Types of Machine Learning Papers

TYPES OF Machine learning PAPER Baseline is all you need Our gridworld experiments prove AGI is already here We got more data and it works better We got more compute and it works better We spent $1M on compute and it looks really cool Cherry-picked results look great We figured out how deep learning generalizes this time, I swear We proved a thing that’s been known empirically for 5 years We plugged this lego block into this other one Results are 0.3% better than that other paper! Results are 0.1% better than that other paper Are the results better than that other paper?

A 3-by-4 grid parodies the recurring styles and claims found in machine learning research papers. Each panel shows a different exaggerated justification for novelty, from using more data or compute to cherry-picking results and making tiny improvements over prior work. The final panel turns the whole process into a self-aware comparison against another paper, making it useful as a reaction to overhyped or incremental AI research.

• @profunctor_io / 6623 2021-04-30