Machine learning for improving computer architecture like memory management techniques

In this work, we propose an ML-empowered memory management framework focused on improving performancein computer architecture by applying predictive, adaptive, and preemptive memory optimization techniques. Byintegrating supervised learning to predict memory access, reinforcement learning for adaptive cache management,and unsupervised learning for prefetching, the framework will be able to conduct dynamic management of memoryresources with significant latency reduction, improved cache hit rates, and better energy efficiency. Indeed, tested ona wide array of workloads comprising machine learning inference, database operations, and scientific computing,the framework demonstrated substantive gains in performance, hence showing its adaptability to high-demandenvironments. Our results show that the framework can indeed perform runtime intelligent optimization of memorymanagement. Thus, the ground is paved for testing on most contemporary computing systems with high demands ofdata and computations.

کلیدواژه ها:

machine learning ، memory management ، cache optimization ، hybrid memory ، computer architecture