The ubiquitous complex networks are often composed of regular and irregular components, which makes uncovering the complexity of network structure into a fundamental challenge in network science. Exploring the regular information and identifying the roles of microscopic elements in network organization can help practitioners to recognize the universal principles of network formation and facilitate network data mining.Despite many algorithms having been proposed for link prediction and network reconstruction, estimating and regulating the reconstructability of complex networks remains an inadequately explored problem. With the practical assumption that there has consistence between local structures of networks and the corresponding adjacency matrices are approximately low rank, we obtain a self-representation network model in which the organization principles of networks are captured by representation matrix. According to the model, original networks can be reconstructed based on observed structure. What's more, the model enables us to estimate to what extent networks are regulable, in other words, measure the reconstructability of complex networks. In addition, the model enables us to measure the importance of network links for network regularity thereby allowing us to regulate the reconstructability of networks. The extensive experiments on disparate networks demonstrate the effectiveness of the proposed algorithm and measure. Specifically, the structural regularity reflects the reconstructability of networks, and the reconstruction accuracy can be promoted via the deleting of irregular network links independent of specific algorithms.

The central problem of this paper is the cost-sensitive binary classification problem, where different costs are associated with different types of mistakes. Several important machine learning applications such as medical decision making, targeted marketing, and intrusion detection can be naturally formalized as costsensitive classification setup ([1]). In these domains, the cost of missing a target is much higher than that of a false-positive, and classifiers that do not take misclassification costs into account do not perform well. The cost-sensitive classification problem has been extensively studied, and people have developed efficient algorithms with provable guarantees on the (generalization) error [6, 9, 26, 27, 11, 4]. These methods primarily take existing classification methods based on empirical risk minimization and try to adapt them in various ways to be sensitive to these misclassification costs. Despite all these efforts, the understanding of the fundamental limits of this problem is still missing. In this paper, we study the hardness of this problem by obtaining minimax lower bounds. In particular, we are interested in understanding how the cost parameter influences the hardness or complexity of the cost-sensitive classification. Minimax Lower Bounds Understanding the hardness or fundamental limits of a learning problem is important for practice for the following reasons: - They give an estimate on the number of samples required for a good performance of a learning algorithm.

Meta-learning approaches have been proposed to tackle the few-shot learning problem.Typically, a meta-learner is trained on a variety of tasks in the hopes of being generalizable to new tasks. However, the generalizability on new tasks of a meta-learner could be fragile when it is over-trained on existing tasks during meta-training phase. In other words, the initial model of a meta-learner could be too biased towards existing tasks to adapt to new tasks, especially when only very few examples are available to update the model. To avoid a biased meta-learner and improve its generalizability, we propose a novel paradigm of Task-Agnostic Meta-Learning (TAML) algorithms. Specifically, we present an entropy-based approach that meta-learns an unbiased initial model with the largest uncertainty over the output labels by preventing it from over-performing in classification tasks. Alternatively, a more general inequality-minimization TAML is presented for more ubiquitous scenarios by directly minimizing the inequality of initial losses beyond the classification tasks wherever a suitable loss can be defined.Experiments on benchmarked datasets demonstrate that the proposed approaches outperform compared meta-learning algorithms in both few-shot classification and reinforcement learning tasks.

This paper presents a locally decoupled network parameter learning with local propagation. Three elements are taken into account: (i) sets of nonlinear transforms that describe the representations at all nodes, (ii) a local objective at each node related to the corresponding local representation goal, and (iii) a local propagation model that relates the nonlinear error vectors at each node with the goal error vectors from the directly connected nodes. The modeling concepts (i), (ii) and (iii) offer several advantages, including (a) a unified learning principle for any network that is represented as a graph, (b) understanding and interpretation of the local and the global learning dynamics, (c) decoupled and parallel parameter learning, (d) a possibility for learning in infinitely long, multi-path and multi-goal networks. Numerical experiments validate the potential of the learning principle. The preliminary results show advantages in comparison to the state-of-the-art methods, w.r.t. the learning time and the network size while having comparable recognition accuracy.

