Menu Close

AI put in charge of creating fairness training of AI

Researchers say they have developed a way to create data “pipelines” to supply training models with high-quality software-fabricated faces. Their results, however, show that, at least in this case, artificial intelligence does not perform better in training AI than debiasing models.

The faces can be used to automatically generate and supplement training datasets, according to a recent paper. In either case, the generated faces can address attributes underrepresented in a database without a human designing the faces.

The work is designed to make AI more accurate at identifying non-white, middle-aged males. It was never the case that people of color and females routinely were represented in the proportion they occupy in reality, making AI training incomplete.

And, as biometric privacy rights become more of a hurdle, it has only gotten more difficult to find large, representative datasets.

The paper, written by two University of Southern California scientists, has not yet been peer-reviewed. Their research was partly funded by the Intelligence Advanced Research Projects Activity unit within the Office of the Director of National Intelligence.

The researchers say their technique allows face-generating algorithms to create larger or smaller attribute classes, making more “chubby” (their description) females with baggy eyes, wavy blond bangs and heavy makeup, for example.

More work is needed, however.

According to the report, “although the network trained with the proposed synthetic data does not outperform the debiasing models in recognition performance, it is impressive that synthetic images can achieve consistent performance with real data and further yields better fairness.” Researchers say they have developed a way to create data “pipelines” to supply training models with high-quality software-fabricated faces. Their results, however, show that, at least in this case, artificial intelligence does not perform better in training AI than debiasing models.

The faces can be used to automatically generate and supplement training datasets, according to a recent paper. In either case, the generated faces can address attributes underrepresented in a database without a human designing the faces.

The work is designed to make AI more accurate at identifying non-white, middle-aged males. It was never the case that people of color and females routinely were represented in the proportion they occupy in reality, making AI training incomplete.

And, as biometric privacy rights become more of a hurdle, it has only gotten more difficult to find large, representative datasets.

The paper, written by two University of Southern California scientists, has not yet been peer-reviewed. Their research was partly funded by the Intelligence Advanced Research Projects Activity unit within the Office of the Director of National Intelligence.

The researchers say their technique allows face-generating algorithms to create larger or smaller attribute classes, making more “chubby” (their description) females with baggy eyes, wavy blond bangs and heavy makeup, for example.

More work is needed, however.

According to the report, “although the network trained with the proposed synthetic data does not outperform the debiasing models in recognition performance, it is impressive that synthetic images can achieve consistent performance with real data and further yields better fairness.”  Read More   

Generated by Feedzy

Disclaimer

Innov8 is owned and operated by Rolling Rock Ventures. The information on this website is for general information purposes only. Any information obtained from this website should be reviewed with appropriate parties if there is any concern about the details reported herein. Innov8 is not responsible for its contents, accuracies, and any inaccuracies. Nothing on this site should be construed as professional advice for any individual or situation. This website includes information and content from external sites that is attributed accordingly and is not the intellectual property of Innov8. All feeds ("RSS Feed") and/or their contents contain material which is derived in whole or in part from material supplied by third parties and is protected by national and international copyright and trademark laws. The Site processes all information automatically using automated software without any human intervention or screening. Therefore, the Site is not responsible for any (part) of this content. The copyright of the feeds', including pictures and graphics, and its content belongs to its author or publisher.  Views and statements expressed in the content do not necessarily reflect those of Innov8 or its staff. Care and due diligence has been taken to maintain the accuracy of the information provided on this website. However, neither Innov8 nor the owners, attorneys, management, editorial team or any writers or employees are responsible for its content, errors or any consequences arising from use of the information provided on this website. The Site may modify, suspend, or discontinue any aspect of the RSS Feed at any time, including, without limitation, the availability of any Site content.  The User agrees that all RSS Feeds and news articles are for personal use only and that the User may not resell, lease, license, assign, redistribute or otherwise transfer any portion of the RSS Feed without attribution to the Site and to its originating author. The Site does not represent or warrant that every action taken with regard to your account and related activities in connection with the RSS Feed, including, without limitation, the Site Content, will be lawful in any particular jurisdiction. It is incumbent upon the user to know the laws that pertain to you in your jurisdiction and act lawfully at all times when using the RSS Feed, including, without limitation, the Site Content.  

Close Bitnami banner
Bitnami