Understanding Deep Learning requires rethinking generalization

Very entertaining article which could be viewed either as philosophical discoveries, or as "machine learning specialists rediscover information theory" :)
https://arxiv.org/abs/1611.03530