Tom Merritt-Smith - of our new sponsors Caspian - is with us this month to teach us about Weak Supervision in AI.
The availability of machine learning packages with friendly APIs, alongside ever
cheaper technology, has made building a gigantic neural network child’s play.
However, developing a useful machine learning classifier is often dependent on
having an accurately labelled data set which can be used to train the
classifier, and the bigger the neural network, the more hungry it is. This comes
at a significant cost - even generating 1 label per second, 100 000 labels would
require ~28 hours of work for one person. In this talk, we’ll cover some
approaches to solve this, some challenges, and have a look at a python package
released by one of the key research groups on weak supervision:
There will be some code-along / code-play opportunities during and after the talk, so bring along your laptop. The package we’ll look at is babble, if you want to try installing it in advance.