This paper introduces the subgraph nomination inference task, in which example subgraphs of interest are used to query a network for similarly interesting subgraphs. This type of problem appears time and again in real world problems connected to, for example, user recommendation systems and structural retrieval tasks in social and biological/connectomic networks. We formally define the subgraph nomination framework with an emphasis on the notion of a user-in-the-loop in the subgraph nomination pipeline. In this setting, a user can provide additional post-nomination light supervision that can be incorporated into the retrieval task. After introducing and formalizing the retrieval task, we examine the nuanced effect that user-supervision can have on performance, both analytically and across real and simulated data examples.

After challenging the validity of these assumptions in real-world applications, we propose ways to move forward when they are violated. First, we show that group fairness criteria purely based on statistical properties of observed data are fundamentally limited. Revisiting this limitation from a causal viewpoint we develop a more versatile conceptual framework, causal fairness criteria, and first algorithms to achieve them. We also provide tools to analyze how sensitive a believed-to-be causally fair algorithm is to misspecifications of the causal graph. Second, we overcome the assumption that sensitive data is readily available in practice. To this end we devise protocols based on secure multi-party computation to train, validate, and contest fair decision algorithms without requiring users to disclose their sensitive data or decision makers to disclose their models. Finally, we also accommodate the fact that outcome labels are often only observed when a certain decision has been made. We suggest a paradigm shift away from training predictive models towards directly learning decisions to relax the traditional assumption that labels can always be recorded. The main contribution of this thesis is the development of theoretically substantiated and practically feasible methods to move research on fair machine learning closer to real-world applications.

We consider the problem of predicting the covariance of a zero mean Gaussian vector, based on another feature vector. We describe a covariance predictor that has the form of a generalized linear model, i.e., an affine function of the features followed by an inverse link function that maps vectors to symmetric positive definite matrices. The log-likelihood is a concave function of the predictor parameters, so fitting the predictor involves convex optimization. Such predictors can be combined with others, or recursively applied to improve performance.

We consider the problem of forecasting multiple values of the future of a vector time series, using some past values. This problem, and related ones such as one-step-ahead prediction, have a very long history, and there are a number of well-known methods for it, including vector auto-regressive models, state-space methods, multi-task regression, and others. Our focus is on low rank forecasters, which break forecasting up into two steps: estimating a vector that can be interpreted as a latent state, given the past, and then estimating the future values of the time series, given the latent state estimate. We introduce the concept of forecast consistency, which means that the estimates of the same value made at different times are consistent. We formulate the forecasting problem in general form, and focus on linear forecasters, for which we propose a formulation that can be solved via convex optimization. We describe a number of extensions and variations, including nonlinear forecasters, data weighting, the inclusion of auxiliary data, and additional objective terms. We illustrate our methods with several examples.

With the goal of designing novel inhibitors for SARS-CoV-1 and SARS-CoV-2, we propose the general molecule optimization framework, Molecular Neural Assay Search (MONAS), consisting of three components: a property predictor which identifies molecules with specific desirable properties, an energy model which approximates the statistical similarity of a given molecule to known training molecules, and a molecule search method. In this work, these components are instantiated with graph neural networks (GNNs), Deep Energy Estimator Networks (DEEN) and Monte Carlo tree search (MCTS), respectively. This implementation is used to identify 120K molecules (out of 40-million explored) which the GNN determined to be likely SARS-CoV-1 inhibitors, and, at the same time, are statistically close to the dataset used to train the GNN.

For infinitesimal learning rates, stochastic gradient descent (SGD) follows the path of gradient flow on the full batch loss function. However moderately large learning rates can achieve higher test accuracies, and this generalization benefit is not explained by convergence bounds, since the learning rate which maximizes test accuracy is often larger than the learning rate which minimizes training loss. To interpret this phenomenon we prove that for SGD with random shuffling, the mean SGD iterate also stays close to the path of gradient flow if the learning rate is small and finite, but on a modified loss. This modified loss is composed of the original loss function and an implicit regularizer, which penalizes the norms of the minibatch gradients. Under mild assumptions, when the batch size is small the scale of the implicit regularization term is proportional to the ratio of the learning rate to the batch size. We verify empirically that explicitly including the implicit regularizer in the loss can enhance the test accuracy when the learning rate is small. In the limit of vanishing learning rates, stochastic gradient descent with minibatch gradients (SGD) follows the path of gradient flow on the full batch loss function (Yaida, 2019). However in deep networks, SGD often achieves higher test accuracies when the learning rate is moderately large (LeCun et al., 2012; Keskar et al., 2017). This generalization benefit is not explained by convergence rate bounds (Ma et al., 2018; Zhang et al., 2019), because it arises even for large compute budgets for which smaller learning rates often achieve lower training losses (Smith et al., 2020).

