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