> The images are then sent to be processed and decoded, which leans on machine learning model to convert analog signals to digital data.
Looking at the paper [1] (section 3.2.):
> Machine learning models are able to better learn and account for any noise properties inherent in the end-to-end write and read processes including: inter-symbol interference between adjacent voxels in the glass, scattered light from neighbouring layers during readout, variability between optical components, and more. By contrast, traditional signal processing techinques require extensive understanding of all these characteristics and careful (often hand-crafted) processing to remove their effects.
It feels weird to me, that they target "sensitive industries including finance, scientific research and healthcare", but decoding/reading the data is not really deterministic.
[1] https://www.microsoft.com/en-us/research/uploads/prod/2023/0...