Much of recent machine learning has focused on deep learning, in which neural network weights are trained through variants of stochastic gradient descent. An alternative approach comes from the field of neuroevolution, which harnesses evolutionary algorithms to optimize neural networks, inspired by the fact that natural brains themselves are the products of an evolutionary process. Neuroevolution enables important capabilities that are typically unavailable to gradient-based approaches, including learning neural network building blocks (for example activation functions), hyperparameters, architectures and even the algorithms for learning themselves. Neuroevolution also differs from deep learning (and deep reinforcement learning) by maintaining a population of solutions during search, enabling extreme exploration and massive parallelization. Finally, because neuroevolution research has (until recently) developed largely in isolation from gradient-based neural network research, it has developed many unique and effective techniques that should be effective in other machine learning areas too.

I'm happy to announce that version 0.8.0 of my ggeffects-package is on CRAN now. The update has fixed some bugs from the previous version and comes along with many new features or improvements. One major part that was addressed in the latest version are fixed and improvements for mixed models, especially zero-inflated mixed models (fitted with the glmmTMB-package). In this post, I want to demonstrate the different options to calculate and visualize marginal effects from mixed models. Basically, the type of predictions, i.e. whether to account for the uncertainty of random effects or not, can be set with the type-argument. The default, type "fe", means that predictions are on the population-level and do not account for the random effect variances.

Recent years have witnessed a widespread increase of interest in network representation learning (NRL). By far most research efforts have focused on NRL for homogeneous networks like social networks where vertices are of the same type, or heterogeneous networks like knowledge graphs where vertices (and/or edges) are of different types. There has been relatively little research dedicated to NRL for bipartite networks. Arguably, generic network embedding methods like node2vec and LINE can also be applied to learn vertex embeddings for bipartite networks by ignoring the vertex type information. However, these methods are suboptimal in doing so, since real-world bipartite networks concern the relationship between two types of entities, which usually exhibit different properties and patterns from other types of network data. For example, E-Commerce recommender systems need to capture the collaborative filtering patterns between customers and products, and search engines need to consider the matching signals between queries and webpages. This work addresses the research gap of learning vertex representations for bipartite networks. We present a new solution BiNE, short for Bipartite Network Embedding}, which accounts for two special properties of bipartite networks: long-tail distribution of vertex degrees and implicit connectivity relations between vertices of the same type. Technically speaking, we make three contributions: (1) We design a biased random walk generator to generate vertex sequences that preserve the long-tail distribution of vertices; (2) We propose a new optimization framework by simultaneously modeling the explicit relations (i.e., observed links) and implicit relations (i.e., unobserved but transitive links); (3) We explore the theoretical foundations of BiNE to shed light on how it works, proving that BiNE can be interpreted as factorizing multiple matrices.

Improving predictive understanding of Earth system variability and change requires data-model integration. Efficient data-model integration for complex models requires surrogate modeling to reduce model evaluation time. However, building a surrogate of a large-scale Earth system model (ESM) with many output variables is computationally intensive because it involves a large number of expensive ESM simulations. In this effort, we propose an efficient surrogate method capable of using a few ESM runs to build an accurate and fast-to-evaluate surrogate system of model outputs over large spatial and temporal domains. We first use singular value decomposition to reduce the output dimensions, and then use Bayesian optimization techniques to generate an accurate neural network surrogate model based on limited ESM simulation samples. Our machine learning based surrogate methods can build and evaluate a large surrogate system of many variables quickly. Thus, whenever the quantities of interest change such as a different objective function, a new site, and a longer simulation time, we can simply extract the information of interest from the surrogate system without rebuilding new surrogates, which significantly saves computational efforts. We apply the proposed method to a regional ecosystem model to approximate the relationship between 8 model parameters and 42660 carbon flux outputs. Results indicate that using only 20 model simulations, we can build an accurate surrogate system of the 42660 variables, where the consistency between the surrogate prediction and actual model simulation is 0.93 and the mean squared error is 0.02. This highly-accurate and fast-to-evaluate surrogate system will greatly enhance the computational efficiency in data-model integration to improve predictions and advance our understanding of the Earth system.

Datasets often contain biases which unfairly disadvantage certain groups, and classifiers trained on such datasets can inherit these biases. In this paper, we provide a mathematical formulation of how this bias can arise. We do so by assuming the existence of underlying, unknown, and unbiased labels which are overwritten by an agent who intends to provide accurate labels but may have biases against certain groups. Despite the fact that we only observe the biased labels, we are able to show that the bias may nevertheless be corrected by re-weighting the data points without changing the labels. We show, with theoretical guarantees, that training on the re-weighted dataset corresponds to training on the unobserved but unbiased labels, thus leading to an unbiased machine learning classifier. Our procedure is fast and robust and can be used with virtually any learning algorithm. We evaluate on a number of standard machine learning fairness datasets and a variety of fairness notions, finding that our method outperforms standard approaches in achieving fair classification.

