4 ms·
Robust learning from untrusted sources
- amitport 7y agoSome of the comparisons do not seem fair. Here the optimizer has all the data (though it was collected from untrusted sources) while in some of the mentioned works the optimization steps are done at data owners. Specifically this means that trust related regularizers aren't as useful.
- ivan_ah 7y agoMore recently, the Snorkel framework has been released, which seems to be solving the dataset-melding problem too: https://www.snorkel.org/ https://www.snorkel.org/ https://github.com/snorkel-team/snorkel https://github.com/snorkel-team/snorkel
- scribu 7y agoSnorkel is mentioned in the article and is at least 1 year old.