WOODS: Benchmarks for Out-of-Distribution Generalization in Time Series

Jean-Christophe Gagnon-Audet, Kartik Ahuja, Mohammad Javad Darvishi Bayazi et al.

Action editor: Antoni Chan.

https://openreview.net/forum?id=mvftzofTYQ

#generalization #generalize #datasets

WOODS: Benchmarks for Out-of-Distribution Generalization in Time Series

https://openreview.net/forum?id=mvftzofTYQ

#generalization #generalize #datasets

I would argue the named entity recognition (#NER) is the simplest, best defined #NLP task with a current, practical application. One might think that NER is a solved problem. However, the best existing NER models still fail to #generalize well to named entities (names) that were not in their training data (despite decades of work).

I don't know what #sentience is, but I'm, on principle, not willing to grant #personhood to something that can't identify names it has never seen before.

Perception people, i'd be very interested to learn about distributions of time constants in #PrimarySensoryCortices. There's a particular choice (that we think we see in the #hippocampus) that lets #DeepNetworks #generalize in what seems like a pretty adaptive way. Distribution is #WeberFechnerLaw:
https://proceedings.mlr.press/v162/jacques22a.html