The pursuit of open science is supported by data sharing, which fosters a culture of transparency, enables data-driven discovery, and promotes knowledge dissemination. However, concerns about privacy breaches and institutional data sharing policies can create barriers to sharing sensitive human subject and patient data, thereby limiting the potential of open science initiatives. In situations where sharing real data is not possible, generating and sharing synthetic data can provide a useful workaround. In this blog, we provide an implementation of Variational Autoencoders (VAE) Generative Models leveraging encoded input data within a Deep Latent Space to generate synthetic Functional Magnetic Resonance Imaging (fMRI) data that can be shared in place of an actual patient’s scanned data. We will discuss the trends in accelerating the adoption of GPU compute to further empower fMRI data reconstruction from a data dimension reduction to produce optimal latent variables and ultimately apply stochastic techniques to the data in latent space and generate new synthetic fMRI images.

Registration is limited to 60.

Presenter:

• Amir Bahmanyari, Advisory AI & Data Science Solutions Architect, Dell

Accommodation requests related to a disability should be made to Jared Haddock at jared.haddock@oregonstate.edu or 541-737-2367 at least five (5) business days prior to the event.