At IDVerse, generative AI (GenAI) is part of our native development stack—and something that’s been central to our core platforms since our company’s founding in 2014. We’ve pioneered the use of synthetic media to train and debias facial recognition systems.
In this blog post, we’d like to show you how (and why) we have used this transformative technology to build the most certified IDV solution on the market.
1) Create vast datasets, no PII
One of the major benefits of GenAI is its ability to create vast datasets for training and testing facial data without the need for personally identifiable information (PII). Our in-house synth generator enables us to generate millions of unique faces, essentially a large facial model (LFM), covering a wide range of environmental lighting conditions, face shapes, ages, head poses, and skin tones.
Using this method, we can guarantee the source, quality, and reliability of the training data used for our algorithms. Having this ability not only eliminates concerns about data protection but furthermore allows us to create targeted datasets to address specific challenges, such as heavy beards or low-quality vector images.
2) Reduce algorithmic bias
GenAI also plays a key role in helping us identify and mitigate algorithmic bias. Through the creation of tailored datasets that focus on specific age ranges, genders, or skin tones, we can rigorously test our algorithms for potential biases and immediately take corrective measures through targeted training. This practice ensures that our identity verification solutions provide fair and accurate results for all users, regardless of their demographic background.
With over 50 million faces generated and 1,000 points automatically labeled on each face, our neural networks can effectively characterize and learn from this extensive dataset.
Unlike traditional approaches that rely on extracting facial features, IDVerse’s system favors mathematical-based models that utilize more accurate and purpose-driven face extraction and Euclidean distance measurements. This approach reduces the risk of demographic bias and ensures that our algorithms perform consistently across diverse populations.
3) Combat the deepfake threat
GenAI also enables us to recreate known attack vectors and generate hundreds of small variations, such as differing poses, lighting conditions, ages, and genders. Training our algorithms on these diverse datasets lets us enhance their ability to detect and defend against sophisticated fraud attempts. In taking this proactive approach, we can stay one step ahead of fraudsters and maintain the very highest levels of security for our clients.
As always, we are looking to the future, and we anticipate that bad actors will increasingly adopt GenAI techniques to create deepfakes and synthetic identities in an attempt to bypass identity verification systems.
Fortunately, IDVerse is well-prepared to combat this threat. Our proprietary Deepfake Defender technology, trained on a large dataset of AI-generated synthetic faces, excels at identifying and blocking bad actors who attempt to gain access using deepfakes.
4) Improve capabilities in real time
We also use GenAI and deep neural networks to continuously enhance our identity verification capabilities in real-time, allowing our solution to analyze vast amounts of data and adapt to new fraud patterns and techniques as they emerge.
One key aspect of IDVerse’s approach is the use of generative adversarial networks (GANs) to create synthetic documents for training purposes. These synthetic documents, which closely mimic real ID docs, are used to train the system’s neural networks to detect even the most sophisticated forgeries. As the GANs generate increasingly complex and realistic synthetic documents, the neural networks become more and more adept at flagging fraudulent ones.
The neural network technology we employ is specifically designed to “learn” and improve with each verification process. Every time a document is analyzed, the system gains valuable insights that are applied to refine its algorithms and enhance its accuracy. This continuous improvement process ensures that IDVerse stays well ahead of the curve when it comes to detecting identity fraud and providing our clients with the highest possible level of security.
The future of IDV with GenAI
The potential applications of GenAI in identity verification are vast and exciting. In parallel with the technology’s evolution, we expect to see ever more innovative solutions that enhance security, reduce fraud, and provide seamless user experiences. IDVerse is committed to exploring these possibilities and collaborating with industry partners to shape the future of the industry.
Our strategic utilization of generative AI sets us apart as a leader in the IDV space. Using GenAI, we are able to create robust, bias-free algorithms that provide industry-leading accuracy and security. IDVerse will continue to push the boundaries of what is possible with GenAI, and we will remain dedicated to our mission of providing secure, reliable, and fair identity verification solutions that protect businesses, individuals, and—without exaggeration—society at large.
About the post:
Images are generative AI-created. Prompt: A captivating image of an amorphous cloud composed entirely of diverse smiling human faces, representing the rich tapestry of humanity. The faces vary in age, gender, ethnicity, and emotion, interconnected in a fluid and dynamic structure. The semi-transparent faces blend and overlap, creating unity and interconnectedness against a soft, neutral background. Tool: Midjourney.
About the author:
Matthew Horgan, now the PM of Core Technologies at IDVerse, recently moved from overseeing the IDVerse Enterprise product. A self-proclaimed science nerd, he loves working with new technologies that power amazing products.
