Next version

We plan to include the following changes to future versions:

• Transfer learning module

• Tensorflow graph computation of spectrograms

• New neural network architectures

  • Sequence model

  • Similarity model (e.g. Siamese network)

  • Object-detection network (e.g. R-CNN)

• Overhaul of html docs (style, structure and content)

• Review all tutorials, and add new ones as needed

• Publish peer-reviewed software paper about Ketos (e.g. in JOSS)

• Functionalities to support active learning

• Methods to calibrate neural networks so that the output scores correspond more closely to probabilities/confidences.

• Neural architecture search capabilities, allowing recommendation of the optimal architecture to be used for a given task.