Estimating school completion is crucial for monitoring SDG 4 on education. The recently introduced SDG indicator 4.1.2, defined as the percentage of children aged 3-5 years above the expected completion age of a given level of education that have completed the respective level, differs from enrolment indicators in that it relies primarily on household surveys. This introduces a number of challenges including gaps between survey waves, conflicting estimates, age misreporting, and delayed completion. Our Adjusted Bayesian Completion Rates (ABC) model addresses these challenges to produce the first complete and consistent time series for SDG indicator 4.1.2, by school level and sex, for 157 countries. The ABC model estimates unobserved true completion rates using a latent ARIMA(1,1,0) with drift process. The model adjusts observations for late completion and age misreporting effects, and also accounts for survey level differences in bias and non-sampling variance. Validation exercises indicate that the model appears well-calibrated and offers a meaningful improvement over simpler approaches in predictive performance.

We introduce a family of Markov Chain Monte Carlo (MCMC) methods designed to sample from target distributions with irregular geometry using an adaptive scheme. In cases where targets exhibit non-Gaussian behaviour, we propose that adaption should be regional in nature as opposed to global. Our algorithms minimize the information projection side of the Kullback-Leibler (KL) divergence between the proposal distribution class and the target to encourage proposals distributed similarly to the regional geometry of the target. Unlike traditional adaptive MCMC, this procedure rapidly adapts to the geometry of the current position as it explores the space without the need for a large batch of samples. We extend this approach to multimodal targets by introducing a heavily tempered chain to enable faster mixing between regions of interest. The divergence minimization algorithms are tested on target distributions with multiple irregularly shaped modes and we provide results demonstrating the effectiveness of our methods.

Facial verification is a core problem studied by researchers in computer vision. Recently published one-to-one comparison models have successfully achieved accuracy results that surpass the abilities of humans. A natural extension to the one-to-one facial verification problem is a one-to-many classification. In this abstract, we present our exploration of different methods of performing one-to-many facial verification using low-resolution images. The CSEye model introduces a direct comparison between the features extracted from each of the candidate images and the suspect before performing the classification task. Initial experiments using 10-to-1 comparisons of faces from the Labelled Faces of the Wild dataset yield promising results.

The out-of-school rate is a critical indicator for monitoring global progress towards universal education. It quantifies the population of children and youth excluded from each level of the education system. As with many education indicators, historical out-of-school reporting has relied exclusively on imperfect administrative data. Recently, the education community has turned to survey data as a supplement to administrative data to overcome its gaps and weaknesses. Producing such consolidated estimates globally in the out-of-school rate context, however, is a challenging task due to the diversity in enrolment patterns, systematic differences in the nature and reliability of administrative and survey-based data, and the heavy presence of invalid administrative observations resulting from enrolment counts that exceed corresponding population estimates. In this paper we introduce a cohort-based Bayesian hierarchical model to address these challenges and produce complete time series of out-of-school rates for 192 countries. The model uses a flexible spline-based process for underlying cohort out-of-school rate curves that are smoothed through cohort progression and over time. Observations are related to these values using a dual likelihood setup where each data source has distinct bias and variance components. The administrative side includes a structure that propagates uncertainty information contained in invalid data to avoid understating uncertainty. Validation exercises and sensitivity analysis suggest that the model is reasonably well calibrated and offers a material improvement over simpler approaches. The model is currently used by UNESCO to monitor out-of-school rates for all countries with available data.

Our goal is to develop a general strategy to decompose a random variable X into multiple independent random variables, without sacrificing any information about unknown parameters. A recent paper showed that for some well-known natural exponential families, X can be thinned into independent random variables X (1) , . . . , X (K) , such that X = K k=1 X (k) . In this paper, we generalize their procedure by relaxing this summation requirement and simply asking that some known function of the independent random variables exactly reconstruct X. This generalization of the procedure serves two purposes. First, it greatly expands the families of distributions for which thinning can be performed. Second, it unifies sample splitting and data thinning, which on the surface seem to be very different, as applications of the same principle. This shared principle is sufficiency. We use this insight to perform generalized thinning operations for a diverse set of families.