Abstract
Systems for data indexing can involve a trade-off between the upfront cost of static indexes and potentially suboptimal performance from on-demand data scanning, while some reactive adaptive methods may incur an initial query latency penalty. This disclosure describes systems and methods for predictive, learned adaptive indexing. A forecasting component, which may use a sequence-aware machine learning model, can analyze historical query patterns to predict future data access ranges. Based on these predictions, a system can speculatively trigger just-in-time index construction on anticipated data partitions, for example, using available system resources. This proactive approach, which can be combined with a reactive learned sorting mechanism for unpredicted queries, may help reduce the initial latency penalty for predictable workloads. Such a technique can provide performance that may be similar to a pre-indexed database with the flexibility and low initialization cost of an adaptive system.
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 License.
Recommended Citation
Marodia, Mukesh Kumar; Singla, Virender Kumar; and Uttam, Saurabh, "Proactive Learned Indexing Driven by Query Workload Forecasting", Technical Disclosure Commons, (January 29, 2026)
https://www.tdcommons.org/dpubs_series/9250