37 - On Statistical Significance, Training Variance, and Why Reporting Score Distributions Matters - a podcast by Allen Institute for Artificial Intelligence

In this episode we talk about a couple of recent papers that get at the issue of training variance, and why we should not just take the max from a training distribution when reporting results. Sadly, our current focus on performance in leaderboards only exacerbates these issues, and (in my opinion) encourages bad science.

Papers:
https://www.semanticscholar.org/paper/Reporting-Score-Distributions-Makes-a-Difference-P-Reimers-Gurevych/0eae432f7edacb262f3434ecdb2af707b5b06481
https://www.semanticscholar.org/paper/Deep-Reinforcement-Learning-that-Matters-Henderson-Islam/90dad036ab47d683080c6be63b00415492b48506

Further episodes of NLP Highlights

Further podcasts by Allen Institute for Artificial Intelligence

Website of Allen Institute for Artificial Intelligence