In many hierarchical inverse problems, not only do we want to estimate high- or infinite-dimensional model parameters in the parameter-to-observable maps, but we also have to estimate hyperparameters that represent critical assumptions in the statistical and mathematical modeling processes. As a joint effect of high-dimensionality, nonlinear dependence, and non-concave structures in the joint posterior posterior distribution over model parameters and hyperparameters, solving inverse problems in the hierarchical Bayesian setting poses a significant computational challenge. In this work, we aim to develop scalable optimization-based Markov chain Monte Carlo (MCMC) methods for solving hierarchical Bayesian inverse problems with nonlinear parameter-to-observable maps and a broader class of hyperparameters. Our algorithmic development is based on the recently developed scalable randomize-then-optimize (RTO) method [4] for exploring the high- or infinite-dimensional model parameter space. By using RTO either as a proposal distribution in a Metropolis-within-Gibbs update or as a biasing distribution in the pseudo-marginal MCMC [2], we are able to design efficient sampling tools for hierarchical Bayesian inversion. In particular, the integration of RTO and the pseudo-marginal MCMC has sampling performance robust to model parameter dimensions. We also extend our methods to nonlinear inverse problems with Poisson-distributed measurements. Numerical examples in PDE-constrained inverse problems and positron emission tomography (PET) are used to demonstrate the performance of our methods.

Uncertainty estimation and ensembling methods go hand-in-hand. Uncertainty estimation is one of the main benchmarks for assessment of ensembling performance. At the same time, deep learning ensembles have provided state-of-the-art results in uncertainty estimation. In this work, we focus on in-domain uncertainty for image classification. We explore the standards for its quantification and point out pitfalls of existing metrics. Avoiding these pitfalls, we perform a broad study of different ensembling techniques. To provide more insight in this study, we introduce the deep ensemble equivalent score (DEE) and show that many sophisticated ensembling techniques are equivalent to an ensemble of only few independently trained networks in terms of test performance.

Problem: The problem we intend to solve is modelled as a binary classification problem. We intend to find the relation in the words and the context in which the words appear within the text and how it could be used to classify texts as real (negative cases) or fake (positive). High-level description: Many news sources contain false information and are therefore "fake news." Because there is a lot of "fake news" articles and fabricated, misleading information on the web, we would like to determine which texts are legitimate (real) and which are illegitimate (fake). To solve this as a binary classification problem, we investigate the effectiveness of different Natural Language Processing models which are used to convert character based texts into numeric representations such as TFIDF, CountVectorizer and Word2Vec models and find out which model is able to preserve most of the contextual information about the text used in a fake news data set and how helpful and effective it is in detecting whether the text is a fake news or not.

Current reading comprehension models generalise well to in-distribution test sets, yet perform poorly on adversarially selected inputs. Most prior work on adversarial inputs studies oversensitivity: semantically invariant text perturbations that cause a model's prediction to change when it should not. In this work we focus on the complementary problem: excessive prediction undersensitivity, where input text is meaningfully changed but the model's prediction does not, even though it should. We formulate a noisy adversarial attack which searches among semantic variations of the question for which a model erroneously predicts the same answer, and with even higher probability. Despite comprising unanswerable questions, both SQuAD2.0 and NewsQA models are vulnerable to this attack. This indicates that although accurate, models tend to rely on spurious patterns and do not fully consider the information specified in a question. We experiment with data augmentation and adversarial training as defences, and find that both substantially decrease vulnerability to attacks on held out data, as well as held out attack spaces. Addressing undersensitivity also improves results on AddSent and AddOneSent, and models furthermore generalise better when facing train/evaluation distribution mismatch: they are less prone to overly rely on predictive cues present only in the training set, and outperform a conventional model by as much as 10.9% F1.

Global information is essential for dense prediction problems, whose goal is to compute a discrete or continuous label for each pixel in the images. Traditional convolutional layers in neural networks, originally designed for image classification, are restrictive in these problems since their receptive fields are limited by the filter size. In this work, we propose autoregressive moving-average (ARMA) layer, a novel module in neural networks to allow explicit dependencies of output neurons, which significantly expands the receptive field with minimal extra parameters. We show experimentally that the effective receptive field of neural networks with ARMA layers expands as autoregressive coefficients become larger. In addition, we demonstrate that neural networks with ARMA layers substantially improve the performance of challenging pixel-level video prediction tasks as our model enlarges the effective receptive field.

