skip to main content


Title: A Stein-Papangelou Goodness-of-Fit Test for Point Processes.
Point processes provide a powerful framework for modeling the distribution and interactions of events in time or space. Their flexibility has given rise to a variety of sophisticated models in statistics and machine learning, yet model diagnostic and criticism techniques re- main underdeveloped. In this work, we pro- pose a general Stein operator for point pro- cesses based on the Papangelou conditional intensity function. We then establish a kernel goodness-of-fit test by defining a Stein dis- crepancy measure for general point processes. Notably, our test also applies to non-Poisson point processes whose intensity functions con- tain intractable normalization constants due to the presence of complex interactions among points. We apply our proposed test to sev- eral point process models, and show that it outperforms a two-sample test based on the maximum mean discrepancy.  more » « less
Award ID(s):
1816499
NSF-PAR ID:
10105515
Author(s) / Creator(s):
; ;
Date Published:
Journal Name:
Artificial Intelligence and Statistics (AISTATS 2019)
Format(s):
Medium: X
Sponsoring Org:
National Science Foundation
More Like this
  1. Gradient-based approximate inference methods, such as Stein variational gradient descent (SVGD), provide simple and general-purpose inference engines for differentiable continuous distributions. However, existing forms of SVGD cannot be directly applied to discrete distributions. In this work, we fill this gap by proposing a simple yet general framework that transforms discrete distributions to equivalent piecewise continuous distributions, on which the gradient-free SVGD is applied to perform efficient approximate inference. The empirical results show that our method outperforms traditional algorithms such as Gibbs sampling and discontinuous Hamiltonian Monte Carlo on various challenging benchmarks of discrete graphical models. We demonstrate that our method provides a promising tool for learning ensembles of binarized neural network (BNN), outperforming other widely used ensemble methods on learning binarized AlexNet on CIFAR-10 dataset. In addition, such transform can be straightforwardly employed in gradient-free kernelized Stein discrepancy to perform goodness-of-fit (GOF) test on discrete distributions. Our proposed method outperforms existing GOF test methods for intractable discrete distributions. 
    more » « less
  2. We present a multimodel analysis for mechanistic hypothesis testing in landscape evolution theory. The study site is a watershed with well‐constrained initial and boundary conditions in which a river network locally incised 50 m over the last 13 ka. We calibrate and validate a set of 37 landscape evolution models designed to hierarchically test elements of complexity from four categories: hillslope processes, channel processes, surface hydrology, and representation of geologic materials. Comparison of each model to a base model, which uses stream power channel incision, uniform lithology, hillslope transport by linear diffusion, and surface water discharge proportional to drainage area, serves as a formal test of which elements of complexity improve model performance. Model fit is assessed using an objective function based on a direct difference between observed and simulated modern topography. A hybrid optimization scheme identifies optimal parameters and uncertainty. Multimodel analysis determines which elements of complexity improve simulation performance. Validation tests which model improvements persist when models are applied to an independent watershed. The three most important model elements are (1) spatial variation in lithology (differentiation between shale and glacial till), (2) a fluvial erosion threshold, and (3) a nonlinear relationship between slope and hillslope sediment flux. Due to nonlinear interactions between model elements, some process representations (e.g., nonlinear hillslopes) only become important when paired with the inclusion of other processes (e.g., erosion thresholds). This emphasizes the need for caution in identifying the minimally sufficient process set. Our approach provides a general framework for hypothesis testing in landscape evolution.

     
    more » « less
  3. We characterize the asymptotic performance of nonparametric goodness of fit testing. The exponential decay rate of the type-II error probability is used as the asymptotic performance metric, and a test is optimal if it achieves the maximum rate subject to a constant level constraint on the type-I error probability. We show that two classes of Maximum Mean Discrepancy (MMD) based tests attain this optimality on Rd, while the quadratictime Kernel Stein Discrepancy (KSD) based tests achieve the maximum exponential decay rate under a relaxed level constraint. Under the same performance metric, we proceed to show that the quadratic-time MMD based two-sample tests are also optimal for general two-sample problems, provided that kernels are bounded continuous and characteristic. Key to our approach are Sanov’s theorem from large deviation theory and the weak metrizable properties of the MMD and KSD. 
    more » « less
  4. While open-source software has become ubiquitous, its sustainability is in question: without a constant supply of contributor effort, open-source projects are at risk. While prior work has extensively studied the motivations of open-source contributors in general, relatively little is known about how people choose which project to contribute to, beyond personal interest. This question is especially relevant in transparent social coding environments like GitHub, where visible cues on personal pro"le and repository pages, known as signals, are known to impact impression formation and decision making. In this paper, we report on a mixed-methods empirical study of the signals that influence the contributors’ decision to join a GitHub project. We first interviewed 15 GitHub contributors about their project evaluation processes and identified the important signals they used, including the structure of the README and the amount of recent activity. Then, we proceeded quantitatively to test out the impact of each signal based on the data of 9,977 GitHub projects. We reveal that many important pieces of information lack easily observable signals, and that some signals may be both attractive and unattractive. Our findings have direct implications for open-source maintainers and the design of social coding environments, e.g., features to be added to facilitate better project searching experience 
    more » « less
  5. Abstract

    We develop a prior probability model for temporal Poisson process intensities through structured mixtures of Erlang densities with common scale parameter, mixing on the integer shape parameters. The mixture weights are constructed through increments of a cumulative intensity function which is modeled nonparametrically with a gamma process prior. Such model specification provides a novel extension of Erlang mixtures for density estimation to the intensity estimation setting. The prior model structure supports general shapes for the point process intensity function, and it also enables effective handling of the Poisson process likelihood normalizing term resulting in efficient posterior simulation. The Erlang mixture modeling approach is further elaborated to develop an inference method for spatial Poisson processes. The methodology is examined relative to existing Bayesian nonparametric modeling approaches, including empirical comparison with Gaussian process prior based models, and is illustrated with synthetic and real data examples.

     
    more » « less