# US 2022/0044121 A1 · Training device, inferring device, training method, inferring method, and non-transitory computer readable medium > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2022/0044121 A1 - Document ID: US20220044121A1 - Status: Published application - Kind: A1 - Application number: 17/508,592 - Filed: 2021-10-22 - Published: 2022-02-10 - Earliest claimed: 2019-04-24 - Assignees: Preffered Networks, Inc. - Applicants: Preffered Networks, Inc. - Inventors: Kaushalya Madhawa PITUWALAKANKANAMGE; Kosuke NAKAGO; Katsuhiko ISHIGURO - Web page: https://patentgrep.com/patent/US20220044121A1 ## CPC - G06N 3/08: Learning methods - G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting - G06N 3/045: Combinations of networks - G06N 3/0455: Auto-encoder networks; Encoder-decoder networks - G06N 3/0475: Generative networks - G06N 3/09: Supervised learning - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) ## Abstract A training device comprises one or more memories and one or more processors. The one or more processors are configured to train a first converter that converts a first feature amount regarding a node of a graph and a second feature amount regarding a structure of the graph into a first latent value through conversion capable of defining inverse conversion, and a second converter that converts the second feature amount into a second latent value through conversion capable of defining inverse conversion, based on the first latent value and the second latent value. ## Claims ### Claim 1 A training device comprising: one or more memories; and one or more processors configured to train a first converter that converts a first feature amount regarding a node of a graph and a second feature amount regarding a structure of the graph into a first latent value through conversion capable of defining inverse conversion, and a second converter that converts the second feature amount into a second latent value through conversion capable of defining inverse conversion, based on the first latent value and the second latent value. ### Claim 2 (depends on 1) The training device according to claim 1, wherein: the first converter includes a first neural network model; the second converter includes a second neural network model; and the one or more processors train the first neural network model and the second neural network model. ### Claim 3 (depends on 1) The training device according to claim 1, wherein the one or more processors make the first converter and the second converter perform conversion based on nonlinear mapping. ### Claim 4 (depends on 1) The training device according to claim 1, wherein the one or more processors make the first converter and the second converter execute mapping based on NVP (Non-volume Preserving) mapping. ### Claim 5 (depends on 4) The training device according to claim 4, wherein: the one or more processors make the first converter calculate the first latent value by applying the NVP mapping plural times on the first feature amount and the second feature amount; and the one or more processors make the second converter calculate the second latent value by applying the NVP mapping plural times on the second feature amount. ### Claim 6 (depends on 1) The training device according to claim 1, wherein: the first feature amount is a first tensor indicating a node feature amount of the graph; and the second feature amount is a second tensor indicating the structure of the graph. ### Claim 7 (depends on 6) The training device according to claim 6, wherein the one or more processors further process acquired data regarding the graph into the first tensor and the second tensor including an adjacency matrix of the graph. ### Claim 8 (depends on 1) The training device according to claim 1, wherein the one or more processors train the first converter and the second converter based on a prior distribution with respect to the first latent value and the second latent value. ### Claim 9 An generation device comprising: one or more memories; and one or more processors configured to: convert second data into a second feature amount regarding a structure of a first graph by using a second inverse converter; and convert first data into a first feature amount regarding a node of the first graph by using a first inverse converter and the second feature amount. ### Claim 10 (depends on 9) The generation device according to claim 9, wherein: the first data is obtained when the one or more processors convert a first feature amount regarding a node of a second graph and a second feature amount regarding a structure of the second graph into a first latent value by using a first converter capable of defining inverse conversion; and the second data is obtained when the one or more processors convert the second feature amount regarding the structure of the second graph into a second latent value by using a second converter capable of defining inverse conversion. ### Claim 11 (depends on 10) The generation device according to claim 10, wherein: the one or more processors make the first inverse converter execute inverse conversion of the first converter; and the one or more processors make the second inverse converter execute inverse conversion of the second converter. ### Claim 12 (depends on 9) The generation device according to claim 9, wherein the first data and the second data are generated based on random numbers. ### Claim 13 (depends on 9) The generation device according to claim 9, wherein the one or more processors further apply noise to at least one of the first data and the second data before converting the first data and the second data. ### Claim 14 (depends on 10) The generation device according to claim 10, wherein the first graph and the second graph are the same graph. ### Claim 15 (depends on 9) The generation device according to claim 9, wherein the one or more processors further generate data of the first graph based on the first feature amount converted by using the first inverse converter and the second feature amount converted by using the second inverse converter. ### Claim 16 (depends on 9) The generation device according to claim 9, wherein: a learned model included in the first inverse converter is a model trained based on a prior distribution with respect to a latent value in inverse conversion of the conversion executed by the first inverse converter; and a learned model included in the second inverse converter is a model trained based on a prior distribution with respect to a latent value in inverse conversion of the conversion executed by the second inverse converter. ### Claim 17 (depends on 16) The generation device according to claim 16, wherein: the learned model included in the first inverse converter is a neural network model; and the learned model included in the second inverse converter is a neural network model. ### Claim 18 (depends on 9) The generation device according to claim 9, wherein: the first feature amount regarding the node of the first graph is a first tensor indicating a node feature amount of the first graph; and the second feature amount regarding the structure of the first graph is a second tensor indicating the structure of the first graph. ### Claim 19 A training method comprising training, by one or more processors, a first converter that converts a first feature amount regarding a node of a graph and a second feature amount regarding a structure of the graph into a first latent value through conversion capable of defining inverse conversion, and second converter that converts the second feature amount into a second latent value through conversion capable of defining inverse conversion, based on the first latent value and the second latent value. ### Claim 20 An generation method comprising: performing, by one or more processors, second inverse conversion on second data into a second feature amount regarding a structure of a first graph; and performing, by the one or more processors, first inverse conversion on first data into a first feature amount regarding a node of the first graph by using the second feature amount.