# US 12,183,054 B2 · Feature vector generation using neural networks > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,183,054 B2 - Document ID: US12183054B2 - Status: Granted - Kind: B2 - Application number: 17/770,674 - Filed: 2019-10-21 - Published: 2022-11-17 - Granted: 2024-12-31 - Earliest claimed: 2019-10-21 - Assignees: Telefonaktiebolaget LM Ericsson (publ) - Applicants: Telefonaktiebolaget LM Ericsson (publ) - Inventors: Bin Sun; Rafia Inam; Yifei Jin - Web page: https://patentgrep.com/patent/US12183054B2 ## CPC - G06N 3/08: Learning methods - G06V 10/454: {Integrating the filters into a hierarchical structure, e.g. convolutional neural networks [CNN]} - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/0495: Quantised networks; Sparse networks; Compressed networks - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract The present invention enables accelerated feature vector crafting with a high level of trustworthiness. Heretofore, a system (10) is set up from a data source verification apparatus (12) for verifying the at least one data source (14-1, . . . , 14-n) that generates the new specimen data and a feature vector crafting apparatus (18) for crafting at least one new feature vector from the new specimen data. The feature vector crafting apparatus (18) receiving data after verification of data sources (14-1, . . . , 14-n). The feature vector crafting apparatus (18) automatically generates new feature vectors from new data for forwarding there to an artificial intelligence infrastructure (20). ## Claims ### Claim 1 A method of verifying at least one data source that generates new specimen data intended for use in a machine learning infrastructure, the method comprising: obtaining at least one feature vector representing the new specimen data at a data source verifying apparatus; executing a correlation check between the at least one feature vector and at least one reference feature vector representing trusted specimen data for determination of a correlation measure; and registering the at least one data source as at least one trusted data source at a subscription management apparatus when the correlation measure is below a first predetermined threshold and the new specimen data is different from the trusted specimen data. ### Claim 2 A data source verifying apparatus for verifying at least one data source that generates new specimen data intended for use in a machine learning infrastructure, the data source verifying apparatus comprising processing circuitry and a memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to: obtain at least one feature vector representing the new specimen data at the data source verification apparatus; execute a correlation check between the at least one feature vector and at least one reference feature vector representing trusted specimen data for determination of a correlation measure; and to register the at least one data source as trusted data source at a subscription management apparatus when the correlation measure is below a first predetermined threshold and the new specimen data is different from the trusted specimen data. ### Claim 3 (depends on 2) The data source verifying apparatus according to claim 2, the memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to: receive a plurality of feature vectors representing the new specimen data as anchor data; evaluate whether the plurality of feature vectors have a continuous occurrence; and conclude on the new specimen data being related to a new object upon continuous occurrence of the plurality of feature vectors. ### Claim 4 (depends on 2) The data source verifying apparatus according to claim 2, the memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to execute a similarity/distance check. ### Claim 5 (depends on 2) The data source verifying apparatus according to claim 2, the memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to detach the at least one data source from the data source verifying apparatus when the correlation measure is equal to or higher than the predetermined threshold. ### Claim 6 (depends on 2) The data source verifying apparatus according to claim 2, the memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to obtain the at least one feature vector as output of a Siamese network processing input source data. ### Claim 7 (depends on 6) The data source verifying apparatus according to claim 6, the memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to compress image data by filter and polling layers of the Siamese network for generation of the input source data. ### Claim 8 (depends on 2) The data source verifying apparatus according to claim 2, the memory containing instructions executable by the processing circuitry whereby the data source verifying apparatus is configured to obtain at least one feature vector which is encrypted and/or compressed. ### Claim 9 A feature vector crafting apparatus comprising processing circuitry and a memory containing instructions executable by the processing circuitry whereby the feature vector crafting apparatus is configured to: receive an identification of at least one trusted data source at the feature vector crafting apparatus; acquire at least one saliency map in relation to new specimen data generated at the at least one trusted data source, wherein the at least one saliency map indicates which part of the new specimen data is to be used for obtaining at least one new feature vector; and craft the at least one new feature vector from the part of new specimen data identified by the at least one saliency map. ### Claim 10 (depends on 9) The feature vector crafting apparatus according to claim 9, the memory containing instructions executable by the processing circuitry whereby the feature vector crafting apparatus is configured to acquire the at least one saliency map by querying the at least one trusted data source for transfer of the at least one saliency map in relation new specimen data. ### Claim 11 (depends on 9) The feature vector crafting apparatus according to claim 9, the memory containing instructions executable by the processing circuitry whereby the feature vector crafting apparatus is configured to acquire the at least one saliency map by local generation of the at least one saliency map from input data transferred by the at least one trusted data source to the feature vector crafting apparatus. ### Claim 12 (depends on 9) The feature vector crafting apparatus according to claim 9, the memory containing instructions executable by the processing circuitry whereby the feature vector crafting apparatus is configured to infer at least one feature area in relation to background for the new specimen data. ### Claim 13 (depends on 9) The feature vector crafting apparatus according to claim 9, the memory containing instructions executable by the processing circuitry whereby the feature vector crafting apparatus is configured to execute a correlation check between at least one feature area in the new specimen data and existing features maintained in a machine learning model and to identify a most uncorrelated feature area in the new specimen data with respect to existing features. ### Claim 14 (depends on 13) The feature vector crafting apparatus according to claim 13, wherein the machine learning model is a boosted machine learning model. ### Claim 15 A system for automatic generation of at least one feature vector for use in a machine learning infrastructure, the system comprising: a data source verifying apparatus for verifying at least one data source that generates new specimen data, wherein the data source verifying apparatus comprises processing circuitry configured to: obtain at least one feature vector representing the new specimen data; execute a correlation check between the at least one obtained feature vector and at least one reference feature vector representing trusted specimen data for determination of a correlation measure; and register the at least one data source as at least one trusted data source at a subscription management apparatus when the correlation measure is below a first predetermined threshold and the new specimen data is different from the trusted specimen data; and a feature vector crafting apparatus that comprises processing circuitry configured to: receive an identification of the at least one trusted data source; acquire at least one saliency map in relation to the new specimen data generated at the at least one trusted data source, wherein the at least one saliency map indicates which part of the new specimen data is to be used for obtaining at least one new feature vector; and craft the at least one new feature vector from the part of new specimen data identified by the at least one saliency map. ### Claim 16 (depends on 15) The system according to claim 15, wherein the data source verifying apparatus comprises processing circuitry configured to: receive a plurality of feature vectors representing the new specimen data as anchor data; evaluate whether the plurality of feature vectors have a continuous occurrence; and conclude on the new specimen data being related to a new object upon continuous occurrence of the plurality of feature vectors. ### Claim 17 (depends on 15) The system according to claim 15, wherein the data source verifying apparatus comprises processing circuitry configured to execute a similarity/distance check. ### Claim 18 (depends on 15) The system according to claim 15, wherein the data source verifying apparatus comprises processing circuitry configured to detach the at least one data source from the data source verifying apparatus when the correlation measure is equal to or higher than the predetermined threshold. ### Claim 19 (depends on 15) The system according to claim 15, wherein the data source verifying apparatus comprises processing circuitry configured to obtain the at least one feature vector as output of a Siamese network processing input source data. ### Claim 20 (depends on 19) The system according to claim 19, wherein the data source verifying apparatus comprises processing circuitry configured to compress image data by filter and polling layers of the Siamese network for generation of the input source data.