Basically, we have been working on a few European projects and during these projects we have tested federated learning. Basically, this technology allows us to train artificial intelligence models for diagnosis and prognosis on our datasets without data sharing. And what happens is that we share the model at the end of the training and this allows us to create a more powerful centralized model between different centers without data sharing...
Basically, we have been working on a few European projects and during these projects we have tested federated learning. Basically, this technology allows us to train artificial intelligence models for diagnosis and prognosis on our datasets without data sharing. And what happens is that we share the model at the end of the training and this allows us to create a more powerful centralized model between different centers without data sharing. And this is awesome because of course it doesn’t break any GDPR violation or anything like that. And at a certain point, we wonder what happens if we apply this to generative AI. Usually with these rare hematological diseases, if you want to train a model to generate synthetic patients, you need a lot of patients. So this technology fits very well with that purpose. Basically we trained three different models, a VAE, a Bayesian network, and also a GAN model. And basically the performance shows that the model created in this federated way creates better patients than the ones that we would have been able to create with the data in each singular institution and also we use the framework for the validation of these synthetic data, we validated that the data were clinically correct but also that the privacy of the patient was preserved and it wasn’t just a copy-paste, you know, otherwise that would have been of course a problem. And so, yeah, this is the whole concept and it was really interesting. We’re still working on these European grants and we plan to have more results by the end of these grants, but so far these results are really promising.
This transcript is AI-generated. While we strive for accuracy, please verify this copy with the video.