Automatic speaker verification systems are vulnerable to audio replay attacks which bypass security by replaying recordings of authorized speakers. Replay attack detection (RA) detection systems built upon Residual Neural Networks (ResNet)s have yielded astonishing results on the public benchmark ASVspoof 2019 Physical Access challenge. With most teams using fine-tuned feature extraction pipelines and model architectures, the generalizability of such systems remains questionable though. In this work, we analyse the effect of discriminative feature learning in a multi-task learning (MTL) setting can have on the generalizability and discriminability of RA detection systems. We use a popular ResNet architecture optimized by the cross-entropy criterion as our baseline and compare it to the same architecture optimized by MTL using Siamese Neural Networks (SNN). It can be shown that SNN outperform the baseline by relative 26.8 % Equal Error Rate (EER). We further enhance the model's architecture and demonstrate that SNN with additional reconstruction loss yield another significant improvement of relative 13.8 % EER.

The universal approximation theorem, in one of its most general versions, says that if we consider only continuous activation functions $\sigma$, then a standard feedforward neural network with one hidden layer is able to approximate any continuous multivariate function $f$ to any given approximation threshold $\varepsilon$, if and only if $\sigma$ is non-polynomial. In this paper, we give a direct algebraic proof of the theorem. Furthermore we shall explicitly quantify the number of hidden units required for approximation. Specifically, if $X\subseteq \mathbb{R}^n$ is compact, then a neural network with $n$ input units, $m$ output units, and a single hidden layer with $\binom{n+d}{d}$ hidden units (independent of $m$ and $\varepsilon$), can uniformly approximate any polynomial function $f:X \to \mathbb{R}^m$ whose total degree is at most $d$ for each of its $m$ coordinate functions. In the general case that $f$ is any continuous function, we show there exists some $N\in \mathcal{O}(\varepsilon^{-n})$ (independent of $m$), such that $N$ hidden units would suffice to approximate $f$. We also show that this uniform approximation property (UAP) still holds even under seemingly strong conditions imposed on the weights. We highlight several consequences: (i) For any $\delta > 0$, the UAP still holds if we restrict all non-bias weights $w$ in the last layer to satisfy $|w| < \delta$. (ii) There exists some $\lambda>0$ (depending only on $f$ and $\sigma$), such that the UAP still holds if we restrict all non-bias weights $w$ in the first layer to satisfy $|w|>\lambda$. (iii) If the non-bias weights in the first layer are \emph{fixed} and randomly chosen from a suitable range, then the UAP holds with probability $1$.

Robust loss minimization is an important strategy for handling robust learning issue on noisy labels. Current robust loss functions, however, inevitably involve hyperparameter(s) to be tuned, manually or heuristically through cross validation, which makes them fairly hard to be generally applied in practice. Besides, the non-convexity brought by the loss as well as the complicated network architecture makes it easily trapped into an unexpected solution with poor generalization capability. To address above issues, we propose a meta-learning method capable of adaptively learning hyperparameter in robust loss functions. Specifically, through mutual amelioration between robust loss hyperparameter and network parameters in our method, both of them can be simultaneously finely learned and coordinated to attain solutions with good generalization capability. Four kinds of SOTA robust loss functions are attempted to be integrated into our algorithm, and comprehensive experiments substantiate the general availability and effectiveness of the proposed method in both its accuracy and generalization performance, as compared with conventional hyperparameter tuning strategy, even with carefully tuned hyperparameters.

This work considers two distinct settings: imitation learning and goal-conditioned reinforcement learning. In either case, effective solutions require the agent to reliably reach a specified state (a goal), or set of states (a demonstration). Drawing a connection between probabilistic long-term dynamics and the desired value function, this work introduces an approach which utilizes recent advances in density estimation to effectively learn to reach a given state. As our first contribution, we use this approach for goal-conditioned reinforcement learning and show that it is both efficient and does not suffer from hindsight bias in stochastic domains. As our second contribution, we extend the approach to imitation learning and show that it achieves state-of-the art demonstration sample-efficiency on standard benchmark tasks.

We present an efficient coreset construction algorithm for large-scale Support Vector Machine (SVM) training in Big Data and streaming applications. A coreset is a small, representative subset of the original data points such that a models trained on the coreset are provably competitive with those trained on the original data set. Since the size of the coreset is generally much smaller than the original set, our preprocess-then-train scheme has potential to lead to significant speedups when training SVM models. We prove lower and upper bounds on the size of the coreset required to obtain small data summaries for the SVM problem. As a corollary, we show that our algorithm can be used to extend the applicability of any off-the-shelf SVM solver to streaming, distributed, and dynamic data settings. We evaluate the performance of our algorithm on real-world and synthetic data sets. Our experimental results reaffirm the favorable theoretical properties of our algorithm and demonstrate its practical effectiveness in accelerating SVM training.