# US 11,113,576 B2 · Information processing apparatus for training neural network for recognition task and method thereof > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 11,113,576 B2 - Document ID: US11113576B2 - Status: Granted - Kind: B2 - Application number: 16/389,546 - Filed: 2019-04-19 - Published: 2019-10-31 - Granted: 2021-09-07 - Earliest claimed: 2018-04-26 - Assignees: Canon Kabushiki Kaisha - Applicants: Canon Kabushiki Kaisha - Inventors: Takayuki Saruta - Web page: https://patentgrep.com/patent/US11113576B2 ## CPC - G06N 20/10: using kernel methods, e.g. support vector machines [SVM] - G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting - G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques} - G06N 3/045: Combinations of networks - G06N 3/0464: Convolutional networks [CNN, ConvNet] - G06N 3/048: Activation functions - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N 3/09: Supervised learning - G06V 10/764: using classification, e.g. of video objects - G06V 10/82: using neural networks - G06V 20/10: Terrestrial scenes (scenes under surveillance with static cameras G06V20/52; scenes perceived from the exterior of a vehicle G06V20/56; scenes perceived from the interior of a vehicle G06V20/59) - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS - G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N) - G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING ## Abstract An apparatus includes a multitask setting unit configured to set a plurality of recognition tasks for which a multilayer neural network is trained, a neural network (NN) training unit configured to train the multilayer NN for the set plurality of recognition tasks, an NN evaluation unit configured to evaluate a training result of the multilayer NN, and a training parameter setting unit configured to set a training parameter in training the multilayer NN for the plurality of recognition tasks, based on a result of evaluation by the NN evaluation unit. ## Claims ### Claim 1 An apparatus comprising: a processor; and a memory coupled to the processor storing instructions that, when executed by the processor, cause the processor to function as: a task setting unit configured to set a plurality of recognition tasks for which a neural network or a classifier is trained, wherein the plurality of recognition tasks is to recognize targets different from each other; a training unit configured to train the neural network or the classifier for the plurality of tasks based on training data and a teaching value for the plurality of recognition tasks; an evaluation unit configured to evaluate a training result of the neural network or of the classifier; and a parameter setting unit configured to set a training parameter including weight assigned to errors in the plurality of recognition tasks in training the neural network or the classifier for the plurality of recognition tasks, based on a result of evaluation by the evaluation unit, wherein the training unit performs second training on the neural network or the classifier according to the training parameter set by the parameter setting unit based on the result of evaluation of first training by the evaluation unit. ### Claim 2 (depends on 1) The apparatus according to claim 1, wherein the evaluation unit is further configured to evaluate training accuracy of the neural network or the classifier in each of the plurality of recognition tasks. ### Claim 3 (depends on 2) The apparatus according to claim 2, wherein the evaluation unit is further configured to evaluate an amount of displacement in the training accuracy of the neural network or the classifier in each of the plurality of recognition tasks at a predetermined interval. ### Claim 4 (depends on 1) The apparatus according to claim 1, wherein the training parameter includes a degree of significance to each of the plurality of recognition tasks in training the neural network or the classifier. ### Claim 5 (depends on 1) The apparatus according to claim 1, wherein the task setting unit is configured to set, as the plurality of recognition tasks, a first recognition task for which the neural network or the classifier is trained and a second recognition task generated from the first recognition task. ### Claim 6 (depends on 5) The apparatus according to claim 5, wherein the instructions further cause the processor to function as a teaching value setting unit configured to set a teaching value for the second recognition task, based on training data and a teaching value for the first recognition task. ### Claim 7 (depends on 1) The apparatus according to claim 1, wherein the evaluation unit is configured to evaluate the training result of the neural network or the classifier by using the training data for the plurality of recognition tasks. ### Claim 8 (depends on 1) The apparatus according to claim 1, wherein the evaluation unit is configured to evaluate the training result of the neural network or the classifier by using evaluation data different from the training data for the plurality of recognition tasks. ### Claim 9 (depends on 1) The apparatus according to claim 1, wherein the plurality of recognition tasks set by the task setting unit includes a classification task for identifying whether an object is included in an image. ### Claim 10 (depends on 1) The apparatus according to claim 1, wherein the plurality of recognition tasks include a detection task for detecting a position of an object. ### Claim 11 (depends on 1) The apparatus according to claim 1, wherein the plurality of recognition tasks include a segmentation task for extracting an object region. ### Claim 12 (depends on 1) The apparatus according to claim 1, wherein the parameter setting unit is further configured to select one of the plurality of recognition tasks for which the neural network or the classifier is trained, and set the training parameter based on training accuracy for the selected recognition task. ### Claim 13 (depends on 1) The apparatus according to claim 1, wherein the instructions further cause the processor to function as a task display unit configured to display the plurality of recognition tasks. ### Claim 14 (depends on 1) The apparatus according to claim 1, wherein the instructions further cause the processor to function as a result display unit configured to display the evaluated training result. ### Claim 15 (depends on 1) An information processing apparatus comprising: a processor; and a memory coupled to the processor storing instructions that, when executed by the processor, cause the processor to function as: a storage unit configured to store information about a structure and a training parameter of a neural network or a classifier trained by the apparatus according to claim 1; an input unit configured to input an identification target image into the neural network or the classifier based on the stored information; and an output unit configured to output an identification result of the identification target image by the neural network or the classifier. ### Claim 16 A method comprising: setting a plurality of recognition tasks for which a neural network or a classifier is trained, wherein the plurality of recognition tasks is to recognize targets different from each other; training the neural network or the classifier for the plurality of recognition tasks based on training data and a teaching value for the plurality of recognition tasks; evaluating a training result of the neural network or of the classifier; setting a training parameter including weight assigned to errors in the plurality of recognition tasks in training the neural network or the classifier for the plurality of recognition tasks, based on a result of the evaluating; and performing second training on the neural network or the classifier according to the training parameter set by the setting based on the result of evaluation of first training by the evaluating. ### Claim 17 A non-transitory computer-readable storage medium storing a program for causing a computer to perform: setting a plurality of recognition tasks for which a neural network or a classifier is trained, wherein the plurality of recognition tasks is to recognize targets different from each other; training the neural network or the classifier for the plurality of recognition tasks based on training data and a teaching value for the plurality of recognition tasks; evaluating a training result of the neural network or of the classifier; and setting a training parameter including weight assigned to errors in the plurality of recognition tasks in training the neural network or the classifier for the plurality of recognition tasks, based on a result of the evaluating; and performing second training on the neural network or the classifier according to the training parameter set by the setting based on the result of evaluation of first training by the evaluating. ### Claim 18 (depends on 15) The information processing apparatus according to claim 15, wherein the plurality of recognition tasks set by the task setting unit includes a classification task for identifying whether an object is included in an image. ### Claim 19 (depends on 16) The method according to claim 16, wherein the plurality of recognition tasks set by the task setting includes a classification task for identifying whether an object is included in an image. ### Claim 20 (depends on 17) The non-transitory computer-readable storage medium according to claim 17, wherein the plurality of recognition tasks set by the task setting includes a classification task for identifying whether an object is included in an image.