SQL performance (leaving out details because self-explantory): Caching Auxiliary data, like min and max values of different columns in a file Data layout optimizations ML and analytics workloads: Declarative Dataframe APIs that also leverage the transaction log, caching, auxiliary data and data layout optimizations already described above Possible future work Come up with completely new storage formats for enabling more data layout optimizations or better suited for modern hardware More possibilities in caching, storing auxiliary data, and data layout optimizations Where and how to use serverless computing systems “Factorized ML” frameworks that push ML logic into SQL joins Can “feature stores” also benefit from the table versioning that Lakehouses offer?