Parsimonious Learning-Augmented Caching

Authors: Sungjin Im, Ravi Kumar, Aditya Petety, Manish Purohit

ICML 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We experimentally evaluate our algorithm on a real-world dataset and demonstrate the empirical dependence of the competitive ratio on the number of queries made as well as on the prediction errors.
Researcher Affiliation Collaboration 1UC Merced 2Google Mountain View.
Pseudocode Yes Algorithm 1: A generic marking algorithm. Algorithm 2: Naive eviction. Algorithm 3: Adaptive query eviction.
Open Source Code No The paper does not provide a specific link or statement indicating that the source code for the described methodology is publicly available.
Open Datasets Yes We use the Citi Bike dataset as in Lykouris & Vassilvitskii (2018). The dataset comes from a publicly-available1 bike sharing platform in New York City.
Dataset Splits No The paper describes using each month's data as an input sequence of 25,000 events to evaluate the online caching algorithms. It does not provide specific train/validation/test splits for these sequences.
Hardware Specification No The paper does not provide specific details regarding the hardware used to run the experiments.
Software Dependencies No The paper does not provide specific software dependencies with version numbers.
Experiment Setup Yes Finally we set the cache size k = 500. For each page p in the cache, its predicted next request time is set to its actual next request time plus a noise, which is drawn i.i.d. from a lognormal distribution whose underlying normal distribution has mean 0 and standard deviation σ. Our algorithm is parameterized by b, the number of queries made per cache miss.