Generally, the present disclosure is directed to training machine learning models, e.g., deep learning models, such that the impact of any implicit bias in the training dataset on the trained model is eliminated or minimized. In particular, in some implementations, the systems and methods of the present disclosure can include or otherwise leverage a generative adversarial configuration with a generator model and a discriminator model such that the resultant trained model (generator) performs its function free of any implicit bias that may be present in the training dataset. The model as trained herein can be any type of machine learning model, e.g., a neural network or other type of model, and can be trained for any suitable purpose.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
N/A, "Removing Bias from Deep Learning Systems", Technical Disclosure Commons, (July 29, 2021)