Q-Learning Bell Curve Meme

Hihi Q-learning make agent go vroom vroom Noooo! Humans build models of the world with analogies and relationships. You can't just take a black box approach to solving intelligence! You need regularize the entropy of the policy function or else the training will be probably unstable! You can't just try out random hyper-parameters! You need to construct reasonable priors! Simple techniques like Q-learning are the easiest to scale Most benchmarks are BS 1st year Student ML Researcher Cited ML Researcher

A bell-curve meme contrasts a beginner, a machine-learning researcher, and a highly cited researcher discussing Q-learning. The beginner celebrates the simple “make agent go vroom vroom” approach, while the middle figure overcomplicates reinforcement-learning theory and the expert argues that simple techniques are easiest to scale and many benchmarks are meaningless. It is used as a reaction to academic overthinking and skepticism toward machine-learning research benchmarks.

• @profunctor_io / 6148 2021-01-16