Detecting overfitting in generative models is an important challenge in machine learning. In this work, we formalize a form of overfitting that we call {\em{data-copying}} -- where the generative model memorizes and outputs training samples or small variations thereof. We provide a three sample non-parametric test for detecting data-copying that uses the training set, a separate sample from the target distribution, and a generated sample from the model, and study the performance of our test on several canonical models and datasets. For code \& examples, visit https://github.com/casey-meehan/data-copying

In machine learning, one of the frequently encountered problems is grouping similar data together. You know the income level of people and now you want to group people with similar income levels together. You want to know who are the people with low income power, the people with high or very high income power, which you think can be helpful in devising a perfect marketing strategy. You have the shopping data of customers with you and now you want to group customers with similar shopping preferences together or you being the bio student want to know which cells share similar properties based upon the data about the cells you have at your hand. All the above stated problems come under the domain of an unsupervised machine learning method called clustering. Although there are a number of clustering algorithms there but when it comes to the simplest one, the award will go to K-Means clustering.

In my former article, I've been introducing the concept of Principal Components Analysis, a powerful technique used in Machine Learning to reduce the dimensionality of your input data. An important mathematical feature of PCA is its link with the Eigenvectors (and Eigenvalues) of the original features matrix X, and in this article, I'm going to show the ratio behind this link. So let's start with a brief recap of PCA. Differently from features selection techniques, PCA does not use a subset of the initial features: it rather computes a new set of features obtained as linear combinations of the original ones, living in a lower-dimensional space. Where M is at most as big as p, but of course, because of the purpose of the task (dimensionality reduction), it is chosen to be less than p. The coefficients of the first 2 lines are known as loading factors and are computed in such a way that these new variables are uncorrelated and most of the information within the initial variables is stored into the first components.

Quadratic discriminant analysis (QDA) is a widely used statistical tool to classify observations from different multivariate Normal populations. The generalized quadratic discriminant analysis (GQDA) classification rule/classifier, which generalizes the QDA and the minimum Mahalanobis distance (MMD) classifiers to discriminate between populations with underlying elliptically symmetric distributions competes quite favorably with the QDA classifier when it is optimal and performs much better when QDA fails under non-Normal underlying distributions, e.g. Cauchy distribution. However, the classification rule in GQDA is based on the sample mean vector and the sample dispersion matrix of a training sample, which are extremely non-robust under data contamination. In real world, since it is quite common to face data highly vulnerable to outliers, the lack of robustness of the classical estimators of the mean vector and the dispersion matrix reduces the efficiency of the GQDA classifier significantly, increasing the misclassification errors. The present paper investigates the performance of the GQDA classifier when the classical estimators of the mean vector and the dispersion matrix used therein are replaced by various robust counterparts. Applications to various real data sets as well as simulation studies reveal far better performance of the proposed robust versions of the GQDA classifier. A Comparative study has been made to advocate the appropriate choice of the robust estimators to be used in a specific situation of the degree of contamination of the data sets.

We address the challenging problem of deep representation learning--the efficient adaption of a pre-trained deep network to different tasks. Specifically, we propose to explore gradient-based features. These features are gradients of the model parameters with respect to a task-specific loss given an input sample. Our key innovation is the design of a linear model that incorporates both gradient and activation of the pre-trained network. We show that our model provides a local linear approximation to an underlying deep model, and discuss important theoretical insights. Moreover, we present an efficient algorithm for the training and inference of our model without computing the actual gradient. Our method is evaluated across a number of representation-learning tasks on several datasets and using different network architectures. Strong results are obtained in all settings, and are well-aligned with our theoretical insights.

Recently, there has been a surge of interest in representation learning in hyperbolic spaces, driven by their ability to represent hierarchical data with significantly fewer dimensions than standard Euclidean spaces. However, the viability and benefits of hyperbolic spaces for downstream machine learning tasks have received less attention. In this paper, we present, to our knowledge, the first theoretical guarantees for learning a classifier in hyperbolic rather than Euclidean space. Specifically, we consider the problem of learning a large-margin classifier for data possessing a hierarchical structure. Our first contribution is a hyperbolic perceptron algorithm, which provably converges to a separating hyperplane. We then provide an algorithm to efficiently learn a large-margin hyperplane, relying on the careful injection of adversarial examples. Finally, we prove that for hierarchical data that embeds well into hyperbolic space, the low embedding dimension ensures superior guarantees when learning the classifier directly in hyperbolic space.

The field of meta-learning, or learning-to-learn, has seen a dramatic rise in interest in recent years. Contrary to conventional approaches to AI where a given task is solved from scratch using a fixed learning algorithm, meta-learning aims to improve the learning algorithm itself, given the experience of multiple learning episodes. This paradigm provides an opportunity to tackle many of the conventional challenges of deep learning, including data and computation bottlenecks, as well as the fundamental issue of generalization. In this survey we describe the contemporary meta-learning landscape. We first discuss definitions of meta-learning and position it with respect to related fields, such as transfer learning, multi-task learning, and hyperparameter optimization. We then propose a new taxonomy that provides a more comprehensive breakdown of the space of meta-learning methods today. We survey promising applications and successes of meta-learning including few-shot learning, reinforcement learning and architecture search. Finally, we discuss outstanding challenges and promising areas for future research.

Variational inference methods have been shown to lead to significant improvements in the computational efficiency of approximate Bayesian inference in mixed multinomial logit models when compared to standard Markov-chain Monte Carlo (MCMC) methods without compromising accuracy. However, despite their demonstrated efficiency gains, existing methods still suffer from important limitations that prevent them to scale to very large datasets, while providing the flexibility to allow for rich prior distributions and to capture complex posterior distributions. In this paper, we propose an Amortized Variational Inference approach that leverages stochastic backpropagation, automatic differentiation and GPU-accelerated computation, for effectively scaling Bayesian inference in Mixed Multinomial Logit models to very large datasets. Moreover, we show how normalizing flows can be used to increase the flexibility of the variational posterior approximations. Through an extensive simulation study, we empirically show that the proposed approach is able to achieve computational speedups of multiple orders of magnitude over traditional MSLE and MCMC approaches for large datasets without compromising estimation accuracy.

In the near future, more and more machines will perform tasks in the vicinity of human spaces or support them directly in their spatially bound activities. In order to simplify the verbal communication and the interaction between robotic units and/or humans, reliable and robust systems w.r.t. noise and processing results are needed. This work builds a foundation to address this task. By using a continuous representation of spatial perception in interiors learned from trajectory data, our approach clusters movement in dependency to its spatial context. We propose an unsupervised learning approach based on a neural autoencoding that learns semantically meaningful continuous encodings of spatio-temporal trajectory data. This learned encoding can be used to form prototypical representations. We present promising results that clear the path for future applications.

Multilayer networks proved to be suitable in extracting and providing dependency information of different complex systems. The construction of these networks is difficult and is mostly done with a static approach, neglecting time delayed interdependences. Tensors are objects that naturally represent multilayer networks and in this paper, we propose a new methodology based on Tucker tensor autoregression in order to build a multilayer network directly from data. This methodology captures within and between connections across layers and makes use of a filtering procedure to extract relevant information and improve visualization. We show the application of this methodology to different stationary fractionally differenced financial data. We argue that our result is useful to understand the dependencies across three different aspects of financial risk, namely market risk, liquidity risk, and volatility risk. Indeed, we show how the resulting visualization is a useful tool for risk managers depicting dependency asymmetries between different risk factors and accounting for delayed cross dependencies. The constructed multilayer network shows a strong interconnection between the volumes and prices layers across all the stocks considered while a lower number of interconnections between the uncertainty measures is identified.