The Gaussian Process (GP) framework flexibility has enabled its use in several data modeling scenarios. The setting where we have unavailable or uncertain inputs that generate possibly noisy observations is usually tackled by the well known Gaussian Process Latent Variable Model (GPLVM). However, the standard variational approach to perform inference with the GPLVM presents some expressions that are tractable for only a few kernel functions, which may hinder its general application. While other quadrature or sampling approaches could be used in that case, they usually are very slow and/or non-deterministic. In the present paper, we propose the use of the unscented transformation to enable the use of any kernel function within the Bayesian GPLVM. Our approach maintains the fully deterministic feature of tractable kernels and presents a simple implementation with only moderate computational cost. Experiments on dimensionality reduction and multistep-ahead prediction with uncertainty propagation indicate the feasibility of our proposal.

Automatic cover detection -- the task of finding in an audio database all the covers of one or several query tracks -- has long been seen as a challenging theoretical problem in the MIR community and as an acute practical problem for authors and composers societies. Original algorithms proposed for this task have proven their accuracy on small datasets, but are unable to scale up to modern real-life audio corpora. On the other hand, faster approaches designed to process thousands of pairwise comparisons resulted in lower accuracy, making them unsuitable for practical use. In this work, we propose a neural network architecture that is trained to represent each track as a single embedding vector. The computation burden is therefore left to the embedding extraction -- that can be conducted offline and stored, while the pairwise comparison task reduces to a simple Euclidean distance computation. We further propose to extract each track's embedding out of its dominant melody representation, obtained by another neural network trained for this task. We then show that this architecture improves state-of-the-art accuracy both on small and large datasets, and is able to scale to query databases of thousands of tracks in a few seconds.

In medical real-world study (RWS), how to fully utilize the fragmentary and scarce information in model training to generate the solid diagnosis results is a challenging task. In this work, we introduce a novel multi-instance neural network, AMI-Net+, to train and predict from the incomplete and extremely imbalanced data. It is more effective than the state-of-art method, AMI-Net. First, we also implement embedding, multi-head attention and gated attention-based multi-instance pooling to capture the relations of symptoms themselves and with the given disease. Besides, we propose var-ious improvements to AMI-Net, that the cross-entropy loss is replaced by focal loss and we propose a novel self-adaptive multi-instance pooling method on instance-level to obtain the bag representation. We validate the performance of AMI-Net+ on two real-world datasets, from two different medical domains. Results show that our approach outperforms other base-line models by a considerable margin.

We study the problem of finding the best linear model that can minimize least-squares loss given a data-set. While this problem is trivial in the low dimensional regime, it becomes more interesting in high dimensions where the population minimizer is assumed to lie on a manifold such as sparse vectors. We propose projected gradient descent (PGD) algorithm to estimate the population minimizer in the finite sample regime. We establish linear convergence rate and data dependent estimation error bounds for PGD. Our contributions include: 1) The results are established for heavier tailed sub-exponential distributions besides sub-gaussian. 2) We directly analyze the empirical risk minimization and do not require a realizable model that connects input data and labels. 3) Our PGD algorithm is augmented to learn the bias terms which boosts the performance. The numerical experiments validate our theoretical results.

Information-theoretic quantities, such as mutual information and conditional entropy, are useful statistics for measuring the dependence between two random variables. However, estimating these quantities in a non-parametric fashion is difficult, especially when the variables are high-dimensional, a mixture of continuous and discrete values, or both. In this paper, we propose a decision forest method, Conditional Forests (CF), to estimate these quantities. By combining quantile regression forests with honest sampling, and introducing a finite sample correction, CF improves finite sample bias in a range of settings. We demonstrate through simulations that CF achieves smaller bias and variance in both low- and high-dimensional settings for estimating posteriors, conditional entropy, and mutual information. We then use CF to estimate the amount of information between neuron class and other ceulluar feautres.

