# US 2022/0044152 A1 · Signal selection device, learning device, and signal selection method and program > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0044152 A1 - Document ID: US20220044152A1 - Status: Published application - Kind: A1 - Application number: 17/423,915 - Filed: 2019-03-28 - Published: 2022-02-10 - Earliest claimed: 2019-03-28 - Assignees: Mitsubishi Electric Corporation - Applicants: Mitsubishi Electric Corporation - Inventors: Takahiko MASUZAKI; Osamu NASU - Web page: https://patentgrep.com/patent/US20220044152A1 ## CPC - G06N 20/00: Machine learning - G06F 2218/08: Feature extraction - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract A signal selection device (10) selects, from a plurality of candidate signals, a training signal for learning a model (20) usable for data analysis. The signal selection device (10) includes a first acquirer (11) that acquires the plurality of candidate signals and training labels associated with the plurality of candidate signals and being status values corresponding to results of analysis performed using the model, a feature calculator (13) that calculates one or more features for each of the plurality of candidate signals, and a selector (16) that selects the training signal from the plurality of candidate signals based on a degree of relevance between the one or more features and the training labels. ## Claims ### Claim 1 A learning device comprising: a signal selection device to select, from a plurality of candidate signals, a training signal for learning a model usable for data analysis, the signal selection device comprising a first acquirer to acquire the plurality of candidate signals and training labels associated with the plurality of candidate signals, the training labels being status values corresponding to results of analysis performed using the model, a feature calculator to calculate one or more features using the plurality of candidate signals, and a selector to select the training signal from the plurality of candidate signals based on a degree of relevance between the one or more features and the training labels; a learner to learn the model based on the training signal selected by the signal selection device; and a contribution outputter to output a degree of contribution indicating a degree of contribution of a plurality of the training signals selected by the signal selection device to label assignment performed using the model, wherein the learner repeats changing a value of each of the plurality of training signals to generate a new model, and the contribution outputter calculates and outputs the degree of contribution based on a comparison between the training labels and labels assigned to the plurality of training signals in accordance with the new model generated through the repeated changing. ### Claim 2 (depends on 1) The learning device according to claim 1, wherein the feature calculator calculates the one or more features using a partial signal extracted from each of the plurality of candidate signals for a predetermined interval. ### Claim 3 (depends on 1) The learning device according to claim 1, wherein the selector selects, as the training signal, a candidate signal corresponding to a feature with a higher degree of relevance to the training label than other features from the plurality of candidate signals. ### Claim 4 (depends on 1) The learning device according to claim 1, wherein further comprising: the signal selection device further comprises a provider to provide, to another device, the plurality of candidate signals and the one or more features in a manner associated with each other. ### Claim 5 (depends on 1) The learning device according to claim 1, wherein: the signal selection device further comprises a second acquirer to acquire a plurality of additional signals to be added as the plurality of candidate signals and one or more features associated with the plurality of additional signals. ### Claim at position 6 -8. (canceled) ### Claim 9 A learning method for selecting, from a plurality of candidate signals, a training signal for learning a model usable for data analysis and learning the model based on the training signal, the method comprising: calculating, by a feature calculator, a feature using the plurality of candidate signals; selecting, by a selector, the training signal from the plurality of candidate signals based on a degree of relevance between the feature and training labels being status values corresponding to results of analysis performed using the model; repeating, by a learner, changing a value of each of a plurality of the training signals to generate a new model; and calculating and outputting, by a contribution outputter, a degree of contribution based on a comparison between the training labels and labels assigned to the plurality of training signals in accordance with the new model generated through the repeated changing, the degree of contribution indicating a degree of contribution of the plurality of training signals selected in the selecting to label assignment performed using the model. ### Claim 10 A non-transitory computer-readable recording medium storing a program, the program causing a computer to select, from a plurality of candidate signals, a training signal for learning a model usable for data analysis and learning the model based on the training signal, the program causing the computer to perform operations comprising: calculating a feature using the plurality of candidate signals; selecting the training signal from the plurality of candidate signals based on a degree of relevance between the feature and training labels being status values corresponding to results of analysis performed using the model repeating changing a value of each of a plurality of the training signals to generate a new model; and calculating and outputting a degree of contribution based on a comparison between the training labels and labels assigned to the plurality of training signals in accordance with the new model generated through the repeated changing, the degree of contribution indicating a degree of contribution of the selected plurality of training signals to label assignment performed using the model.