In image-based camera localization systems, information about the environment is usually stored in some representation, which can be referred to as a map. Conventionally, most maps are built upon hand-crafted features. Recently, neural networks have attracted attention as a data-driven map representation, and have shown promising results in visual localization. However, these neural network maps are generally hard to interpret by human. A readable map is not only accessible to humans, but also provides a way to be verified when the ground truth pose is unavailable. To tackle this problem, we propose Generative Map, a new framework for learning human-readable neural network maps, by combining a generative model with the Kalman filter, which also allows it to incorporate additional sensor information such as stereo visual odometry. For evaluation, we use real world images from the 7-Scenes and Oxford RobotCar datasets. We demonstrate that our Generative Map can be queried with a pose of interest from the test sequence to predict an image, which closely resembles the true scene. For localization, we show that Generative Map achieves comparable performance with current regression models. Moreover, our framework is trained completely from scratch, unlike regression models which rely on large ImageNet pretrained networks.

This paper uses a branching classifier mechanism in an unsupervised scenario, to enable it to self-organise data into unknown categories. A teaching phase is then able to help the classifier to learn the true category for each input row, using a reduced number of training steps. The pattern ensembles are learned in an unsupervsised manner that use a closest-distance clustering. This is done without knowing what the actual output category is and leads to each actual category having several clusters associated with it. One measure of success is then that each of these sub-clusters is coherent, which means that every data row in the cluster belongs to the same category. The total number of clusters is also important and a teaching phase can then teach the classifier what the correct actual category is. During this phase, any classifier can also learn or infer correct classifications from some other classifier's knowledge, thereby reducing the required number of presentations. As the information is added, cross-referencing between the two structures allows it to be used more widely. With this process, a unique structure can build up that would not be possible by either method separately. The lower level is a nested ensemble of patterns created by self-organisation. The upper level is a hierarchical tree, where each end node represents a single category only, so there is a transition from mixed ensemble masses to specific categories. The structure also has relations to brain-like modelling.

In predictive process analytics, current and historical process data in event logs are used to predict future. E.g., to predict the next activity or how long a process will still require to complete. Recurrent neural networks (RNN) and its subclasses have been demonstrated to be well suited for creating prediction models. Thus far, event attributes have not been fully utilized in these models. The biggest challenge in exploiting them in prediction models is the potentially large amount of event attributes and attribute values. We present a novel clustering technique which allows for trade-offs between prediction accuracy and the time needed for model training and prediction. As an additional finding, we also found that this clustering method combined with having raw event attribute values provides even better prediction accuracy at the cost of additional time required for training and prediction. We also built a highly configurable test framework that can be used to efficiently evaluate different prediction approaches and parameterizations.

Behavioral decision theories aim to explain human behavior. Can they help predict it? An open tournament for prediction of human choices in fundamental economic decision tasks is presented. The results suggest that integration of certain behavioral theories as features in machine learning systems provides the best predictions. Surprisingly, the most useful theories for prediction build on basic properties of human and animal learning and are very different from mainstream decision theories that focus on deviations from rational choice. Moreover, we find that theoretical features should be based not only on qualitative behavioral insights (e.g. loss aversion), but also on quantitative behavioral foresights generated by functional descriptive models (e.g. Prospect Theory). Our analysis prescribes a recipe for derivation of explainable, useful predictions of human decisions.

The goal of this paper is to study why stochastic gradient descent (SGD) is efficient for neural networks, and how neural net design affects SGD. In particular, we investigate how overparameterization -- an increase in the number of parameters beyond the number of training data -- affects the dynamics of SGD. We introduce a simple concept called gradient confusion. When confusion is high, stochastic gradients produced by different data samples may be negatively correlated, slowing down convergence. But when gradient confusion is low, we show that SGD has better convergence properties than predicted by classical theory. Using theoretical and experimental results, we study how overparameterization affects gradient confusion, and thus the convergence of SGD, on linear models and neural networks. We show that increasing the number of parameters of linear models or increasing the width of neural networks leads to lower gradient confusion, and thus faster and easier model training. We also show how overparameterization by increasing the depth of neural networks results in higher gradient confusion, making deeper models harder to train. Finally, we observe empirically that techniques like batch normalization and skip connections reduce gradient confusion, which helps reduce the training burden of deep networks.

