Ovation Fertility and AI company Presagen have collaborated to develop a novel decentralized artificial intelligence (AI) learning model that enhances AI training while adhering to data privacy laws. When training AI, it’s critical to use multiple, diverse data sources. This results in unbiased and generalized AI. However, privacy laws prevent researchers from moving data outside of the country of origin. To combat this issue, Presagen developed the Decentralized AI Training Algorithm (DAITA).
The new algorithm uses a decentralized federated learning approach and knowledge distillation, allowing each node to operate independently without needing to access external data. This allows researchers to comply with data privacy laws while conducting effective AI training. It also has the added benefits of being scalable and cost-effective. The algorithm works by moving the AI to the location of the data rather than moving the data to the AI in a central location. When distributing any information globally using the model, only general abstract learning is shared, not the individual datasets.
According to Presagen Chief Scientist Dr. Jonathan Hall, “Using DAITA, we can optimize how the AI travels around the world. Doing so, not only minimizes the cost of transfer, but also maximizes the performance of the final AI while adhering to data privacy laws.”
To test this algorithm in a real-world setting, Presagen turned to Ovation Fertility. In doing so, Ovation and Presagen collaborated to assess the viability of embryos with the goal of helping embryologists identify which embryos were the most likely to result in pregnancy. The results of this study showed that AI training using this approach produces comparable results to centralized training. Further, when nodes included poor-quality data, which is common, this algorithm exceeded the performance of traditional centralized training.
Ovation Fertility Vice President of Scientific Advancement, Dr. Matthew (Tex) VerMilyea highlighted the importance of these findings. “At Ovation, our goal is to develop new ways to assess the viability of embryos and help patients have IVF cycles that result in successful pregnancies. AI is one of the tools that we are using to help make this possible, and DAITA highlights a way to better train AI to identify the best embryos for transfer, while adhering to data privacy laws.”
For more on this, head on over to Nature
