Paper 10: Artificial Intelligence’s Impact on Data Structures
Authors : Plaaksha Chaudhry, Tanya Verma, Dr Gaurvi Shukla, Dr Shalini Lamba
Doi: https://doi.org/10.63920/tjths.44010
Abstract
To counter the growing number of modern workload complexities, system designers are increasingly incorporating machine learning (ML) to optimize software performance. Traditional system components like data structures, caches, memory allocators, garbage collectors, and database optimizers employ pre-computed heuristics or analytical models that guarantee worst-case performance but neglect patterns observed in actual workloads. Data-driven methods can boost average case performance by predicting demands, critical resources, and access patterns based on historical execution traces and observations during runtime. These applications include learning-based auto-tuning databases for adapting to varying workloads, ML-assisted memory management for improving locality and reducing fragmentation issues, learning-based data structures with predictive caching for reducing latency and cache misses. While end-to-end systems based on ML learning are still being researched owing to limitations with inference time complexity, retraining expenses, and robustness to distributional shifts, a complementary learning approach utilizing small ML models along with algorithms is preferable to ensure feasible performance improvements. Future research topics on learning-enhanced system design will be discussed in the following sections.
