Approximate inference in complex probabilistic models such as deep Gaussian processes requires the optimisation of doubly stochastic objective functions. These objectives incorporate randomness both from mini-batch subsampling of the data and from Monte Carlo estimation of expectations. If the gradient variance is high, the stochastic optimisation problem becomes difficult with a slow rate of convergence. Control variates can be used to reduce the variance, but past approaches do not take into account how mini-batch stochasticity affects sampling stochasticity, resulting in sub-optimal variance reduction. We propose a new approach in which we use a recognition network to cheaply approximate the optimal control variate for each mini-batch, with no additional model gradient computations. We illustrate the properties of this proposal and test its performance on logistic regression and deep Gaussian processes.

Grade prediction for future courses not yet taken by students is important as it can help them and their advisers during the process of course selection as well as for designing personalized degree plans and modifying them based on their performance. One of the successful approaches for accurately predicting a student's grades in future courses is Cumulative Knowledge-based Regression Models (CKRM). CKRM learns shallow linear models that predict a student's grades as the similarity between his/her knowledge state and the target course. However, prior courses taken by a student can have \black{different contributions when estimating a student's knowledge state and towards each target course, which} cannot be captured by linear models. Moreover, CKRM and other grade prediction methods ignore the effect of concurrently-taken courses on a student's performance in a target course. In this paper, we propose context-aware non-linear and neural attentive models that can potentially better estimate a student's knowledge state from his/her prior course information, as well as model the interactions between a target course and concurrent courses. Compared to the competing methods, our experiments on a large real-world dataset consisting of more than $1.5$M grades show the effectiveness of the proposed models in accurately predicting students' grades. Moreover, the attention weights learned by the neural attentive model can be helpful in better designing their degree plans.

The online analysis of multi-way data stored in a tensor $\mathcal{X} \in \mathbb{R} ^{I_1 \times \dots \times I_N} $ has become an essential tool for capturing the underlying structures and extracting the sensitive features which can be used to learn a predictive model. However, data distributions often evolve with time and a current predictive model may not be sufficiently representative in the future. Therefore, incrementally updating the tensor-based features and model coefficients are required in such situations. A new efficient tensor-based feature extraction, named NeSGD, is proposed for online $CANDECOMP/PARAFAC$ (CP) decomposition. According to the new features obtained from the resultant matrices of NeSGD, a new criteria is triggered for the updated process of the online predictive model. Experimental evaluation in the field of structural health monitoring using laboratory-based and real-life structural datasets show that our methods provide more accurate results compared with existing online tensor analysis and model learning. The results showed that the proposed methods significantly improved the classification error rates, were able to assimilate the changes in the positive data distribution over time, and maintained a high predictive accuracy in all case studies.

The abundance of modern health data provides many opportunities for the use of machine learning techniques to build better statistical models to improve clinical decision making. Predicting time-to-event distributions, also known as survival analysis, plays a key role in many clinical applications. We introduce a variational time-to-event prediction model, named Variational Survival Inference (VSI), which builds upon recent advances in distribution learning techniques and deep neural networks. VSI addresses the challenges of non-parametric distribution estimation by ($i$) relaxing the restrictive modeling assumptions made in classical models, and ($ii$) efficiently handling the censored observations, {\it i.e.}, events that occur outside the observation window, all within the variational framework. To validate the effectiveness of our approach, an extensive set of experiments on both synthetic and real-world datasets is carried out, showing improved performance relative to competing solutions.

Stochastic gradient descent (SGD) has been widely studied in the literature from different angles, and is commonly employed for solving many big data machine learning problems. However, the averaging technique, which combines all iterative solutions into a single solution, is still under-explored. While some increasingly weighted averaging schemes have been considered in the literature, existing works are mostly restricted to strongly convex objective functions and the convergence of optimization error. It remains unclear how these averaging schemes affect the convergence of {\it both optimization error and generalization error} (two equally important components of testing error) for {\bf non-strongly convex objectives, including non-convex problems}. In this paper, we {\it fill the gap} by comprehensively analyzing the increasingly weighted averaging on convex, strongly convex and non-convex objective functions in terms of both optimization error and generalization error. In particular, we analyze a family of increasingly weighted averaging, where the weight for the solution at iteration $t$ is proportional to $t^{\alpha}$ ($\alpha > 0$). We show how $\alpha$ affects the optimization error and the generalization error, and exhibit the trade-off caused by $\alpha$. Experiments have demonstrated this trade-off and the effectiveness of polynomially increased weighted averaging compared with other averaging schemes for a wide range of problems including deep learning.

