Markov chain Monte Carlo (MCMC) algorithms for hidden Markov models often rely on the forward-backward sampler. This makes them computationally slow as the length of the time series increases, motivating the recent development of sub-sampling-based approaches. These approximate the full posterior by using small random subsequences of the data at each MCMC iteration within stochastic gradient MCMC. In the presence of imbalanced data resulting from rare latent states, subsequences often exclude rare latent state data, leading to inaccurate inference and prediction/detection of rare events. We propose a targeted sub-sampling (TASS) approach that over-samples observations corresponding to rare latent states when calculating the stochastic gradient of parameters associated with them. TASS uses an initial clustering of the data to construct subsequence weights that reduce the variance in gradient estimation. This leads to improved sampling efficiency, in particular in settings where the rare latent states correspond to extreme observations. We demonstrate substantial gains in predictive and inferential accuracy on real and synthetic examples.

With the increasing popularity of graph-based methods for dimensionality reduction and representation learning, node embedding functions have become important objects of study in the literature. In this paper, we take an axiomatic approach to understanding node embedding methods, first stating three properties for embedding dissimilarity networks, then proving that all three cannot be satisfied simultaneously by any node embedding method. Similar to existing results on the impossibility of clustering under certain axiomatic assumptions, this points to fundamental difficulties inherent to node embedding tasks. Once these difficulties are identified, we then relax these axioms to allow for certain node embedding methods to be admissible in our framework.

Many real-world optimisation problems such as hyperparameter tuning in machine learning or simulation-based optimisation can be formulated as expensive-to-evaluate black-box functions. A popular approach to tackle such problems is Bayesian optimisation (BO), which builds a response surface model based on the data collected so far, and uses the mean and uncertainty predicted by the model to decide what information to collect next. In this paper, we propose a novel variant of the well-known Knowledge Gradient acquisition function that allows it to handle constraints. We empirically compare the new algorithm with four other state-of-the-art constrained Bayesian optimisation algorithms and demonstrate its superior performance. We also prove theoretical convergence in the infinite budget limit.

Knowledge distillation, i.e., one classifier being trained on the outputs of another classifier, is an empirically very successful technique for knowledge transfer between classifiers. It has even been observed that classifiers learn much faster and more reliably if trained with the outputs of another classifier as soft labels, instead of from ground truth data. So far, however, there is no satisfactory theoretical explanation of this phenomenon. In this work, we provide the first insights into the working mechanisms of distillation by studying the special case of linear and deep linear classifiers. Specifically, we prove a generalization bound that establishes fast convergence of the expected risk of a distillation-trained linear classifier. From the bound and its proof we extract three key factors that determine the success of distillation: * data geometry -- geometric properties of the data distribution, in particular class separation, has a direct influence on the convergence speed of the risk; * optimization bias -- gradient descent optimization finds a very favorable minimum of the distillation objective; and * strong monotonicity -- the expected risk of the student classifier always decreases when the size of the training set grows.

New research out of China is offering a novel method for Natural Language Processing (NLP) to perform sentiment analysis on social media forums and language research datasets – by categorizing and labeling animated GIFs that are posted in response to text announcements. The researchers, led by Boaz Shmueli of National Tsing Hua University at Taiwan, have used Twitter's in-built database of reaction GIFs as an index to quantify the affective state of a user's response, obviating the need to negotiate multiple language responses, the challenge of detecting sarcasm, or of identifying core emotional temperature from ambiguous or excessively brief responses. Clicking the'GIF' button when composing a Twitter post offers a standard set of labeled animated GIFs that are potentially easier for NLP to parse into'identified' emotions than plain-text language. The paper characterizes the use of reaction GIFs in this way as'a new type of label, not yet available in NLP emotion datasets', and notes that existing datasets either use the dimensional model of emotion or the discrete emotions model, neither of which offers this kind of insight. An animated GIF response to a user post.

