Robust Learning-Augmented Caching: An Experimental Study
Authors: Jakub Chłędowski, Adam Polak, Bartosz Szabucki, Konrad Tomasz Żołna
ICML 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We are the first to comprehensively evaluate these learning-augmented algorithms on real-world caching datasets and state-of-the-art machine-learned predictors. We show that a straightforward method blindly following either a predictor or a classical robust algorithm, and switching whenever one becomes worse than the other has only a low overhead over a wellperforming predictor, while competing with classical methods when the coupled predictor fails, thus providing a cheap worst-case insurance. |
| Researcher Affiliation | Collaboration | 1Jagiellonian University, Kraków, Poland 2EPFL, Lausanne, Switzerland 3Deep Mind, London, United Kingdom. |
| Pseudocode | No | The paper describes algorithms verbally but does not include any pseudocode or clearly labeled algorithm blocks. |
| Open Source Code | Yes | The source code is publicly available at https://github.com/chledowski/ml_caching_with_guarantees. |
| Open Datasets | Yes | Our datasets come from the 2nd Cache Replacement Championship (CRC, 2017) and consist of real-world memory access traces from the SPEC CPU2006 benchmark (Henning, 2006). ... made them publicly available at https://github.com/chledowski/Robust-Learning-Augmented-Caching-An-Experimental-Study-Datasets. |
| Dataset Splits | Yes | The first 80% of this sequence is used for training, followed by 10% used for validation, and the last 10% for testing. |
| Hardware Specification | No | The paper mentions "constrained computing resources" but does not provide any specific details about the hardware used for the experiments (e.g., GPU/CPU models, memory specifications). |
| Software Dependencies | No | The paper refers to using existing codebases like Parrot and mentioning components like LSTM and Transformer, but it does not specify software versions (e.g., Python version, deep learning framework version like TensorFlow or PyTorch, CUDA version) for reproducibility. |
| Experiment Setup | Yes | Due to constrained computing resources, we limited the number of training steps to 20 000. ... As a result, our models are trained for 20 000 steps on each dataset, with a batch size of 32. The best model is chosen to be the one with the highest validation cache hit rate among the evaluated checkpoints (done every 5000 steps). |