There has recently been an explosion of work on spoken dialogue systems, along with an increased interest in open-domain systems that engage in casual conversations on popular topics such as movies, books and music. These systems aim to socially engage, entertain, and even empathize with their users. Since the achievement of such social goals is hard to measure, recent research has used dialogue length or human ratings as evaluation metrics, and developed methods for automatically calculating novel metrics, such as coherence, consistency, relevance and engagement. Here we develop a PARADISE model for predicting the performance of Athena, a dialogue system that has participated in thousands of conversations with real users, while competing as a finalist in the Alexa Prize. We use both user ratings and dialogue length as metrics for dialogue quality, and experiment with predicting these metrics using automatic features that are both system dependent and independent. Our goal is to learn a general objective function that can be used to optimize the dialogue choices of any Alexa Prize system in real time and evaluate its performance. Our best model for predicting user ratings gets an R$^2$ of .136 with a DistilBert model, and the best model for predicting length with system independent features gets an R$^2$ of .865, suggesting that conversation length may be a more reliable measure for automatic training of dialogue systems.

An intrinsic time-scale decomposition (ITD) based method for power transformer fault diagnosis is proposed. Dissolved gas analysis (DGA) parameters are ranked according to their skewness, and then ITD based features extraction is performed. An optimal set of PRC features are determined by an XGBoost classifier. For classification purpose, an XGBoost classifier is used to the optimal PRC features set. The proposed method's performance in classification is studied using publicly available DGA data of 376 power transformers and employing an XGBoost classifier. The Proposed method achieves more than 95% accuracy and high sensitivity and F1-score, better than conventional methods and some recent machine learning-based fault diagnosis approaches. Moreover, it gives better Cohen Kappa and F1-score as compared to the recently introduced EMD-based hierarchical technique for fault diagnosis in power transformers.

We design step-size schemes that make stochastic gradient descent (SGD) adaptive to (i) the noise $\sigma^2$ in the stochastic gradients and (ii) problem-dependent constants. When minimizing smooth, strongly-convex functions with condition number $\kappa$, we first prove that $T$ iterations of SGD with Nesterov acceleration and exponentially decreasing step-sizes can achieve a near-optimal $\tilde{O}(\exp(-T/\sqrt{\kappa}) + \sigma^2/T)$ convergence rate. Under a relaxed assumption on the noise, with the same step-size scheme and knowledge of the smoothness, we prove that SGD can achieve an $\tilde{O}(\exp(-T/\kappa) + \sigma^2/T)$ rate. In order to be adaptive to the smoothness, we use a stochastic line-search (SLS) and show (via upper and lower-bounds) that SGD converges at the desired rate, but only to a neighbourhood of the solution. Next, we use SGD with an offline estimate of the smoothness and prove convergence to the minimizer. However, its convergence is slowed down proportional to the estimation error and we prove a lower-bound justifying this slowdown. Compared to other step-size schemes, we empirically demonstrate the effectiveness of exponential step-sizes coupled with a novel variant of SLS.

Noise-contrastive estimation (NCE) is a statistically consistent method for learning unnormalized probabilistic models. It has been empirically observed that the choice of the noise distribution is crucial for NCE's performance. However, such observations have never been made formal or quantitative. In fact, it is not even clear whether the difficulties arising from a poorly chosen noise distribution are statistical or algorithmic in nature. In this work, we formally pinpoint reasons for NCE's poor performance when an inappropriate noise distribution is used. Namely, we prove these challenges arise due to an ill-behaved (more precisely, flat) loss landscape. To address this, we introduce a variant of NCE called "eNCE" which uses an exponential loss and for which normalized gradient descent addresses the landscape issues provably when the target and noise distributions are in a given exponential family.

Understanding when and why interpolating methods generalize well has recently been a topic of interest in statistical learning theory. However, systematically connecting interpolating methods to achievable notions of optimality has only received partial attention. In this paper, we investigate the question of what is the optimal way to interpolate in linear regression using functions that are linear in the response variable (as the case for the Bayes optimal estimator in ridge regression) and depend on the data, the population covariance of the data, the signal-to-noise ratio and the covariance of the prior for the signal, but do not depend on the value of the signal itself nor the noise vector in the training data. We provide a closed-form expression for the interpolator that achieves this notion of optimality and show that it can be derived as the limit of preconditioned gradient descent with a specific initialization. We identify a regime where the minimum-norm interpolator provably generalizes arbitrarily worse than the optimal response-linear achievable interpolator that we introduce, and validate with numerical experiments that the notion of optimality we consider can be achieved by interpolating methods that only use the training data as input in the case of an isotropic prior. Finally, we extend the notion of optimal response-linear interpolation to random features regression under a linear data-generating model that has been previously studied in the literature.

