Making Machine Learning Reproducible with CodaLab

Опубликовано 21 июня 2016, 21:00

We often come across work in Machine Learning which is difficult to reproduce because either the parameters used were not reported in the literature or the experiment setup was not the same. In my intern work I show that CodaLab Worksheets can be used to solve this problem. I was able to reproduce results for two papers: - Machine Comprehension Test, by M. Richardson, C. J.C. Burges, and E. Renshaw, EMNLP 2013 - FActored Spectrally Transformed Linear Mixed Models C. Lippert, J. Listgarten, Y. Liu, C.M. Kadie, R.I. Davidson, and D. Heckerman, Nature 2011. I also showed that CodaLab Worksheets can be used to put forward a tutorial as I will demonstrate by presenting a tutorial on Calibration with two algorithms for Calibration. I will end the talk with the research work that I am pursuing on intelligible models for multi-task learning.