Unconstrained Dynamic Regret via Sparse Coding

Authors: Zhiyu Zhang, Ashok Cutkosky, Yannis Paschalidis

NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The paper concludes with an application in fine-tuning time series forecasters, where unconstrained dynamic OCO is naturally motivated. Due to limited space, this is deferred to Appendix E, with experiments that support our theoretical results.
Researcher Affiliation Academia Zhiyu Zhang Harvard University zhiyuz@seas.harvard.edu Ashok Cutkosky Boston University ashok@cutkosky.com Ioannis Ch. Paschalidis Boston University yannisp@bu.edu
Pseudocode Yes Algorithm 1 Sparse coding with size 1 dictionary. ... Algorithm 2 Sparse coding with general dictionary. ... Algorithm 3 FREEGRAD [MK20, Definition 4]: scale-free and gradient adaptive unconstrained static OLO. ... Algorithm 4 Haar OLR with known time horizon. ... Algorithm 5 Anytime Haar OLR (Algorithm 4 with doubling trick).
Open Source Code Yes Code is available at https://github.com/zhiyuzz/Neur IPS2023-Sparse-Coding.
Open Datasets Yes Here we use the Jena weather forecasting dataset,16 which records the weather data at a German city, Jena, every 10 minutes. ... 16Available at https://www.bgc-jena.mpg.de/wetter/.
Dataset Splits No The paper uses synthetic data and a weather dataset, but does not explicitly provide information on train/validation/test splits, percentages, or sample counts for these datasets.
Hardware Specification No The paper does not specify any hardware details (e.g., CPU, GPU models, or memory) used for running the experiments.
Software Dependencies No The paper discusses algorithms and frameworks but does not specify any software dependencies with version numbers (e.g., Python, PyTorch, TensorFlow versions).
Experiment Setup Yes Both algorithms require a confidence hyperparameter ε, and we set it to 1.