1 paper · 1 filter
Zixuan Wang, Xingyu Dang, Jason D. Lee +1
Natural language data follows a power-law distribution, with most knowledge and skills appearing at very low frequency. While a common intuition suggests that reweighting or curati…