System and methods to generate synthetic data for network using generative adversarial network
A generator model initially receives seed data for a telecommunications network. Using the seed data, the generator model generates, in one or more simulations cycles, synthetic data. A discriminator model, trained to classify the synthetic data, communicates the classified data to the generator to update the generator model. An evaluator evaluates the synthetic data that has been combined with the seed data using evaluation metrics. In various aspects, the generator model: extracts semantic information from text embeddings of the seed data; extracts feature data from the semantic information; determines a relationship between the feature data and the semantic information; and disentangles the feature data to generate the synthetic data. In some aspects, the generator model adds noise at the generator to vary the feature data prior to each simulation cycle of the one or more simulation cycles.
1 . A system for generating synthetic data in a telecommunications network, the system comprising:
a generator model implemented using a generative adversarial network (GAN), the generator model being executed by one or more processors and configured to receive seed data for a telecommunications network and to generate, in one or more simulations cycles, synthetic data;
a discriminator neural network executed by the one or more processors and trained to classify the synthetic data and to communicate classification results to the generator model to update the generator model during the one or more simulation cycles; and
an evaluator executed by the one or more processors and configured to evaluate combined data comprising the seed data and the synthetic data using one or more evaluation metrics.
2 . The system of claim 1 , wherein the one or more evaluation metrics comprise one or more of: structural and statistical similarity, model capability, and privacy risk.
3 . The system of claim 1 , further comprising training the generator model to extract semantic information from text embeddings of the seed data.
4 . The system of claim 3 , further comprising training the generator model to extract feature data from the semantic information.
5 . The system of claim 4 , further comprising, determining a relationship, at the generator model, between the feature data and the semantic information.
6 . The system of claim 5 , further comprising disentangling, at the generator model, the feature data to generate the synthetic data.
7 . The system of claim 6 , further comprising, adding noise at the generator model to vary the feature data prior to each simulation cycle of the one or more simulation cycles.
8 . The system of claim 1 , utilizing the classified data to update the generator model.
9 . A method for generating synthetic data in a telecommunications network, the method comprising:
receiving, at a generator model implemented using a generative adversarial network (GAN) and executed by one or more processors, seed data for a telecommunications network and generating, in one or more simulation cycles, synthetic data;
classifying, at a discriminator neural network executed by the one or more processors and trained to classify the synthetic data, and communicating classification results to the generator model to update the generator model during the one or more simulation cycles; and
evaluating, at an evaluator executed by the one or more processors, combined data comprising the seed data and the synthetic data using one or more evaluation metrics.
10 . The method of claim 9 , wherein the one or more evaluation metrics comprise one or more of: structural and statistical similarity, model capability, and privacy risk.
11 . The method of claim 9 , further comprising training the generator model to extract semantic information from text embeddings of the seed data.
12 . The method of claim 11 , further comprising training the generator model to extract feature data from the semantic information.
13 . The method of claim 12 , further comprising, determining a relationship, at the generator model, between the feature data and the semantic information.
14 . The method of claim 13 , further comprising disentangling, at the generator model, the feature data to generate the synthetic data.
15 . The method of claim 14 , further comprising, adding noise at the generator model to vary the feature data prior to each simulation cycle of the one or more simulation cycles.
16 . The method of claim 9 , utilizing the classified data to update the generator model.
17 . A non-transitory computer readable media having computer executable instructions stored thereon that, when executed by one or more computer processing components, cause the one or more computer processing components to perform operations for generating synthetic data in a telecommunications network:
receiving, at a generator model implemented using a generative adversarial network (GAN), seed data for a telecommunications network training the generator model to:
extract semantic information from text embeddings of the seed data;
extract feature data from the semantic information;
determine a relationship between the feature data and the semantic information; and
disentangle the feature data to generate, in one or more simulation cycles, synthetic data;
classifying, at a discriminator neural network trained to classify the synthetic data and communicating classification results to the generator model to update the generator model; and
evaluating, at an evaluator, combined data comprising the seed data and the synthetic data using one or more evaluation metrics.
18 . The media of claim 17 , wherein the one or more evaluation metrics comprise one or more of: structural and statistical similarity, model capability, and privacy risk.
19 . The media of claim 17 , further comprising, adding noise at the generator model to vary the feature data prior to each simulation cycle of the one or more simulation cycles.
20 . The media of claim 17 , further comprising utilizing the classified data to update the generator model.