← Back to Sessions

EarthquakeNPP: Benchmarking Neural Point Processes in California and China

Description: 

Recent advancements in point process models within the machine learning community have led to the development of Neural Point Processes (NPPs), which promise greater flexibility and improvements over classical models such as the Epidemic-Type Aftershock Sequence (ETAS) model. EarthquakeNPP is a benchmarking platform designed to direct the development of NPPs for earthquake forecasting. The platform hosts a suite of NPP models alongside a benchmark implementation of the ETAS model, enabling standardized forecasting experiments. The platform defines consistent training and testing partitions of earthquake catalogs and evaluates model performance using log-likelihood and CSEP generative evaluation metrics. We present initial results from benchmarking experiments on datasets from California, including high-resolution catalogs, and China using the China Earthquake Networks Center (CENC) catalog. While "off-the-shelf" NPP models currently do not surpass the ETAS model, we highlight ongoing efforts to adapt NPPs specifically for earthquake forecasting.

Session: Building and Decoding High-resolution Earthquake Catalogs With Statistical and Machine-learning Tools - I

Type: Oral

Date: 4/15/2025

Presentation Time: 04:30 PM (local time)

Presenting Author: Samuel

Student Presenter: No

Invited Presentation: Yes

Poster Number:

Authors