Coherent reasoning under uncertainty can be represented in a very general manner by coherent sets of desirable gambles. In a context that does not allow for indecision, this leads to an approach that is mathematically equivalent to working with coherent conditional probabilities. If we do allow for indecision, this leads to a more general foundation for coherent (imprecise-)probabilistic inference. In this framework, and for a given finite category set, coherent predictive inference under exchangeability can be represented using Bernstein coherent cones of multivariate polynomials on the simplex generated by this category set. This is a powerful generalisation of de Finetti's Representation Theorem allowing for both imprecision and indecision. We define an inference system as a map that associates a Bernstein coherent cone of polynomials with every finite category set. Many inference principles encountered in the literature can then be interpreted, and represented mathematically, as restrictions on such maps. We discuss, as particular examples, two important inference principles: representation insensitivitya strengthened version of Walley's representation invarianceand specificity. We show that there is an infinity of inference systems that satisfy these two principles, amongst which we discuss in particular the skeptically cautious inference system, the inference systems corresponding to (a modified version of) Walley and Bernard's Imprecise Dirichlet Multinomial Models (IDMM), the skeptical IDMM inference systems, and the Haldane inference system. We also prove that the latter produces the same posterior inferences as would be obtained using Haldane's improper prior, implying that there is an infinity of proper priors that produce the same coherent posterior inferences as Haldane's improper one. Finally, we impose an additional inference principle that allows us to characterise uniquely the immediate predictions for the IDMM inference systems.

A natural approach to analyze interaction data of form "what-connects-to-what-when" is to create a time-series (or rather a sequence) of graphs through temporal discretization (bandwidth selection) and spatial discretization (vertex contraction). Such discretization together with non-negative factorization techniques can be useful for obtaining clustering of graphs. Motivating application of performing clustering of graphs (as opposed to vertex clustering) can be found in neuroscience and in social network analysis, and it can also be used to enhance community detection (i.e., vertex clustering) by way of conditioning on the cluster labels. In this paper, we formulate a problem of clustering of graphs as a model selection problem. Our approach involves information criteria, non-negative matrix factorization and singular value thresholding, and we illustrate our techniques using real and simulated data.

In certain situations that shall be undoubtedly more and more common in the Big Data era, the datasets available are so massive that computing statistics over the full sample is hardly feasible, if not unfeasible. A natural approach in this context consists in using survey schemes and substituting the "full data" statistics with their counterparts based on the resulting random samples, of manageable size. It is the main purpose of this paper to investigate the impact of survey sampling with unequal inclusion probabilities on stochastic gradient descent-based M-estimation methods in large-scale statistical and machine-learning problems. Precisely, we prove that, in presence of some a priori information, one may significantly increase asymptotic accuracy when choosing appropriate first order inclusion probabilities, without affecting complexity. These striking results are described here by limit theorems and are also illustrated by numerical experiments.

Directed graphs have asymmetric connections, yet the current graph clustering methodologies cannot identify the potentially global structure of these asymmetries. We give a spectral algorithm called di-sim that builds on a dual measure of similarity that correspond to how a node (i) sends and (ii) receives edges. Using di-sim, we analyze the global asymmetries in the networks of Enron emails, political blogs, and the c elegans neural connectome. In each example, a small subset of nodes have persistent asymmetries; these nodes send edges with one cluster, but receive edges with another cluster. Previous approaches would have assigned these asymmetric nodes to only one cluster, failing to identify their sending/receiving asymmetries. Regularization and "projection" are two steps of di-sim that are essential for spectral clustering algorithms to work in practice. The theoretical results show that these steps make the algorithm weakly consistent under the degree corrected Stochastic co-Blockmodel, a model that generalizes the Stochastic Blockmodel to allow for both (i) degree heterogeneity and (ii) the global asymmetries that we intend to detect. The theoretical results make no assumptions on the smallest degree nodes. Instead, the theorem requires that the average degree grows sufficiently fast and that the weak consistency only applies to the subset of the nodes with sufficiently large leverage scores. The results results also apply to bipartite graphs.

Finite chase, or alternatively chase termination, is an important condition to ensure the decidability of existential rule languages. In the past few years, a number of rule languages with finite chase have been studied. In this work, we propose a novel approach for classifying the rule languages with finite chase. Using this approach, a family of decidable rule languages, which extend the existing languages with the finite chase property, are naturally defined. We then study the complexity of these languages. Although all of them are tractable for data complexity, we show that their combined complexity can be arbitrarily high. Furthermore, we prove that all the rule languages with finite chase that extend the weakly acyclic language are of the same expressiveness as the weakly acyclic one, while rule languages with higher combined complexity are in general more succinct than those with lower combined complexity.

