Classifying signals and simulations representing data from the Allen Telescope Array (ATA), in partnership with Adam Cox (@gadamc) from IBM.

Representative (clean) signal types:

The main challenge was that a large fraction of samples have a very poor SNR, with many examples hard to discern by eye.

SETINet V3 Activations for poor SNR example:

Final presentation: link

Final report: link

commonutils (library)/

• datautils: utilities for pre-processing, creating dataset, visualizing activations
• modelspecs: keras model specifications SETINet V1, V2, V3
• traceutils: utilities for dynamic-programming based feature extractor (path-trace)
• nputils, modelutils: misc utility functions for pre-processing, debugging

Akash/

• Various scripts / notebooks for training models, documenting results

matlab-exploration/

• experimentation with image / signal processing techniques