Scalable approximate Bayesian inference for particle tracking data

Authors: Ruoxi Sun, Liam Paninski

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We apply the method to simulated and real data. We show that the method robustly performs approximate Bayesian inference on the observed data, and provides more accurate results than competing methods that output just a single best path.
Researcher Affiliation Academia 1Department of Biological Sciences; 2Departments of Statistics and Neuroscience; Grossman Center for the Statistics of Mind; Center for Theoretical Neuroscience; Columbia University.
Pseudocode Yes Algorithm 1 Conditional sampling network
Open Source Code Yes Code is available here.
Open Datasets Yes The data are TIR-FM imaged clathrin-coated pits in a BSC1 cell (Jaqaman et al., 2008). We trained the network on simulated data whose parameters (signal-to-noise ratio, particle density and speed, psf width, etc.) were coarsely matched to the real data; see the comparison video for details.
Dataset Splits No The paper mentions training and testing data but does not explicitly describe a validation set or specific splits for reproducibility.
Hardware Specification No The paper mentions training times ('on the order of hours') and discusses network architectures but does not specify any hardware details like GPU/CPU models or memory used for experiments.
Software Dependencies No We trained the network (using default learning rate settings in Keras) to minimize the binary cross-entropy between the target mask (zero ex-cept at si t, or all zeros if all the particles in qt were already sampled and no further particles should be added) and the network s output probability mask.
Experiment Setup Yes We trained the network (using default learning rate settings in Keras) to minimize the binary cross-entropy between the target mask (zero ex-cept at si t, or all zeros if all the particles in qt were already sampled and no further particles should be added) and the network s output probability mask. We use M = 2 throughout this paper.