Recent advances in the field of meta-learning have tackled domains consisting of large numbers of small ("few-shot") supervised learning tasks. Meta-learning algorithms must be able to rapidly adapt to any individual few-shot task, fitting to a small support set within a task and using it to predict the labels of the task's query set. This problem setting can be extended to the Bayesian context, wherein rather than predicting a single label for each query data point, a model predicts a distribution of labels capturing its uncertainty. Successful methods in this domain include Bayesian ensembling of MAML-based models, Bayesian neural networks, and Gaussian processes with learned deep kernel and mean functions. While Gaussian processes have a robust Bayesian interpretation in the meta-learning context, they do not naturally model non-Gaussian predictive posteriors for expressing uncertainty. In this paper, we design a theoretically principled method, VMGP, extending Gaussian-process-based meta-learning to allow for high-quality, arbitrary non-Gaussian uncertainty predictions. On benchmark environments with complex non-smooth or discontinuous structure, we find our VMGP method performs significantly better than existing Bayesian meta-learning baselines.

Compressive learning is an approach to efficient large scale learning based on sketching an entire dataset to a single mean embedding (the sketch), i.e. a vector of generalized moments. The learning task is then approximately solved as an inverse problem using an adapted parametric model. Previous works in this context have focused on sketches obtained by averaging random features, that while universal can be poorly adapted to the problem at hand. In this paper, we propose and study the idea of performing sketching based on data-dependent Nystr\"om approximation. From a theoretical perspective we prove that the excess risk can be controlled under a geometric assumption relating the parametric model used to learn from the sketch and the covariance operator associated to the task at hand. Empirically, we show for k-means clustering and Gaussian modeling that for a fixed sketch size, Nystr\"om sketches indeed outperform those built with random features.

Modern machine learning applications often involve learning statistical models of great complexity and datasets of massive size become increasingly available. However, while increasing the size of the training datasets generally offers improvement in model performance, the training process is very computation-intensive and thus time-consuming. Indeed, hardware architectures have physical limits in terms of storage, memory, processing speed and communication. A central challenge is thus to design efficient large-scale algorithms. Distributed learning and parallel computing is a common and simple approach to deal with large datasets. The n observations are evenly split to M machines (or local nodes, workers), each having access to only a subset of n/M training samples. Each machine performs local computations to fit a model and transmits it to a central node for merging. This simple divide and conquer approach having been proposed in e.g.

I'm SUPER EXCITED to show fellow time-series enthusiasts a new way that we can scale time series analysis using an amazing technology called Spark! Without Spark, large-scale forecasting projects of 10,000 time series can take days to run because of long-running for-loops and the need to test many models on each time series. Spark has been widely accepted as a "big data" solution, and we'll use it to scale-out (distribute) our time series analysis to Spark Clusters, and run our analysis in parallel. Spark is an amazing technology for processing large-scale data science workloads. Modeltime is a state-of-the-art forecasting library that I personally developed for "Tidy Forecasting" in R. Modeltime now integrates a Spark Backend with capability of forecasting 10,000 time series using distributed Spark Clusters.

A growing portfolio of research has been reported on the use of machine learning-based architectures and models in the domain of healthcare. The development of data-driven applications and services for the diagnosis and classification of key illness conditions is challenging owing to issues of low volume, low-quality contextual data for the training, and validation of algorithms, which, in turn, compromises the accuracy of the resultant models. Here, a fusion machine learning approach is presented reporting an improvement in the accuracy of the identification of diabetes and the prediction of the onset of critical events for patients with diabetes (PwD). Globally, the cost of treating diabetes, a prevalent chronic illness condition characterized by high levels of sugar in the bloodstream over long periods, is placing severe demands on health providers and the proposed solution has the potential to support an increase in the rates of survival of PwD through informing on the optimum treatment on an individual patient basis. At the core of the proposed architecture is a fusion of machine learning classifiers (Support Vector Machine and Artificial Neural Network).