Avram Golbert - Zero Shot Learning for Kaggle Bengali Handwritten Recognition
Details
Zero Short Learning is the strange setup where we’re required at test time to handle never-before observed classes. While interesting academically, this also occurs in natural scenarios when you consider multi-label and Extreme classification (Such as retail).
Bengali Characters are a composite of letters, diacritics and accents. As such there are ~13000 possible Graphemes (combinations) to classify, most are not present in the train set.
This Kaggle Challenge provides a nice playground to talk about Generalized Zero Shot Learning, data imbalance and more.
*No linguistic background is required. Nor will be taught.
Join Zoom Meeting
https://us02web.zoom.us/j/6043600514?pwd=VTFuU2VSTTNhTE1RRFJTZjhZNTN1Zz09
Meeting ID: 604 360 0514
Password: 703769