This paper provides results of evaluating some text summarisation techniques for the purpose of producing call summaries for contact centre solutions. We specifically focus on extractive summarisation methods, as they do not require any labelled data and are fairly quick and easy to implement for production use. We experimentally compare several such methods by using them to produce summaries of calls, and evaluating these summaries objectively (using ROUGE-L) and subjectively (by aggregating the judgements of several annotators). We found that TopicSum and Lead-N outperform the other summarisation methods, whilst BERTSum received comparatively lower scores in both subjective and objective evaluations. The results demonstrate that even such simple heuristics-based methods like Lead-N ca n produce meaningful and useful summaries of call centre dialogues.

Probabilistic graphical models are a powerful concept for modeling high-dimensional distributions. Besides modeling distributions, probabilistic graphical models also provide an elegant framework for performing statistical inference; because of the high-dimensional nature, however, one must often use approximate methods for this purpose. Belief propagation performs approximate inference, is efficient, and looks back on a long success-story. Yet, in most cases, belief propagation lacks any performance and convergence guarantees. Many realistic problems are presented by graphical models with loops, however, in which case belief propagation is neither guaranteed to provide accurate estimates nor that it converges at all. This thesis investigates how the model parameters influence the performance of belief propagation. We are particularly interested in their influence on (i) the number of fixed points, (ii) the convergence properties, and (iii) the approximation quality.

Data used to train machine learning (ML) models can be sensitive. Membership inference attacks (MIAs), attempting to determine whether a particular data record was used to train an ML model, risk violating membership privacy. ML model builders need a principled definition of a metric to quantify the membership privacy risk of (a) individual training data records, (b) computed independently of specific MIAs, (c) which assesses susceptibility to different MIAs, (d) can be used for different applications, and (e) efficiently. None of the prior membership privacy risk metrics simultaneously meet all these requirements. We present SHAPr, a membership privacy metric based on Shapley values which is a leave-one-out (LOO) technique, originally intended to measure the contribution of a training data record on model utility. We conjecture that contribution to model utility can act as a proxy for memorization, and hence represent membership privacy risk. Using ten benchmark datasets, we show that SHAPr is indeed effective in estimating susceptibility of training data records to MIAs. We also show that, unlike prior work, SHAPr is significantly better in estimating susceptibility to newer, and more effective MIA. We apply SHAPr to evaluate the efficacy of several defenses against MIAs: using regularization and removing high risk training data records. Moreover, SHAPr is versatile: it can be used for estimating vulnerability of different subgroups to MIAs, and inherits applications of Shapley values (e.g., data valuation). We show that SHAPr has an acceptable computational cost (compared to naive LOO), varying from a few minutes for the smallest dataset to ~92 minutes for the largest dataset.

Transportation activity-based simulators (ABMs) represent an individual traveler's activity patterns and trips throughout the day by using nested choice models. The generated trips are then simulated in a traffic flow simulator to learn system-level patterns. These behaviorally-realistic models require a high-resolution representation of network flows and, thus, are computationally expensive. The very same flexibility which makes these simulation models appealing, also makes their calibration problems intractable, with the number of simulations required to find an optimal solution growing exponentially as the input dimension increases [90, 70]. As a result, the use of these simulators is currently limited to what-if analysis. This paper focuses on calibrating the static choice model parameters used in activity-based simulators. The goal of calibration is to find values of the simulator's input parameters θ that minimizes the deviance between observed data and simulator's outputs.

We consider a novel task of automatically generating text descriptions of music. Compared with other well-established text generation tasks such as image caption, the scarcity of well-paired music and text datasets makes it a much more challenging task. In this paper, we exploit the crowd-sourced music comments to construct a new dataset and propose a sequence-to-sequence model to generate text descriptions of music. More concretely, we use the dilated convolutional layer as the basic component of the encoder and a memory based recurrent neural network as the decoder. To enhance the authenticity and thematicity of generated texts, we further propose to fine-tune the model with a discriminator as well as a novel topic evaluator. To measure the quality of generated texts, we also propose two new evaluation metrics, which are more aligned with human evaluation than traditional metrics such as BLEU. Experimental results verify that our model is capable of generating fluent and meaningful comments while containing thematic and content information of the original music.

The challenge of language grounding is to fully understand natural language by grounding language in real-world referents. While AI techniques are available, the widespread adoption and effectiveness of such technologies for human-robot teams relies critically on user trust. This survey provides three contributions relating to the newly emerging field of trust in language grounding, including a) an overview of language grounding research in terms of AI technologies, data sets, and user interfaces; b) six hypothesised trust factors relevant to language grounding, which are tested empirically on a human-robot cleaning team; and c) future research directions for trust in language grounding.

