Presentation
Domain-Aware Performant AI-Based Compression
DescriptionMicroscopes play a critical role in scientific discoveries. A grand challenge in microscopy-based research is to manage the significantly high data volume and velocity of data generation while ensuring real-time analysis and closed-loop microscope operation. The scanning transmission electron microscope (STEM) at PNNL can generate 1100 frames per second with 128x128 pixels per frame, which leads to more than 5 GB of data generation per minute. The future microscopes could well support even higher data rates. The current data processing solutions utilize sampling to manage such a high data volume and velocity. However, it's important to note that sampling can lead to the loss of critical information. Consequently, complete data processing and analytics are necessary. Data compression techniques that reduce data size while retaining key properties on individual images appear to be a promising solution.
