6fe43269967adbb64ec6149852b5cc3e-AuthorFeedback.pdf
–Neural Information Processing Systems
Examples include Xu et al., (2016, arXiv:1602.04511), Mei et al.3 (2017, arXiv:1612.09328) and almost all our cited papers on point process. This is because the prediction of next4 arrival is essentially predicting a distribution, and log-likelihood can give a better characterization of the predicted5 distribution compared toRMSE. For example, in the MathOverflow dataset, a user that just answered one question usually comments on12 severalotheranswers. Toreviewer3: Wewill add more discussion regarding the performance of different models in the next version.
