Product design generator
An embodiment establishes a product design database based at least in part on product design data received from a product image repository, wherein the product design data is representative of a plurality of product designs. The embodiment ranks each product design based on a ranking metric derived from market data corresponding to existing retail products. The embodiment constructs a product design ranking based on a relative ranking between each of the product designs that have been ranked. The embodiment generates a new product design using a machine learning model based on the plurality of product designs and the product design ranking. The embodiment evaluates whether the new product design satisfies a design constraint. The embodiment, upon an evaluation that the new product design satisfies the design constraint, stores the new product design in the product design database. The embodiment displays the new product design via an interface.
1 . A computer-implemented method comprising:
establishing a product design database based at least in part on product design data received from a product image repository, wherein the product design data is representative of a plurality of product designs;
transforming each product design into a vector embedding representing the product design, the transforming producing a set of vector embeddings;
ranking each product design of the plurality of product designs stored on the product design database based on a ranking metric derived from market data corresponding to existing retail products;
constructing a product design ranking based on a relative ranking between each of the product designs that have been ranked;
generating a new product design using a machine learning model based on the plurality of product designs and the product design ranking;
adjusting an influence radius for a class of images generated by the machine learning model to control a frequency of the class of images generated, the adjusting the influence radius causing the machine learning model to generate new product design images according to the influence radius of the class of images;
updating, upon an evaluation that the new product design satisfies a design constraint, the set of vector embeddings, wherein the updating is performed iteratively, wherein each iteration comprises updating the set of vector embeddings based on design features corresponding to the new product design;
displaying one or more new product designs of the set of new product designs via an interface; and
actuating, upon an evaluation that the set of new product designs satisfies an acceptance threshold, a manufacturing system to manufacture at least one of a product or product part according to the one or more new product designs.
2 . The computer-implemented method of claim 1 , further comprising training the machine learning model by iteratively evaluating whether the new product design satisfies the design constraint, wherein a new product design that satisfies the design constraint during a first iteration is used as input for a subsequent iteration, and wherein training the machine learning model ceases upon meeting an exit criterion.
3 . The computer-implemented method of claim 2 , wherein meeting the exit criterion comprises meeting a predetermined acceptable design threshold.
4 . The computer-implemented method of claim 2 , wherein training the machine learning model is performed incrementally.
5 . The computer-implemented method of claim 2 , wherein training the machine learning model comprises a reinforcement learning with human feedback (RLHF) technique.
6 . The computer-implemented method of claim 2 , wherein training the machine learning model further comprises applying a biased entropy loss function to bias the product design ranking.
7 . The computer-implemented method of claim 1 , wherein the market metric is based in part on market demand.
8 . The computer-implemented method of claim 1 , wherein the market metric is based in part on market sentiment.
9 . The computer-implemented method of claim 1 , wherein the market data comprises at least one of browsing data, social network data, and point-of-sale (POS) data.
10 . The computer-implemented method of claim 1 , wherein machine learning model comprises a generative adversarial neural network.
11 . A computer program product for new product design generation comprising one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by a processor to cause the processor to perform operations comprising:
establishing a product design database based at least in part on product design data received from a product image repository, wherein the product design data is representative of a plurality of product designs;
transforming each product design into a vector embedding representing the product design, the transforming producing a set of vector embeddings;
ranking each product design of the plurality of product designs stored on the product design database based on a ranking metric derived from market data corresponding to existing retail products;
constructing a product design ranking based on a relative ranking between each of the product designs that have been ranked;
generating a new product design using a machine learning model based on the plurality of product designs and the product design ranking;
adjusting an influence radius for a class of images generated by the machine learning model to control a frequency of the class of images generated, the adjusting the influence radius causing the machine learning model to generate new product design images according to the influence radius of the class of images;
updating, upon an evaluation that the new product design satisfies a design constraint, the set of vector embeddings, wherein the updating is performed iteratively, wherein each iteration comprises updating the set of vector embeddings based on design features corresponding to the new product design;
displaying one or more new product designs of the set of new product designs via an interface; and
actuating, upon an evaluation that the set of new product designs satisfies an acceptance threshold, a manufacturing system to manufacture at least one of a product or product part according to the one or more new product designs.
12 . The computer program product of claim 11 , wherein the stored program instructions are stored in a computer readable storage device in a data processing system, and wherein the stored program instructions are transferred over a network from a remote data processing system.
13 . The computer program product of claim 11 , wherein the stored program instructions are stored in a computer readable storage device in a server data processing system, and wherein the stored program instructions are downloaded in response to a request over a network to a remote data processing system for use in a computer readable storage device associated with the remote data processing system, further comprising:
program instructions to meter use of the program instructions associated with the request; and
program instructions to generate an invoice based on the metered use.
14 . The computer program product of claim 11 , further comprising training the machine learning model by iteratively evaluating whether the new product design satisfies the design constraint, wherein a new product design that satisfies the design constraint during a first iteration is used as input for a subsequent iteration, and wherein training the machine learning model ceases upon meeting an exit criterion.
15 . A computer system comprising a processor and one or more computer readable storage media, and program instructions collectively stored on the one or more computer readable storage media, the program instructions executable by the processor to cause the processor to perform operations comprising:
establishing a product design database based at least in part on product design data received from a product image repository, wherein the product design data is representative of a plurality of product designs;
transforming each product design into a vector embedding representing the product design, the transforming producing a set of vector embeddings;
ranking each product design of the plurality of product designs stored on the product design database based on a ranking metric derived from market data corresponding to existing retail products;
constructing a product design ranking based on a relative ranking between each of the product designs that have been ranked;
generating a new product design using a machine learning model based on the plurality of product designs and the product design ranking;
adjusting an influence radius for a class of images generated by the machine learning model to control a frequency of the class of images generated, the adjusting the influence radius causing the machine learning model to generate new product design images according to the influence radius of the class of images;
updating, upon an evaluation that the new product design satisfies a design constraint, the set of vector embeddings, wherein the updating is performed iteratively, wherein each iteration comprises updating the set of vector embeddings based on design features corresponding to the new product design;
displaying one or more new product designs of the set of new product designs via an interface; and
actuating, upon an evaluation that the set of new product designs satisfies an acceptance threshold, a manufacturing system to manufacture at least one of a product or product part according to the one or more new product designs.
16 . The computer system of claim 15 , further comprising training the machine learning model based in part by iteratively evaluating whether a generated new product design is acceptable, wherein an acceptable generated new product design during a first iteration is used as input for a subsequent iteration, and wherein training the machine learning model ceases upon meeting an exit criterion.
17 . The computer system of claim 16 , wherein training the machine learning model further comprises applying a biased entropy loss function to bias the product design ranking.