In recent years, random matrices have come to play a major role in computational mathematics, but most of the classical areas of random matrix theory remain the province of experts. Over the last decade, with the advent of matrix concentration inequalities, research has advanced to the point where we can conquer many (formerly) challenging problems with a page or two of arithmetic. My aim is to describe the most successful methods from this area along with some interesting examples that these techniques can illuminate. I hope that the results in these pages will inspire future work on applications of random matrices as well as refinements of the matrix concentration inequalities discussed herein. I have chosen to present a coherent body of results based on a generalization of the Laplace transform method for establishing scalar concentration inequalities. In the last two years, Lester Mackey and I, together with our coauthors, have developed an alternative approach to matrix concentration using exchangeable pairs and Markov chain couplings. With some regret, I have chosen to omit this theory because the ideas seem less accessible to a broad audience of researchers. The interested reader will find pointers to these articles in the annotated bibliography. The work described in these notes reflects the influence of many researchers.

This article is an overview of the "SP theory of intelligence". The theory aims to simplify and integrate concepts across artificial intelligence, mainstream computing and human perception and cognition, with information compression as a unifying theme. It is conceived as a brain-like system that receives 'New' information and stores some or all of it in compressed form as 'Old' information. It is realised in the form of a computer model -- a first version of the SP machine. The concept of "multiple alignment" is a powerful central idea. Using heuristic techniques, the system builds multiple alignments that are 'good' in terms of information compression. For each multiple alignment, probabilities may be calculated. These provide the basis for calculating the probabilities of inferences. The system learns new structures from partial matches between patterns. Using heuristic techniques, the system searches for sets of structures that are 'good' in terms of information compression. These are normally ones that people judge to be 'natural', in accordance with the 'DONSVIC' principle -- the discovery of natural structures via information compression. The SP theory may be applied in several areas including 'computing', aspects of mathematics and logic, representation of knowledge, natural language processing, pattern recognition, several kinds of reasoning, information storage and retrieval, planning and problem solving, information compression, neuroscience, and human perception and cognition. Examples include the parsing and production of language including discontinuous dependencies in syntax, pattern recognition at multiple levels of abstraction and its integration with part-whole relations, nonmonotonic reasoning and reasoning with default values, reasoning in Bayesian networks including 'explaining away', causal diagnosis, and the solving of a geometric analogy problem.

Tumors contain multiple, genetically diverse subclonal populations of cells that have evolved from a single progenitor population through successive waves of expansion and selection [1-3]. Reconstructing their evolutionary histories can help identify characteristic driver mutations associated with cancer development and progression [4, 5]; and can provide insight into how tumors might respond to treatment [6, 7]. In some cases, it is possible to genotype the subpopulations present in a tumor, while reconstructing its history, using the population frequencies of mutations that distinguish these subclonal populations [2, 8-21]. Increasingly, tumors are being characterized using whole genome sequencing (WGS) of bulk tumor samples [22] and few automated methods exist to reliably perform this reconstruction on the basis of these data. Subclonal reconstruction algorithms attempt to infer the population structure of heterogeneous tumors based on the measured variant allelic frequency (VAF) of their somatic mutations.

Diverse experimental data ranging from microarray gene expression data in biology to spectrum data in astronomy require to be clustered to signal meaningful correlation of the data. Massive documents or images on internet are also needed to be effectively organized so as to promote the efficiency of search engines. Clustering method as K-means (1) is popular for its simplicity, yet sensitive to noise and initialization and thus is limited by the lack of reliability. Hierarchical clustering (HC) (2) is simple and intuitive and thus widely used especially in biology (3), whereas it needs a large computation (4) and its result is variable to a set of similarity measures between clusters. Moreover, the cluster number for the above methods needs to be prespecified (e.g., K-means) or determined by a threshold (e.g., HC). Some other well-known algorithms either involve complex optimization and postprocessing (5), or have limited range of applications such as the distribution (6) or the attribute of data (7, 8). Although affinity propagation (AP) (9) has much better performance than K-means and the cluster number is determined automatically, it is not good at detecting nonspherical clusters (10). Recently, two effective clustering algorithms (10, 11) were proposed, which can together form a pool of clustering methods based on the in-tree structure (11). But they involve a free parameter.

Regularization is essential when training large neural networks. As deep neural networks can be mathematically interpreted as universal function approximators, they are effective at memorizing sampling noise in the training data. This results in poor generalization to unseen data. Therefore, it is no surprise that a new regularization technique, Dropout, was partially responsible for the now-ubiquitous winning entry to ImageNet 2012 by the University of Toronto. Currently, Dropout (and related methods such as DropConnect) are the most effective means of regularizing large neural networks. These amount to efficiently visiting a large number of related models at training time, while aggregating them to a single predictor at test time. The proposed FaMe model aims to apply a similar strategy, yet learns a factorization of each weight matrix such that the factors are robust to noise.