In many practical applications, machine learning is divided over multiple agents, where each agent learns a different task and/or learns from a different dataset. We present Epsilon Model Spaces (EMS), a framework for learning a global model by aggregating local learnings performed by each agent. Our approach forgoes sharing of data between agents, makes no assumptions on the distribution of data across agents, and requires minimal communication between agents. We empirically validate our techniques on MNIST experiments and discuss how EMS can generalize to a wide range of problem settings, including federated averaging and catastrophic forgetting. We believe our framework to be among the first to lay out a general methodology for "combining" distinct models.

Unsupervised Learning is a class of Machine Learning techniques to find the patterns in data. The data given to unsupervised algorithm are not labelled, which means only the input variables(X) are given with no corresponding output variables. In unsupervised learning, the algorithms are left to themselves to discover interesting structures in the data. In supervised learning, the system tries to learn from the previous examples that are given. So if the dataset is labelled it comes under a supervised problem, it the dataset is unlabelled then it is an unsupervised problem. The image to the left is an example of supervised learning; we use regression techniques to find the best fit line between the features.

Several fundamental problems in robotics, including state estimation, prediction, and motion planning rely on accurate models that can map state to measurements (forward models) or measurements to state (inverse models). Classic examples include the measurement models for global positioning systems, inertial measurement units, or beam sensors that are frequently used in simultaneous localization and mapping [1], or the forward and inverse kinematic models that map joint configurations to workspace and vice-versa. Some of these models can be very difficult to derive analytically, and, in these cases, roboticists have often resorted to machine learning to infer accurate models directly from data. For example, complex nonlinear forward kinematics have been modeled with techniques as diverse as Bayesian networks [2] and Bezier Splines [3], and many researchers have tackled the problem of learning inverse kinematics with nonparametric methods like locally weighted projection regression (LWPR) [4], [5], mixtures of experts [6], and Gaussian Process Regression [7]. While these techniques are able to learn accurate models, they rely heavily on prior knowledge about the kinematic relationship between the robot statespace and workspace. Despite the important role that forward and inverse models have played in robotics, there has been little progress in defining these models for very high-dimensional sensor data like images and video.

We develop a general method for estimating a finite mixture of non-normalized models. Here, a non-normalized model is defined to be a parametric distribution with an intractable normalization constant. Existing methods for estimating non-normalized models without computing the normalization constant are not applicable to mixture models because they contain more than one intractable normalization constant. The proposed method is derived by extending noise contrastive estimation (NCE), which estimates non-normalized models by discriminating between the observed data and some artificially generated noise. We also propose an extension of NCE with multiple noise distributions. Then, based on the observation that conventional classification learning with neural networks is implicitly assuming an exponential family as a generative model, we introduce a method for clustering unlabeled data by estimating a finite mixture of distributions in an exponential family. Estimation of this mixture model is attained by the proposed extensions of NCE where the training data of neural networks are used as noise. Thus, the proposed method provides a probabilistically principled clustering method that is able to utilize a deep representation. Application to image clustering using a deep neural network gives promising results.

Mirek Olšák Charles University We introduce a theorem proving algorithm that uses practically no domain heuristics for guiding its connection-style proof search. Instead, it runs many Monte-Carlo simulations guided by reinforcement learning from previous proof attempts. We produce several versions of the prover, parameterized by different learning and guiding algorithms. The strongest version of the system is trained on a large corpus of mathematical problems and evaluated on previously unseen problems. The trained system solves within the same number of inferences over 40% more problems than a baseline prover, which is an unusually high improvement in this hard AI domain. To our knowledge this is the first time reinforcement learning has been convincingly applied to solving general mathematical problems on a large scale.

This work includes a number of novel contributions for the multiple-source adaptation problem. We present new normalized solutions with strong theoretical guarantees for the cross-entropy loss and other similar losses. We also provide new guarantees that hold in the case where the conditional probabilities for the source domains are distinct. Moreover, we give new algorithms for determining the distribution-weighted combination solution for the cross-entropy loss and other losses. We report the results of a series of experiments with real-world datasets. We find that our algorithm outperforms competing approaches by producing a single robust model that performs well on any target mixture distribution. Altogether, our theory, algorithms, and empirical results provide a full solution for the multiple-source adaptation problem with very practical benefits.