Theoretical results show that Bayesian methods can achieve lower bounds on regret for online logistic regression. In practice, however, such techniques may not be feasible especially for very large feature sets. Various approximations that, for huge sparse feature sets, diminish the theoretical advantages, must be used. Often, they apply stochastic gradient methods with hyper-parameters that must be tuned on some surrogate loss, defeating theoretical advantages of Bayesian methods. The surrogate loss, defined to approximate the mixture, requires techniques as Monte Carlo sampling, increasing computations per example. We propose low complexity analytical approximations for sparse online logistic and probit regressions. Unlike variational inference and other methods, our methods use analytical closed forms, substantially lowering computations. Unlike dense solutions, as Gaussian Mixtures, our methods allow for sparse problems with huge feature sets without increasing complexity. With the analytical closed forms, there is also no need for applying stochastic gradient methods on surrogate losses, and for tuning and balancing learning and regularization hyper-parameters. Empirical results top the performance of the more computationally involved methods. Like such methods, our methods still reveal per feature and per example uncertainty measures.

This paper introduces the unsupervised learning problem of playable video generation (PVG). In PVG, we aim at allowing a user to control the generated video by selecting a discrete action at every time step as when playing a video game. The difficulty of the task lies both in learning semantically consistent actions and in generating realistic videos conditioned on the user input. We propose a novel framework for PVG that is trained in a self-supervised manner on a large dataset of unlabelled videos. We employ an encoder-decoder architecture where the predicted action labels act as bottleneck. The network is constrained to learn a rich action space using, as main driving loss, a reconstruction loss on the generated video. We demonstrate the effectiveness of the proposed approach on several datasets with wide environment variety. Further details, code and examples are available on our project page willi-menapace.github.io/playable-video-generation-website.

Condition-based maintenance (CBM) strategies exploit machine learning models to assess the health status of systems based on the collected data from the physical environment, while machine learning models are vulnerable to adversarial attacks. A malicious adversary can manipulate the collected data to deceive the machine learning model and affect the CBM system's performance. Adversarial machine learning techniques introduced in the computer vision domain can be used to make stealthy attacks on CBM systems by adding perturbation to data to confuse trained models. The stealthy nature causes difficulty and delay in detection of the attacks. In this paper, adversarial machine learning in the domain of CBM is introduced. A case study shows how adversarial machine learning can be used to attack CBM capabilities. Adversarial samples are crafted using the Fast Gradient Sign method, and the performance of a CBM system under attack is investigated. The obtained results reveal that CBM systems are vulnerable to adversarial machine learning attacks and defense strategies need to be considered.

The most common supervised task in machine learning is to learn a single-task, single-output prediction model. However, such a setting can be ill-adapted to some problems and applications. On the one hand, producing a single output can be undesirable when data is scarce and when producing reliable, possibly set-valued predictions is important (for instance in the medical domain where examples are very hard to collect for specific targets, and where predictions are used for critical decisions). Such an issue can be solved by using conformal prediction approaches [1]. It was initially proposed as a transductive online learning approach to provide set predictions (in the classification case) or interval predictions (in the case of regression) with a statistical guarantee depending on the probability of error tolerated by the user, but was then extended to handle inductive processes [2]. On the other hand, there are many situations where there are multiple, possibly correlated output variables to predict at once, and it is then natural to try to leverage such correlations to improve predictions. Such learning tasks are commonly called Multi-task in the literature [3]. Most research work on conformal prediction for multi-task learning focuses on the problem of multi-label prediction [4, 5], where each task is a binary classification one. Conformal prediction for multi-target regression has been less explored, with only a few studies dealing with it: Kuleshov et al. [6] provide a theoretical framework to use conformal predictors within manifold (e.g., to provide a mono-dimensional embedding of the multi-variate output), while Neeven and Smirnov [7] use a straightforward multi-target extension of a conformal single-output k-nearest neighbor regressor [8] to provide weather forecasts.