Diverse applications - particularly in tumour subtyping - have demonstrated the importance of integrative clustering as a means to combine information from multiple high-dimensional omics datasets. Cluster-Of-Clusters Analysis (COCA) is a popular integrative clustering method that has been widely applied in the context of tumour subtyping. However, the properties of COCA have never been systematically explored, and the robustness of this approach to the inclusion of noisy datasets, or datasets that define conflicting clustering structures, is unclear. We rigorously benchmark COCA, and present Kernel Learning Integrative Clustering (KLIC) as an alternative strategy. KLIC frames the challenge of combining clustering structures as a multiple kernel learning problem, in which different datasets each provide a weighted contribution to the final clustering. This allows the contribution of noisy datasets to be down-weighted relative to more informative datasets. We show through extensive simulation studies that KLIC is more robust than COCA in a variety of situations. R code to run KLIC and COCA can be found at https://github.com/acabassi/klic

Blended courses that mix in-person instruction with online platforms are increasingly popular in secondary education. These tools record a rich amount of data on students' study habits and social interactions. Prior research has shown that these metrics are correlated with students' performance in face to face classes. However, predictive models for blended courses are still limited and have not yet succeeded at early prediction or cross-class predictions even for repeated offerings of the same course. In this work, we use data from two offerings of two different undergraduate courses to train and evaluate predictive models on student performance based upon persistent student characteristics including study habits and social interactions. We analyze the performance of these models on the same offering, on different offerings of the same course, and across courses to see how well they generalize. We also evaluate the models on different segments of the courses to determine how early reliable predictions can be made. This work tells us in part how much data is required to make robust predictions and how cross-class data may be used, or not, to boost model performance. The results of this study will help us better understand how similar the study habits, social activities, and the teamwork styles are across semesters for students in each performance category. These trained models also provide an avenue to improve our existing support platforms to better support struggling students early in the semester with the goal of providing timely intervention.

The increased usage of solar energy places additional importance on forecasts of solar radiation. Solar panel power production is primarily driven by the amount of solar radiation and it is therefore important to have accurate forecasts of solar radiation. Accurate forecasts that also give information on the forecast uncertainties can help users of solar energy to make better solar radiation based decisions related to the stability of the electrical grid. To achieve this, we apply statistical post-processing techniques that determine relationships between observations of global radiation (made within the KNMI network of automatic weather stations in the Netherlands) and forecasts of various meteorological variables from the numerical weather prediction (NWP) model HARMONIE-AROME (HA) and the atmospheric composition model CAMS. Those relationships are used to produce probabilistic forecasts of global radiation. We compare 7 different statistical post-processing methods, consisting of two parametric and five non-parametric methods. We find that all methods are able to generate probabilistic forecasts that improve the raw global radiation forecast from HA according to the root mean squared error (on the median) and the potential economic value. Additionally, we show how important the predictors are in the different regression methods. We also compare the regression methods using various probabilistic scoring metrics, namely the continuous ranked probability skill score, the Brier skill score and reliability diagrams. We find that quantile regression and generalized random forests generally perform best. In (near) clear sky conditions the non-parametric methods have more skill than the parametric ones.

This paper considers a new family of variational distributions motivated by Sklar's theorem. This family is based on new copula-like densities on the hypercube with non-uniform marginals which can be sampled efficiently, i.e. with a complexity linear in the dimension of state space. Then, the proposed variational densities that we suggest can be seen as arising from these copula-like densities used as base distributions on the hypercube with Gaussian quantile functions and sparse rotation matrices as normalizing flows. The latter correspond to a rotation of the marginals with complexity $\mathcal{O}(d \log d)$. We provide some empirical evidence that such a variational family can also approximate non-Gaussian posteriors and can be beneficial compared to Gaussian approximations. Our method performs largely comparably to state-of-the-art variational approximations on standard regression and classification benchmarks for Bayesian Neural Networks.

The ability to evaluate the performance of a computational model is a vital requirement for driving algorithm research. This is often particularly difficult for generative models such as generative adversarial networks (GAN) that model a data manifold only specified indirectly by a finite set of training examples. In the common case of image data, the samples live in a high-dimensional embedding space with little structure to help assessing either the overall quality of samples or the coverage of the underlying manifold. We present an evaluation metric with the ability to separately and reliably measure both of these aspects in image generation tasks by forming explicit non-parametric representations of the manifolds of real and generated data. We demonstrate the effectiveness of our metric in StyleGAN and BigGAN by providing several illustrative examples where existing metrics yield uninformative or contradictory results. Furthermore, we analyze multiple design variants of StyleGAN to better understand the relationships between the model architecture, training methods, and the properties of the resulting sample distribution. In the process, we identify new variants that improve the state-of-the-art. We also perform the first principled analysis of truncation methods and identify an improved method. Finally, we extend our metric to estimate the perceptual quality of individual samples, and use this to study latent space interpolations.