CNI PhD Scholar Ankita Koley recently presented her joint work with Prof. Chandramani Singh on Dynamic Content Caching at IEEE MASS 2024. Dynamic content, such as YouTube recommendations and Facebook newsfeeds, is continuously updated on central servers, and its timely delivery is essential to maintain user relevance.
Their research addresses the dynamic content caching problem, where a local cache at a base station (BS) temporarily stores selected content to reduce delays. The BS must decide whether to serve content from its local cache or fetch a fresh version from the central server, balancing two key costs: fetching cost and ageing cost, which reflects the outdatedness of cached content. Adding to the complexity, the BS often has only partial information about how outdated its cached content is.
To solve this, they formulated an optimal content fetching and caching problem to minimize the average cost subject to cache capacity constraints. The problem suffers from the curse of dimensionality and is provably hard to solve. They model it as a continuous-time, restless multi-armed bandit process (RMAB), with each content modelled as a partially observable Markov decision process.
They proved the model's indexability, derived explicit expressions for the Whittle index, and developed a Whittle index-based solution. This approach, is further validated through simulations, and demonstrated asymptotically optimal performance, outperforming recent methods.
Check out their paper to learn more!!