Manifold regularization is a technique that penalizes the complexity of learned functions over the intrinsic geometry of input data. We develop a connection to learning functions which are "locally stable", and propose new regularization terms for training deep neural networks that are stable against a class of local perturbations. These regularizers enable us to train a network to state-of-the-art robust accuracy of 70% on CIFAR-10 against a PGD adversary using $\ell_\infty$ perturbations of size $\epsilon = 8/255$. Furthermore, our techniques do not rely on the construction of any adversarial examples, thus running orders of magnitude faster than standard algorithms for adversarial training.

This paper introduces a novel perspective about error in machine learning and proposes inverse feature learning (IFL) as a representation learning approach that learns a set of high-level features based on the representation of error for classification or clustering purposes. The proposed perspective about error representation is fundamentally different from current learning methods, where in classification approaches they interpret the error as a function of the differences between the true labels and the predicted ones or in clustering approaches, in which the clustering objective functions such as compactness are used. Inverse feature learning method operates based on a deep clustering approach to obtain a qualitative form of the representation of error as features. The performance of the proposed IFL method is evaluated by applying the learned features along with the original features, or just using the learned features in different classification and clustering techniques for several data sets. The experimental results show that the proposed method leads to promising results in classification and especially in clustering. In classification, the proposed features along with the primary features improve the results of most of the classification methods on several popular data sets. In clustering, the performance of different clustering methods is considerably improved on different data sets. There are interesting results that show some few features of the representation of error capture highly informative aspects of primary features. We hope this paper helps to utilize the error representation learning in different feature learning domains.

Domain-specific image collections present potential value in various areas of science and business but are often not curated nor have any way to readily extract relevant content. To employ contemporary supervised image analysis methods on such image data, they must first be cleaned and organized, and then manually labeled for the nomenclature employed in the specific domain, which is a time consuming and expensive endeavor. To address this issue, we designed and implemented the Plud system. Plud provides an iterative semi-supervised workflow to minimize the effort spent by an expert and handles realistic large collections of images. We believe it can support labeling datasets regardless of their size and type. Plud is an iterative sequence of unsupervised clustering, human assistance, and supervised classification. With each iteration 1) the labeled dataset grows, 2) the generality of the classification method and its accuracy increases, and 3) manual effort is reduced. We evaluated the effectiveness of our system, by applying it on over a million images documenting human decomposition. In our experiment comparing manual labeling with labeling conducted with the support of Plud, we found that it reduces the time needed to label data and produces highly accurate models for this new domain.

Right to be forgotten, also known as the right to erasure, is the right of individuals to have their data erased from an entity storing it. The General Data Protection Regulation in the European Union legally solidified the status of this long held notion. As a consequence, there is a growing need for the development of mechanisms whereby users can verify if service providers comply with their deletion requests. In this work, we take the first step in proposing a formal framework to study the design of such verification mechanisms for data deletion requests -- also known as machine unlearning -- in the context of systems that provide machine learning as a service. We propose a backdoor-based verification mechanism and demonstrate its effectiveness in certifying data deletion with high confidence using the above framework. Our mechanism makes a novel use of backdoor attacks in ML as a basis for quantitatively inferring machine unlearning. In our mechanism, each user poisons part of its training data by injecting a user-specific backdoor trigger associated with a user-specific target label. The prediction of target labels on test samples with the backdoor trigger is then used as an indication of the user's data being used to train the ML model. We formalize the verification process as a hypothesis testing problem, and provide theoretical guarantees on the statistical power of the hypothesis test. We experimentally demonstrate that our approach has minimal effect on the machine learning service but provides high confidence verification of unlearning. We show that with a $30\%$ poison ratio and merely $20$ test queries, our verification mechanism has both false positive and false negative ratios below $10^{-5}$. Furthermore, we also show the effectiveness of our approach by testing it against an adaptive adversary that uses a state-of-the-art backdoor defense method.

Current data-driven traffic prediction models are usually trained with large datasets, e.g. several months of speeds and flows. Such models provide very good fit for ordinary road conditions, but often fail just when they are most needed: when traffic suffers a sudden and significant disruption, such as a road incident. In this work, we describe QTIP: a simulation-based framework for quasi-instantaneous adaptation of prediction models upon traffic disruption. In a nutshell, QTIP performs real-time simulations of the affected road for multiple scenarios, analyzes the results, and suggests a change to an ordinary prediction model accordingly. QTIP constructs the simulated scenarios per properties of the incident, as conveyed by immediate distress signals from affected vehicles. Such real-time signals are provided by In-Vehicle Monitor Systems, which are becoming increasingly prevalent world-wide. We experiment QTIP in a case study of a Danish motorway, and the results show that QTIP can improve traffic prediction in the first critical minutes of road incidents.