We explore learned index structures that use machine learning models to predict where data is stored. By combining traditional data structures like arrays and trees with lightweight models, we can make search operations much faster than with conventional methods.

To support frequent updates, we design updatable versions of indexes like RMI, ALEX, and LIPP. We also apply SIMD acceleration to speed up lookups, and build hybrid designs like RB-SkipList, which merges lists and trees for efficient in-memory search. Our work includes practical experiments to measure real-world performance.