Automated analysis of electron microscopy datasets poses multiple challenges, such as limitation in the size of the training dataset, variation in data distribution induced by variation in sample quality and experiment conditions, etc. It is crucial for the trained model to continue to provide acceptable segmentation/classification performance on new data, and quantify the uncertainty associated with its predictions. Among the broad applications of machine learning, various approaches have been adopted to quantify uncertainty, such as Bayesian modeling, Monte Carlo dropout, ensembles, etc. With the aim of addressing the challenges specific to the data domain of electron microscopy, two different types of ensembles of pre-trained neural networks were implemented in this work. The ensembles performed semantic segmentation of ice crystal within a two-phase mixture, thereby tracking its phase transformation to water. The first ensemble (EA) is composed of U-net style networks having different underlying architectures, whereas the second series of ensembles (ER-i) are composed of randomly initialized U-net style networks, wherein each base learner has the same underlying architecture 'i'. The encoders of the base learners were pre-trained on the Imagenet dataset. The performance of EA and ER were evaluated on three different metrics: accuracy, calibration, and uncertainty. It is seen that EA exhibits a greater classification accuracy and is better calibrated, as compared to ER. While the uncertainty quantification of these two types of ensembles are comparable, the uncertainty scores exhibited by ER were found to be dependent on the specific architecture of its base member ('i') and not consistently better than EA. Thus, the challenges posed for the analysis of electron microscopy datasets appear to be better addressed by an ensemble design like EA, as compared to an ensemble design like ER.

Recently, there is a revival of interest in low-rank matrix completion-based unsupervised learning through the lens of dual-graph regularization, which has significantly improved the performance of multidisciplinary machine learning tasks such as recommendation systems, genotype imputation and image inpainting. While the dual-graph regularization contributes a major part of the success, computational costly hyper-parameter tunning is usually involved. To circumvent such a drawback and improve the completion performance, we propose a novel Bayesian learning algorithm that automatically learns the hyper-parameters associated with dual-graph regularization, and at the same time, guarantees the low-rankness of matrix completion. Notably, a novel prior is devised to promote the low-rankness of the matrix and encode the dual-graph information simultaneously, which is more challenging than the single-graph counterpart. A nontrivial conditional conjugacy between the proposed priors and likelihood function is then explored such that an efficient algorithm is derived under variational inference framework. Extensive experiments using synthetic and real-world datasets demonstrate the state-of-the-art performance of the proposed learning algorithm for various data analysis tasks.

We propose a novel approach to perform approximate Bayesian inference in complex models such as Bayesian neural networks. The approach is more scalable to large data than Markov Chain Monte Carlo, it embraces more expressive models than Variational Inference, and it does not rely on adversarial training (or density ratio estimation). We adopt the recent approach of constructing two models: (1) a primary model, tasked with performing regression or classification; and (2) a secondary, expressive (e.g. implicit) model that defines an approximate posterior distribution over the parameters of the primary model. However, we optimise the parameters of the posterior model via gradient descent according to a Monte Carlo estimate of the posterior predictive distribution -- which is our only approximation (other than the posterior model). Only a likelihood needs to be specified, which can take various forms such as loss functions and synthetic likelihoods, thus providing a form of a likelihood-free approach. Furthermore, we formulate the approach such that the posterior samples can either be independent of, or conditionally dependent upon the inputs to the primary model. The latter approach is shown to be capable of increasing the apparent complexity of the primary model. We see this being useful in applications such as surrogate and physics-based models. To promote how the Bayesian paradigm offers more than just uncertainty quantification, we demonstrate: uncertainty quantification, multi-modality, as well as an application with a recent deep forecasting neural network architecture.

ProBoost, a new boosting algorithm for probabilistic classifiers, is proposed in this work. This algorithm uses the epistemic uncertainty of each training sample to determine the most challenging/uncertain ones; the relevance of these samples is then increased for the next weak learner, producing a sequence that progressively focuses on the samples found to have the highest uncertainty. In the end, the weak learners' outputs are combined into a weighted ensemble of classifiers. Three methods are proposed to manipulate the training set: undersampling, oversampling, and weighting the training samples according to the uncertainty estimated by the weak learners. Furthermore, two approaches are studied regarding the ensemble combination. The weak learner herein considered is a standard convolutional neural network, and the probabilistic models underlying the uncertainty estimation use either variational inference or Monte Carlo dropout. The experimental evaluation carried out on MNIST benchmark datasets shows that ProBoost yields a significant performance improvement. The results are further highlighted by assessing the relative achievable improvement, a metric proposed in this work, which shows that a model with only four weak learners leads to an improvement exceeding 12% in this metric (for either accuracy, sensitivity, or specificity), in comparison to the model learned without ProBoost.