https://doi.org/10.1051/epjconf/202429503013
Fast Integration of Poisson Distributions for Dead Sensor Marginalization
1 Skidmore College, Saratoga Springs, New York, USA.
2 Department of Computer and Information Sciences, University of Delaware, Newark, Delaware, USA.
3 Department of Physics and Astronomy, William Marsh Rice University, Houston, Texas, USA.
4 Department of Computer Science, William Marsh Rice University, Houston, Texas, USA.
* e-mail: zobilodeau@gmail.com
Published online: 6 May 2024
In dual-phase time-projection chambers there are photosensor arrays arranged to allow for inference of the positions of interactions within the detector. If there is a gap in data left by a broken or saturated photosensors, the inference of the position is less precise and less accurate. As we are unable to repair or replace photosensors once the experiment has begun, we develop methods to estimate the missing signals. Our group is developing a probabilistic graphical model of the correlations between the number of photons detected by adjacent photosensors that represents the probability distribution over photons detected as a Poisson distribution. Determining the posterior probability distribution over a number of photons detected by a sensor then requires integration over a multivariate Poisson distribution, which is computationally intractable for high-dimensions. In this work, we present an approach to quickly calculate and integrate over a multidimensional Poisson distribution. Our approach uses Zarr, a Python array compression package, to manage large multi-dimensional arrays and approximates the log factorial to quickly calculate the Poisson distribution without overflow.
© The Authors, published by EDP Sciences, 2024
This is an Open Access article distributed under the terms of the Creative Commons Attribution License 4.0, which permits unrestricted use, distribution, and reproduction in any medium, provided the original work is properly cited.
