# US 2020/0401844 A1 · Adversarial multi-binary neural network for multi-class classification > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 2020/0401844 A1 - Document ID: US20200401844A1 - Status: Published application - Kind: A1 - Application number: 17/014,256 - Filed: 2020-09-08 - Published: 2020-12-24 - Earliest claimed: 2019-05-15 - Assignees: Beijing Didi Infinity Technology and Development Co., Ltd. - Applicants: Beijing Didi Infinity Technology and Development Co., Ltd. - Inventors: Kun Han; Haiyang Xu - Web page: https://patentgrep.com/patent/US20200401844A1 ## CPC - G06F 40/216: using statistical methods - G06F 16/355: Creation or modification of classes or clusters - G06F 18/211: Selection of the most significant subset of features - G06F 18/2132: based on discrimination criteria, e.g. discriminant analysis - G06F 18/2431: Multiple classes - G06F 40/284: Lexical analysis, e.g. tokenisation or collocates - G06F 40/30: Semantic analysis - G06N 3/044: Recurrent networks, e.g. Hopfield networks - G06N 3/0442: characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] - G06N 3/08: Learning methods - G06N 3/09: Supervised learning - G06N 3/094: Adversarial learning - G06N 3/045: Combinations of networks - G06N 20/20: Ensemble learning - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract Embodiments of the disclosure provide a multi-class classification system. An exemplary system includes at least one processor and at least one non-transitory memory storing instructions that, when executed by the at least one processor, cause the system to perform operations. The operation includes applying a multi-class classifier to classify a set of objects into multiple classes and applying a plurality of binary classifiers to the set of objects, wherein the plurality of binary classifiers are decomposed from the multi-class classifier, each binary classifier classifying the set of the objects into a first group consisting of one or more classes selected from the multiple classes and a second group consisting of one or more remaining classes of the multiple classes. The operation also includes jointly classifying the set of objects using the multi-class classifier and the plurality of binary classifiers. ## Claims ### Claim 1 A multi-class classification system, comprising: at least one processor; and at least one non-transitory memory storing instructions that, when executed by the at least one processor, cause the system to perform operations comprising: applying a multi-class classifier to classify a set of objects into multiple classes; applying a plurality of binary classifiers to the set of objects, wherein the plurality of binary classifiers are decomposed from the multi-class classifier, each binary classifier classifying the set of the objects into a first group consisting of one or more classes selected from the multiple classes and a second group consisting of one or more remaining classes of the multiple classes; and jointly classifying the set of objects using the multi-class classifier and the plurality of binary classifiers. ### Claim 2 (depends on 1) The system of claim 1, wherein the operations comprise: jointly training the multi-class classifier and the plurality of binary classifiers by minimizing a joint loss comprising a multi-class classification loss corresponding to the multi-class classifier and one or more binary classification losses corresponding to one or more binary classifiers selected from the plurality of binary classifiers. ### Claim 3 (depends on 1) The system of claim 1, wherein the operations comprise: receiving, by an encoder corresponding to at least one binary classifier, the set of objects; and generating, by the encoder, contextual information from the set of the objects. ### Claim 4 (depends on 3) The system of claim 3, wherein the encoder comprises a bidirectional long short memory (BiLSTM). ### Claim 5 (depends on 3) The system of claim 3, wherein the operations further comprise: determining, by a private attention layer corresponding to the at least one binary classifier, class-specific information based on the contextual information; and determining, by a shared attention layer, class-agnostic information based on the contextual information. ### Claim 6 (depends on 5) The system of claim 5, wherein the class-agnostic information comprises a shared feature shared by multiple classes. ### Claim 7 (depends on 5) The system of claim 5, wherein the operations comprise: refining, by a discriminator, the class-agnostic information by minimizing an adversarial loss. ### Claim 8 (depends on 7) The system of claim 7, wherein the operations comprise: jointly training the multi-class classifier and the plurality of binary classifiers by minimizing a joint loss comprising: a multi-class classification loss corresponding to the multi-class classifier; one or more binary classification losses corresponding to one or more binary classifiers selected from the plurality of binary classifiers; and the adversarial loss. ### Claim 9 (depends on 5) The system of claim 5, wherein the operations comprise: generating classification features of the at least one binary classifier by concatenating the class-specific information and the class-agnostic information. ### Claim 10 (depends on 5) The system of claim 5, wherein the operations comprise: generating classification features of the multi-class classifier by concatenating class-specific information corresponding to multiple binary classifiers and the class-agnostic information. ### Claim 11 A multi-class classification method, comprising: applying a multi-class classifier to classify a set of objects into multiple classes; applying a plurality of binary classifiers to the set of objects, wherein the plurality of binary classifiers are decomposed from the multi-class classifier, each binary classifier classifying the set of the objects into a first group consisting of one or more classes selected from the multiple classes and a second group consisting of one or more remaining classes of the multiple classes; and jointly classifying the set of objects using the multi-class classifier and the plurality of binary classifiers. ### Claim 12 (depends on 11) The method of claim 11, comprising: jointly training the multi-class classifier and the plurality of binary classifiers by minimizing a joint loss comprising a multi-class classification loss corresponding to the multi-class classifier and one or more binary classification losses corresponding to one or more binary classifiers selected from the plurality of binary classifiers. ### Claim 13 (depends on 11) The method of claim 11, comprising: receiving, by an encoder corresponding to at least one binary classifier, the set of objects; and generating, by the encoder, contextual information from the set of the objects. ### Claim 14 (depends on 13) The method of claim 13, further comprising: determining, by a private attention layer corresponding to the at least one binary classifier, class-specific information based on the contextual information; and determining, by a shared attention layer, class-agnostic information based on the contextual information. ### Claim 15 (depends on 14) The method of claim 14, wherein the class-agnostic information comprises a shared feature shared by multiple classes. ### Claim 16 (depends on 14) The method of claim 14, further comprising: refining, by a discriminator, the class-agnostic information by minimizing an adversarial loss. ### Claim 17 (depends on 16) The method of claim 16, further comprising: jointly training the multi-class classifier and the plurality of binary classifiers by minimizing a joint loss comprising: a multi-class classification loss corresponding to the multi-class classifier; one or more binary classification losses corresponding to one or more binary classifiers selected from the plurality of binary classifiers; and the adversarial loss. ### Claim 18 (depends on 14) The method of claim 14, further comprising: generating classification features of the at least one binary classifier by concatenating the class-specific information and the class-agnostic information. ### Claim 19 (depends on 14) The method of claim 14, further comprising: generating classification features of the multi-class classifier by concatenating class-specific information corresponding to multiple binary classifiers and the class-agnostic information. ### Claim 20 A non-transitory computer-readable medium having instructions stored thereon that, when executed by at least one processor, causes the processor to perform a method for classifying a set of objects, the method comprising: applying a multi-class classifier to classify the set of objects into multiple classes; applying a plurality of binary classifiers to the set of objects, wherein the plurality of binary classifiers are decomposed from the multi-class classifier, each binary classifier classifying the set of the objects into a first group consisting of one or more classes selected from the multiple classes and a second group consisting of one or more remaining classes of the multiple classes; and jointly classifying the set of objects using the multi-class classifier and the plurality of binary classifiers.