Vinsamlegast notið þetta auðkenni þegar þið vitnið til verksins eða tengið í það: https://hdl.handle.net/1946/39398
AI researchers at Silicon Valley based biotech company, Natera Inc, have developed deep learning neural networks for chromosomal anomaly detection in unborn fetuses, using the cell free DNA from plasma extracted with a simple blood draw from the mother. The algorithms have increased the detection accuracy of various conditions. After years of research and development, the AI team is implementing a machine learning production pipeline to bring these groundbreaking AI tools to users. This will give millions of parents the opportunity to understand the health of their unborn babies more accurately. In this white paper, we will explore the pros and cons of using a particular framework, called TensorFlow Extended (TFx), when deploying machine learning solutions into production. We will look at each component of the TFx pipeline and
discuss its efficacy for the projects and how it could be utilized by
Natera. We will also pay particular attention to the Machine Learning
Metadata Store (MLMD), a database that can be used with or
independent of TensorFlow Extended. Our goal is to explore how
neural network based deep learning projects navigate from R&D into
the production phase. The deliverables from this project
are a white paper, Proposal for Natera Machine Learning production
pipelines using TensorFlow Extended, a custom database API that
utilized a TFx component metadata store database and these supporting
documents.
Skráarnafn | Stærð | Aðgangur | Lýsing | Skráartegund | |
---|---|---|---|---|---|
Proposal for Natera Machine Learning production pipeline using TensorFlow Extended.pdf | 1.89 MB | Opinn | Heildartexti | Skoða/Opna | |
ML Pipelines with TFx - Supporting Documents.pdf | 580.57 kB | Opinn | Fylgiskjöl | Skoða/Opna |