# US 12,217,189 B2 · Hyperparameter adjustment device, non-transitory recording medium in which hyperparameter adjustment program is recorded, and hyperparameter adjustment program > Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt - Number: US 12,217,189 B2 - Document ID: US12217189B2 - Status: Granted - Kind: B2 - Application number: 18/008,500 - Filed: 2020-09-10 - Published: 2023-07-06 - Granted: 2025-02-04 - Earliest claimed: 2020-09-10 - Assignees: Aizoth Inc. - Applicants: Aizoth Inc. - Inventors: Kotaro Kawajiri - Web page: https://patentgrep.com/patent/US12217189B2 ## CPC - G06N 3/0985: Hyperparameter optimisation; Meta-learning; Learning-to-learn - G06N 3/045: Combinations of networks - G06N 3/0499: Feedforward networks - G06N 3/084: Backpropagation, e.g. using gradient descent - G06N 3/086: using evolutionary algorithms, e.g. genetic algorithms or genetic programming - G06N 3/09: Supervised learning - G06N 3/126: Evolutionary algorithms, e.g. genetic algorithms or genetic programming - G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS ## Abstract A learning processing unit that causes a second neural network (NN) to be trained, with a hyperparameter set of a first NN accepted as input, so as to output post-learning performance that is the performance of a trained first NN to which the hyperparameter set is set. A GA processing unit adjusts the hyperparameter set of the first NN by a genetic algorithm, with the hyperparameter set of the first NN handled as entity, the fitness of said algorithm being configured to be a value that corresponds to the post-learning performance of the first NN to which the hyperparameter set is set. In processing in each generation of the genetic algorithm, the post-learning performance of the first NN corresponding to each hyperparameter is acquired using the second NN. ## Claims ### Claim 1 A hyperparameter tuning device including a processor, the hyperparameter tuning device comprising: a learning processing unit operated by the processor and configured to train a learner, using a hyperparameter set of a first neural network as input of the learner, to cause the learner to output post-learning performance, the post-learning performance being performance of the first neural network in which the hyperparameter set is set responsive to the first neural network having been trained; and a hyperparameter tuning unit operated by the processor and configured to tune the hyperparameter set of the first neural network by repeating a process, without training a plurality of the first neural networks in each of which a corresponding one of a plurality of hyperparameter sets is set, of generating a plurality of new hyperparameter sets on the basis of the plurality of hyperparameter sets, and inputting each of the plurality of new hyperparameter sets that have been generated into the learner that has been trained, to acquire the post-leaning performance of the plurality of the first neural networks in each of which a corresponding one of the plurality of new hyperparameter sets is set, wherein the plurality of new hyperparameter sets are generated based on the post-learning performance of a plurality of the first neural networks in each of which a corresponding one of a plurality of hyperparameter sets of the first neural network is set, the post-learning performance of the plurality of the first neural networks being acquired by inputting each of the plurality of hyperparameter sets to the learner that has been trained. ### Claim 2 (depends on 1) The hyperparameter tuning device according to claim 1, wherein the hyperparameter tuning unit is configured to tune the hyperparameter set of the first neural network by using a heuristic search technique, the heuristic search technique using the post-learning performance of a plurality of the first neural networks in each of which a corresponding one of a plurality of hyperparameter sets of the first neural network is set, the post-learning performance of the plurality of the first neural networks being acquired by inputting each of the plurality of hyperparameter sets to the learner that has been trained. ### Claim 3 (depends on 2) The hyperparameter tuning device according to claim 2, wherein the hyperparameter tuning unit is a genetic algorithm processing unit that is configured to tune the hyperparameter set of the first neural network by using a genetic algorithm, the genetic algorithm using, as an individual, the hyperparameter set of the first neural network and using, as fitness, a value corresponding to the post-learning performance of the first neural network in which the hyperparameter set is set and which has been trained, the post-learning performance of the first neural network being acquired by inputting the hyperparameter set to the learner that has been trained, wherein the genetic algorithm repeats a process of generating a plurality of next-generation individuals based on the fitness of each of a plurality of the individuals. ### Claim 4 (depends on 1) The hyperparameter tuning device according to claim 1, further comprising a learner parameter determination unit operated by the processor and configured to determine a hyperparameter of the learner, in accordance with analysis target data which is input to the first neural network. ### Claim 5 (depends on 1) The hyperparameter tuning device according claim 1, wherein the learner includes a second neural network that is different from the first neural network. ### Claim 6 A non-transitory computer-readable recording medium in which a hyperparameter tuning program is recorded, the hyperparameter tuning program causing a computer to function as: a learning processing unit operated by a processor of the computer and configured to train a learner, using a hyperparameter set of a first neural network as input of the learner, to cause the learner to output post-learning performance, the post-learning performance being performance of the first neural network in which the hyperparameter set is set responsive to the first neural network having been trained; and a hyperparameter tuning unit operated by the processor and configured to tune the hyperparameter set of the first neural network by repeating a process, without training a plurality of the first neural networks in each of which a corresponding one of a plurality of hyperparameter sets is set, of generating a plurality of new hyperparameter sets on the basis of the plurality of hyperparameter sets, and inputting each of the plurality of new hyperparameter sets that have been generated into the learner that has been trained, to acquire the post-leaning performance of the plurality of the first neural networks in each of which a corresponding one of the plurality of new hyperparameter sets is set, wherein the plurality of new hyperparameter sets are generated based on the post-learning performance of a plurality of the first neural networks in each of which a corresponding one of a plurality of hyperparameter sets of the first neural network is set, the post-learning performance of the plurality of the first neural networks being acquired by inputting each of the plurality of hyperparameter sets to the learner that has been trained.