Hey guys thank so much for the insight and the example!
Thanks @Eduardo Serna, I chose this network because it is currently one of the most suitable audio feature extractor in the Wolfram Neural Net Repository. The layers that were added are typical layers added to perform classification for sequences and I took inspiration from a classification example in BERT NN repo page for sentiment classification (see section: "Train a classifier model with the subword embeddings"), because it was text and hence a sequence too (audio is a sequence).
Concerning the mushroom classifier idea it would be cool to use Geographic Data and WeatherData to create some more advance classifier with priors. See below a toy example as a start for such mushroom classifier.
Recently, I learnt about an ambitious project trying to identify new species of mushrooms via their spores, which will be trapped automatically from air. It seems that we still don't know much about fungi, mycologists have estimated that only ~10 percent of the predicted number of fungi—which includes mushrooms and yeasts—has been described.
Hey Jofre, this is super fun and cool. Can you give some insight as to why you chose that particular neural net, and the transformations you did on it?
I have wanted for a while an easy way to help classify mushrooms, also using a geographic prior, even seasonal/weather information, smell, etc.
It might exist alread
- Congratulations! This post is now featured in our Staff Pick column as distinguished by a badge on your profile of a Featured Contributor! Thank you, keep it coming, and consider contributing your work to the The Notebook Archive!
A tour de force and model computational essay. The section on machine learning is particularly strong.