Score-matching estimators for continuous-time point-process regression models
Open Access
- 10 November 2016
- conference paper
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- Vol. 6, 1-5
Abstract
We introduce a new class of efficient estimators based on score matching for probabilistic point process models. Unlike discretised likelihood-based estimators, score matching estimators operate on continuous-time data, with computational demands that grow with the number of events rather than with total observation time. Furthermore, estimators for many common regression models can be obtained in closed form, rather than by iteration. This new approach to estimation may thus expand the range of tractable models available for event-based data.Keywords
This publication has 6 references indexed in Scilit:
- The Equivalence of Information-Theoretic and Likelihood-Based Methods for Neural Dimensionality ReductionPLOS Computational Biology, 2015
- Some extensions of score matchingComputational Statistics & Data Analysis, 2007
- Dimensionality reduction in neural models: An information-theoretic generalization of spike-triggered average and covariance analysisJournal of Vision, 2006
- Analyzing Neural Responses to Natural Signals: Maximally Informative DimensionsNeural Computation, 2004
- Approximating Point Process Likelihoods with GLIMJournal of the Royal Statistical Society Series C: Applied Statistics, 1992
- Random Point Processes in Time and Space1991