Due to the truncated Gaussianity of the posterior, its distribution has to be approximated. In this work, we consider Monte Carlo (MC) and Markov chain Monte Carlo (MCMC). However, strictly interpolating the observations may entail expensive computations due to highly restrictive sample spaces. Having (constrained) GP emulators when data are actually noisy is also of interest. We introduce a noise term for the relaxation of the interpolation conditions, and we develop the corresponding approximation of GP emulators under linear inequality constraints. We show with various toy examples that the performance of MC and MCMC samplers improves when considering noisy observations. Finally, on a 5D monotonic example, we show that our framework still provides high effective sample rates with reasonable running times.

The Laplace approximation has been one of the workhorses of Bayesian inference. It often delivers good approximations in practice despite the fact that it does not strictly take into account where the volume of posterior density lies. Variational approaches avoid this issue by explicitly minimising the Kullback-Leibler divergence DKL between a postulated posterior and the true (unnormalised) logarithmic posterior. However, they rely on a closed form DKL in order to update the variational parameters. To address this, stochastic versions of variational inference have been devised that approximate the intractable DKL with a Monte Carlo average. This approximation allows calculating gradients with respect to the variational parameters. However, variational methods often postulate a factorised Gaussian approximating posterior. In doing so, they sacrifice a-posteriori correlations. In this work, we propose a method that combines the Laplace approximation with the variational approach. The advantages are that we maintain: applicability on non-conjugate models, posterior correlations and a reduced number of free variational parameters. Numerical experiments demonstrate improvement over the Laplace approximation and variational inference with factorised Gaussian posteriors.

We study the problem of offline learning in automated decision systems under the contextual bandits model. We are given logged historical data consisting of contexts, (randomized) actions, and (nonnegative) rewards. A common goal is to evaluate what would happen if different actions were taken in the same contexts, so as to optimize the action policies accordingly. The typical approach to this problem, inverse probability weighted estimation (IPWE) [Bottou et al., 2013], requires logged action probabilities, which may be missing in practice due to engineering complications. Even when available, small action probabilities cause large uncertainty in IPWE, rendering the corresponding results insignificant. To solve both problems, we show how one can use policy improvement (PIL) objectives, regularized by policy imitation (IML). We motivate and analyze PIL as an extension to Clipped-IPWE, by showing that both are lower-bound surrogates to the vanilla IPWE. We also formally connect IML to IPWE variance estimation [Swaminathan and Joachims 2015] and natural policy gradients. Without probability logging, our PIL-IML interpretations justify and improve, by reward-weighting, the state-of-art cross-entropy (CE) loss that predicts the action items among all action candidates available in the same contexts. With probability logging, our main theoretical contribution connects IML-underfitting to the existence of either confounding variables or model misspecification. We show the value and accuracy of our insights by simulations based on Simpson's paradox, standard UCI multiclass-to-bandit conversions and on the Criteo counterfactual analysis challenge dataset.

Statistical inference involves estimation of parameters of a model based on observations. Building on the recently proposed Equilibrium Expectation approach and Persistent Contrastive Divergence, we derive a simple and fast Markov chain Monte Carlo algorithm for maximum likelihood estimation (MLE) of parameters of exponential family distributions. The algorithm has good scaling properties and is suitable for Monte Carlo inference on large network data with billions of tie variables. The performance of the algorithm is demonstrated on Markov random fields, conditional random fields, exponential random graph models and Boltzmann machines.

Classical multidimensional scaling is an important tool for dimension reduction in many applications. Yet few theoretical results characterizing its statistical performance exist. In this paper, we provide a theoretical framework for analyzing the quality of embedded samples produced by classical multidimensional scaling. This lays down the foundation for various downstream statistical analysis. As an application, we study its performance in the setting of clustering noisy data. Our results provide scaling conditions on the sample size, ambient dimensionality, between-class distance and noise level under which classical multidimensional scaling followed by a clustering algorithm can recover the cluster labels of all samples with high probability. Numerical simulations confirm these scaling conditions are sharp in low, moderate, and high dimensional regimes. Applications to both human RNAseq data and natural language data lend strong support to the methodology and theory.