A visually-grounded navigation instruction can be interpreted as a sequence of expected observations and actions an agent following the correct trajectory would encounter and perform. Based on this intuition, we formulate the problem of finding the goal location in Vision-And-Language Navigation (VLN) [1] within the framework of Bayesian state tracking - learning observation and motion models conditioned on these expectable events. Together with a mapper that constructs a semantic spatial map on-the-fly during navigation, we formulate an end-to-end differentiable Bayes filter and train it to identify the goal by predicting the most likely trajectory through the map according to the instructions. The resulting navigation policy constitutes a new approach to instruction following that explicitly models a probability distribution over states, encoding strong geometric and algorithmic priors while enabling greater explainability. Our experiments show that our approach outperforms strong baselines when predicting the goal location in VLN.

The choice of how to retain information about past gradients dramatically affects the convergence properties of state-of-the-art stochastic optimization methods, such as Heavy-ball, Nesterov's momentum, RMSprop and Adam. Building on this observation, we use stochastic differential equations (SDEs) to explicitly study the role of memory in gradient-based algorithms. We first derive a general continuous-time model that can incorporate arbitrary types of memory, for both deterministic and stochastic settings. We provide convergence guarantees for this SDE for weakly-quasi-convex and quadratically growing functions. We then demonstrate how to discretize this SDE to get a flexible discrete-time algorithm that can implement a board spectrum of memories ranging from short- to long-term. Not only does this algorithm increase the degrees of freedom in algorithmic choice for practitioners but it also comes with better stability properties than classical momentum in the convex stochastic setting. In particular, no iterate averaging is needed for convergence. Interestingly, our analysis also provides a novel interpretation of Nesterov's momentum as stable gradient amplification and highlights a possible reason for its unstable behavior in the (convex) stochastic setting. Furthermore, we discuss the use of long term memory for second-moment estimation in adaptive methods, such as Adam and RMSprop. Finally, we provide an extensive experimental study of the effect of different types of memory in both convex and nonconvex settings.

Recent literature has shown several benefits of hyperbolic embedding of graph-structured data (GSD) in representing their structures and latent relations. While several studies have explored the ability of hyperbolic embedding to represent data (for example, by quantifying their mean average precision) and their ability to produce better visualisations of clusters, only few works exploited the effectiveness of hyperbolic embedding to perform learning on the initial GSD. Motivated by innovative ideas from the fields of Brain computer interfaces and Radar processing, this paper presents a new scheme for learning GSD based on hyperbolic embedding, Riemannian barycentre (i.e. Fr\'echet or geometric mean) and $K$-means algorithms as a significant tool that derives from it. The main idea is as follows. Relying on the Riemannian barycentre, we define a notion of minimal variance which allows us to choose an embedding between different ones. This embedding is used thereafter together with $K$-means algorithms to perform unsupervised clustering and in combination with the nearest neighbour rule to perform supervised learning. We demonstrate the performance of the proposed framework through several experiments on real-world social networks and hierarchical GSD. The obtained results outperform their counterparts in high-dimensional Euclidean spaces and recent proposed geometric approaches.

Despite continuing medical advances, the rate of newborn morbidity and mortality globally remains high, with over 6 million casualties every year. The prediction of pathologies affecting newborns based on their cry is thus of significant clinical interest, as it would facilitate the development of accessible, low-cost diagnostic tools\cut{ based on wearables and smartphones}. However, the inadequacy of clinically annotated datasets of infant cries limits progress on this task. This study explores a neural transfer learning approach to developing accurate and robust models for identifying infants that have suffered from perinatal asphyxia. In particular, we explore the hypothesis that representations learned from adult speech could inform and improve performance of models developed on infant speech. Our experiments show that models based on such representation transfer are resilient to different types and degrees of noise, as well as to signal loss in time and frequency domains.

Kriging is an efficient machine-learning tool, which allows to obtain an approximate response of an investigated phenomenon on the whole parametric space. Adaptive schemes provide a the ability to guide the experiment yielding new sample point positions to enrich the metamodel. Herein a novel adaptive scheme called Monte Carlo-intersite Voronoi (MiVor) is proposed to efficiently identify binary decision regions on the basis of a regression surrogate model. The performance of the innovative approach is tested for analytical functions as well as some mechanical problems and is furthermore compared to two regression-based adaptive schemes. For smooth problems, all three methods have comparable performances. For highly fluctuating response surface as encountered e.g. for dynamics or damage problems, the innovative MiVor algorithm performs very well and provides accurate binary classification with only a few observation points.