Nonstationary sparse spectral permanental process

Zicheng Sun, Yixuan Zhang, Zenan Ling, Xuhui Fan, Feng Zhou*

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference proceeding contributionpeer-review

Abstract

Existing permanental processes often impose constraints on kernel types or stationarity, limiting the model's expressiveness. To overcome these limitations, we propose a novel approach utilizing the sparse spectral representation of nonstationary kernels. This technique relaxes the constraints on kernel types and stationarity, allowing for more flexible modeling while reducing computational complexity to the linear level. Additionally, we introduce a deep kernel variant by hierarchically stacking multiple spectral feature mappings, further enhancing the model's expressiveness to capture complex patterns in data. Experimental results on both synthetic and real-world datasets demonstrate the effectiveness of our approach, particularly in scenarios with pronounced data nonstationarity. Additionally, ablation studies are conducted to provide insights into the impact of various hyperparameters on model performance. Code is publicly available at https://github.com/SZC20/DNSSPP.

Original languageEnglish
Title of host publicationNeurIPS 2024
Subtitle of host publication38th Conference on Neural Information Processing Systems: proceedings
EditorsA. Globerson, L. Mackey, D. Belgrave, A. Fan, U. Paquet, J. Tomczak, C. Zhang
Place of PublicationSydney, NSW
PublisherCurran Associates
Pages1-24
Number of pages24
ISBN (Print)9798331314385
Publication statusPublished - 2024
EventConference on Neural Information Processing Systems (38th : 2024) - Vancouver, Canada
Duration: 10 Dec 202415 Dec 2024

Publication series

NameAdvances in Neural Information Processing Systems
Volume37
ISSN (Print)1049-5258

Conference

ConferenceConference on Neural Information Processing Systems (38th : 2024)
Abbreviated titleNeurIPS 2024
Country/TerritoryCanada
CityVancouver
Period10/12/2415/12/24

Cite this