CacheMamba: Popularity Prediction for Mobile Edge Caching Networks via Selective State Spaces

TL;DR

This work tackles popularity prediction for Mobile Edge Caching (MEC) by formulating it as a ranking problem to maximize cache-hit rate at edge updates. It introduces CacheMamba, a Mamba-based state-space model (SSM) architecture that efficiently predicts the top-K context files most likely to be requested, and it benchmarks against Transformer-based models. Empirical results on MovieLens-32M-derived data show that CacheMamba achieves higher MAP@K and NDCG@K scores, as well as better cache-hit rates, with markedly lower increases in computational cost as the sequence length grows. The findings suggest CacheMamba offers scalable, accurate, and resource-efficient popularity prediction for MEC, enabling more effective edge caching decisions in latency-sensitive applications like AR/VR and autonomous vehicles.

Abstract

Mobile Edge Caching (MEC) plays a pivotal role in mitigating latency in data-intensive services by dynamically caching frequently requested content on edge servers. This capability is critical for applications such as Augmented Reality (AR), Virtual Reality (VR), and Autonomous Vehicles (AV), where efficient content caching and accurate popularity prediction are essential for optimizing performance. In this paper, we explore the problem of popularity prediction in MEC by utilizing historical time-series request data of intended files, formulating this problem as a ranking task. To this aim, we propose CacheMamba model by employing Mamba, a state-space model (SSM)-based architecture, to identify the top-K files with the highest likelihood of being requested. We then benchmark the proposed model against a Transformer-based approach, demonstrating its superior performance in terms of cache-hit rate, Mean Average Precision (MAP), Normalized Discounted Cumulative Gain (NDCG), and Floating-Point Operations Per Second (FLOPS), particularly when dealing with longer sequences.

Figures (2)

• Figure 1: Illustration of the proposed CacheMamba architecture;(a) Block diagram of Mamba (left), and (b) CacheMamba model (right).
• Figure 2: Comparing the impact of window length $L$ in CacheMamba and Transformer-based models on cache-hit rate and the number of FLOPS.