in any one of the appendices 1 to 5.
[0053] Although this embodiment has been described above, the present invention is not limited to this specific embodiment, and various modifications and changes are possible within the scope of the gist of the invention as described in the claims. [Explanation of Symbols]
[0054] 100 Information Processing Devices 110 Input / output section 120 Arithmetic section 121 Requirements Classification Model 122 Language Models 123 User Request Parameter Conversion Model 124 Request Parameter Adjustment Model 125 Resource Optimization Problem Solving Unit 126 Control Unit 1000 drive unit 1001 Recording media 1002 Auxiliary storage 1003 Memory device 1004 CPU 1005 Interface device 1006 Display device 1007 Input device 1008 Output device
Claims
1. An information processing device for estimating parameters used for resource allocation, A calculation unit that estimates the parameters from prompts representing user requests and adjusts the parameters based on user feedback regarding those parameters. An information processing device equipped with the following features.
2. The calculation unit classifies the prompt into a plurality of prompts and performs parameter estimation for each classified prompt. The information processing apparatus according to claim 1.
3. The calculation unit adjusts the parameters based on the user's feedback if the parameters do not satisfy the feasibility of resource allocation. The information processing apparatus according to claim 1.
4. The calculation unit adjusts the parameters based on the user's feedback regarding the results of resource allocation. The information processing apparatus according to claim 1.
5. The calculation unit, when it is presumed that the user's request has been satisfied, stops processing for updating parameters until a predetermined waiting time has elapsed. The information processing apparatus according to claim 1.
6. A parameter estimation method performed by an information processing device that estimates parameters used for resource allocation, The parameters are estimated from prompts representing user requests, and the parameters are adjusted based on user feedback regarding those parameters. Parameter estimation method.
7. A program for causing a computer to function as an arithmetic unit in an information processing device according to any one of claims 1 to 5.