Notice: The reproducibility variables underlying each score are classified using an automated LLM-based pipeline, validated against a manually labeled dataset. LLM-based classification introduces uncertainty and potential bias; scores should be interpreted as estimates. Full accuracy metrics and methodology are described in Coakley et alK. L. Coakley, T. Snelleman, H. Hoos, and O. E. Gundersen, "The embrace of open science: An analysis of a decade of AI research and 56 800 conference papers," Under Review, 2026..

WaLRUS: Wavelets for Long range Representation Using State Space Methods

Authors: Hossein Babaei, Mel White, Sina Alemohammad, Richard Baraniuk

NeurIPS 2025 | Venue PDF | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We compare Wa LRUS to Hi PPO-based models, and demonstrate improved accuracy and more efficient implementations for online function approximation tasks. ... Empirical results demonstrate that the wavelet-based Wa LRUS model consistently outperforms Legendre and Fourier-based Hi PPO models in reconstruction accuracy, especially on signals with sharp transients.
Researcher Affiliation Academia Hossein Babaei Mel White Sina Alemohammad Richard G. Baraniuk Department of Electrical and Computer Engineering, Rice University EMAIL
Pseudocode No The paper describes the methods mathematically and textually, particularly in Section 2 and Appendix A.1, but does not include any explicitly labeled pseudocode blocks or algorithm figures.
Open Source Code Yes Code to generate the matrices is available at the following repository: https://github.com/echbaba/walrus. ... Code and data are available at https://osf.io/7kjcx/?view_only= 5dc38b9776624deb9d1c0d8f88108658
Open Datasets Yes M4 Forecasting Competition [36]: A diverse collection of univariate time series... Speech Commands [37]: A dataset of one-second audio clips... Wavelet Benchmark Collection [38]: A synthetic benchmark featuring signals with distinct singularity structures...
Dataset Splits No The paper mentions generating or testing on '3,000 random instances' or '1000 random sequences, each containing 10 spikes', but it does not provide specific training/test/validation splits for a fixed dataset, as the evaluation focuses on online function approximation rather than traditional model training with predefined splits.
Hardware Specification No Appendix A.2.5 "Computational resources" states: "Within the scope of this paper, no networks were trained and no parameters were learned. Only CPU resources were utilized, but speed could be improved with parallel resources on a GPU." This statement is too general and does not provide specific details on CPU or GPU models, memory, or other hardware specifications.
Software Dependencies No The paper does not explicitly list specific software libraries, frameworks, or languages with their version numbers.
Experiment Setup Yes Appendix A.2.4 "Wavelet frames used for each experiment" and Table 3 provide detailed parameters for specifying the redundant wavelet frame, including Wavelet Function (D22), L (Frame Length = 2^19), Scale min, Shift, and rcond (0.01), as well as the effective sizes (Neff) for all compared SSMs across different experiments.