The Google-brewed paper has thought-provoking ideas and is poised to be remembered for a long time. The paper’s key question is: can indexing data structures (e.g. B-Trees or HashMaps) be replaced with machine learning models (e.g. Neural Networks) ? (Figure 1). Indexes as predictive modelsThis simple question completely broke mine and others’ previous idea of what an index is: Before: indexes ==
![The Case for Learned Index Structures](https://cdn-ak-scissors.b.st-hatena.com/image/square/618754e29e1612ac04d7540405b765c07cea4f23/height=288;version=1;width=512/https%3A%2F%2Fmiro.medium.com%2Fv2%2Fresize%3Afit%3A1200%2F1%2AyHbgRO_yieWXKZDOXamziQ.png)