Recent studies in the field of Machine Translation (MT) and Natural Language Processing (NLP) have shown that existing models amplify biases observed in the training data. The amplification of biases in language technology has mainly been examined with respect to specific phenomena, such as gender bias. In this work, we go beyond the study of gender in MT and investigate how bias amplification might affect language in a broader sense. We hypothesize that the 'algorithmic bias', i.e. an exacerbation of frequently observed patterns in combination with a loss of less frequent ones, not only exacerbates societal biases present in current datasets but could also lead to an artificially impoverished language: 'machine translationese'. We assess the linguistic richness (on a lexical and morphological level) of translations created by different data-driven MT paradigms - phrase-based statistical (PB-SMT) and neural MT (NMT). Our experiments show that there is a loss of lexical and morphological richness in the translations produced by all investigated MT paradigms for two language pairs (EN<=>FR and EN<=>ES).

In formal concept analysis, it is well-known that the number of formal concepts can be exponential in the worst case. To analyze the average case, we introduce a probabilistic model for random formal contexts and prove that the average number of formal concepts has a superpolynomial asymptotic lower bound.

Counterfactual instances offer human-interpretable insight into the local behaviour of machine learning models. We propose a general framework to generate sparse, in-distribution counterfactual model explanations which match a desired target prediction with a conditional generative model, allowing batches of counterfactual instances to be generated with a single forward pass. The method is flexible with respect to the type of generative model used as well as the task of the underlying predictive model. This allows straightforward application of the framework to different modalities such as images, time series or tabular data as well as generative model paradigms such as GANs or autoencoders and predictive tasks like classification or regression. We illustrate the effectiveness of our method on image (CelebA), time series (ECG) and mixed-type tabular (Adult Census) data.

How many neurons are needed to approximate a target probability distribution using a neural network with a given input distribution and approximation error? This paper examines this question for the case when the input distribution is uniform, and the target distribution belongs to the class of histogram distributions. We obtain a new upper bound on the number of required neurons, which is strictly better than previously existing upper bounds. The key ingredient in this improvement is an efficient construction of the neural nets representing piecewise linear functions. We also obtain a lower bound on the minimum number of neurons needed to approximate the histogram distributions.

A key reason Forecasting - predicting future values of time series, is a key for recent success of deep learning for forecasting is multitask component in many industries (Fildes et al. 2008). Applications univariate forecasting - sharing deep learning model parameters include forecasting supply chain and airline demand across all series, possibly with some series-specific (Fildes et al. 2008; Seeger, Salinas, and Flunkert 2016), financial scaling factors or parametric model components (Salinas, prices (Kim 2003), and energy, traffic or weather Flunkert, and Gasthaus 2019; Smyl 2020; Bandara, Bergmeir, patterns (Chatfield 2000). Forecasts are often required for and Hewamalage 2020; Li et al. 2019; Wen et al. 2017; Rangapuram large numbers of related time series, i.e., multivariate time series et al. 2018; Chen et al. 2018). E.g., the winner of forecasting, as opposed to univariate (single time series) the M4 forecasting competition (Makridakis, Spiliotis, and forecasting. For example, retailers may require sales/demand Assimakopoulos 2020) was a hybrid ES-RNN model (Smyl forecasts for millions of different products at thousands of 2020), in which a single shared univariate RNN model is used different locations - amounting to billions of sales time series.

For a d-ary Boolean function Φ: {0, 1}d → {0, 1} and an assignment to its variables x = (x1, x2, . . . , xd) we consider the problem of finding those subsets of the variables that are sufficient to determine the function value with a given probability δ. This is motivated by the task of interpreting predictions of binary classifiers described as Boolean circuits, which can be seen as special cases of neural networks. We show that the problem of deciding whether such subsets of relevant variables of limited size k ≤ d exist is complete for the complexity class NPPP and thus, generally, unfeasible to solve. We then introduce a variant, in which it suffices to check whether a subset determines the function value with probability at least δ or at most δ − γ for 0 < γ < δ. This promise of a probability gap reduces the complexity to the class NPBPP. Finally, we show that finding the minimal set of relevant variables cannot be reasonably approximated, i.e. with an approximation factor d1−α for α > 0, by a polynomial time algorithm unless P = NP. This holds even with the promise of a probability gap.

Amortised inference enables scalable learning of sequential latent-variable models (LVMs) with the evidence lower bound (ELBO). In this setting, variational posteriors are often only partially conditioned. While the true posteriors depend, e.g., on the entire sequence of observations, approximate posteriors are only informed by past observations. This mimics the Bayesian filter -- a mixture of smoothing posteriors. Yet, we show that the ELBO objective forces partially-conditioned amortised posteriors to approximate products of smoothing posteriors instead. Consequently, the learned generative model is compromised. We demonstrate these theoretical findings in three scenarios: traffic flow, handwritten digits, and aerial vehicle dynamics. Using fully-conditioned approximate posteriors, performance improves in terms of generative modelling and multi-step prediction.

Deep-predictive-coding networks (DPCNs) are hierarchical, generative models that rely on feed-forward and feed-back connections to modulate latent feature representations of stimuli in a dynamic and context-sensitive manner. A crucial element of DPCNs is a forward-backward inference procedure to uncover sparse states of a dynamic model, which are used for invariant feature extraction. However, this inference and the corresponding backwards network parameter updating are major computational bottlenecks. They severely limit the network depths that can be reasonably implemented and easily trained. We therefore propose a optimization strategy, with better empirical and theoretical convergence, based on accelerated proximal gradients. We demonstrate that the ability to construct deeper DPCNs leads to receptive fields that capture well the entire notions of objects on which the networks are trained. This improves the feature representations. It yields completely unsupervised classifiers that surpass convolutional and convolutional-recurrent autoencoders and are on par with convolutional networks trained in a supervised manner. This is despite the DPCNs having orders of magnitude fewer parameters.

As an important problem of causal inference, we discuss the estimation of treatment effects under the existence of unobserved confounding. By representing the confounder as a latent variable, we propose Counterfactual VAE, a new variant of variational autoencoder, based on recent advances in identifiability of representation learning. Combining the identifiability and classical identification results of causal inference, under mild assumptions on the generative model and with small noise on the outcome, we theoretically show that the confounder is identifiable up to an affine transformation and then the treatment effects can be identified. Experiments on synthetic and semi-synthetic datasets demonstrate that our method matches the state-of-the-art, even under settings violating our formal assumptions.

We propose a saliency-based, multi-target detection and segmentation framework for multi-aspect, semi-coherent imagery formed from circular-scan, synthetic-aperture sonar (CSAS). Our framework relies on a multi-branch, convolutional encoder-decoder network (MB-CEDN). The encoder portion extracts features from one or more CSAS images of the targets. These features are then split off and fed into multiple decoders that perform pixel-level classification on the extracted features to roughly mask the target in an unsupervised-trained manner and detect foreground and background pixels in a supervised-trained manner. Each of these target-detection estimates provide different perspectives as to what constitute a target. These opinions are cascaded into a deep-parsing network to model contextual and spatial constraints that help isolate targets better than either solution estimate alone. We evaluate our framework using real-world CSAS data with five broad target classes. Since we are the first to consider both CSAS target detection and segmentation, we adapt existing image and video-processing network topologies from the literature for comparative purposes. We show that our framework outperforms supervised deep networks. It greatly outperforms state-of-the-art unsupervised approaches for diverse target and seafloor types.