Functional data analysis (FDA) concerns the study of observations that can be represented as functions, often residing in an infinite-dimensional space. In the recent decades, FDA saw remarkable progress, with many theoretical and practical problems successfully resolved. Often, however, statistical concepts used in finite-dimensional spaces do not readily generalise to random functions. As a result, alternative definitions and desired properties have to be used [80, 27, 46, 47]. An example of a prominent tool of multivariate analysis that is difficult to generalise to functional data is statistical depth. Developed in the 1980s, statistical depth is an umbrella term for methods introducing orderings, ranks, and by extension, nonparametric statistical inference, to multivariate and more complex datasets.

This paper presents Sparse Tensor Classifier (STC), a supervised classification algorithm for categorical data inspired by the notion of superposition of states in quantum physics. By regarding an observation as a superposition of features, we introduce the concept of wave-particle duality in machine learning and propose a generalized framework that unifies the classical and the quantum probability. We show that STC possesses a wide range of desirable properties not available in most other machine learning methods but it is at the same time exceptionally easy to comprehend and use. Empirical evaluation of STC on structured data and text classification demonstrates that our methodology achieves state-of-theart performances compared to both standard classifiers and deep learning, at the additional benefit of requiring minimal data pre-processing and hyper-parameter tuning. Moreover, STC provides a native explanation of its predictions both for single instances and for each target label globally. All the code is released at https://sparsetensorclassifier.org

We propose a new technique for the automatic generation of optimal ad-hoc ans\"atze for classification by using quantum support vector machine (QSVM). This efficient method is based on NSGA-II multiobjective genetic algorithms which allow both maximize the accuracy and minimize the ansatz size. It is demonstrated the validity of the technique by a practical example with a non-linear dataset, interpreting the resulting circuit and its outputs. We also show other application fields of the technique that reinforce the validity of the method, and a comparison with classical classifiers in order to understand the advantages of using quantum machine learning.

Many important application domains generate distributed collections of heterogeneous local datasets. These local datasets are often related via an intrinsic network structure that arises from domain-specific notions of similarity between local datasets. Different notions of similarity are induced by spatiotemporal proximity, statistical dependencies, or functional relations. We use this network structure to adaptively pool similar local datasets into nearly homogenous training sets for learning tailored models. Our main conceptual contribution is to formulate networked federated learning using the concept of generalized total variation (GTV) minimization as a regularizer. This formulation is highly flexible and can be combined with almost any parametric model including Lasso or deep neural networks. We unify and considerably extend some well-known approaches to federated multi-task learning. Our main algorithmic contribution is a novel federated learning algorithm that is well suited for distributed computing environments such as edge computing over wireless networks. This algorithm is robust against model misspecification and numerical errors arising from limited computational resources including processing time or wireless channel bandwidth. As our main technical contribution, we offer precise conditions on the local models as well on their network structure such that our algorithm learns nearly optimal local models. Our analysis reveals an interesting interplay between the (information-) geometry of local models and the (cluster-) geometry of their network.

Modern vehicles, including connected vehicles and autonomous vehicles, nowadays involve many electronic control units connected through intra-vehicle networks to implement various functionalities and perform actions. Modern vehicles are also connected to external networks through vehicle-to-everything technologies, enabling their communications with other vehicles, infrastructures, and smart devices. However, the improving functionality and connectivity of modern vehicles also increase their vulnerabilities to cyber-attacks targeting both intra-vehicle and external networks due to the large attack surfaces. To secure vehicular networks, many researchers have focused on developing intrusion detection systems (IDSs) that capitalize on machine learning methods to detect malicious cyber-attacks. In this paper, the vulnerabilities of intra-vehicle and external networks are discussed, and a multi-tiered hybrid IDS that incorporates a signature-based IDS and an anomaly-based IDS is proposed to detect both known and unknown attacks on vehicular networks. Experimental results illustrate that the proposed system can detect various types of known attacks with 99.99% accuracy on the CAN-intrusion-dataset representing the intra-vehicle network data and 99.88% accuracy on the CICIDS2017 dataset illustrating the external vehicular network data. For the zero-day attack detection, the proposed system achieves high F1-scores of 0.963 and 0.800 on the above two datasets, respectively. The average processing time of each data packet on a vehicle-level machine is less than 0.6 ms, which shows the feasibility of implementing the proposed system in real-time vehicle systems. This emphasizes the effectiveness and efficiency of the proposed IDS.