Purpose: Malicious web domain identification is of significant importance to the security protection of Internet users. With online credibility and performance data, this paper aims to investigate the use of machine learning tech-niques for malicious web domain identification by considering the class imbalance issue (i.e., there are more benign web domains than malicious ones). Design/methodology/approach: We propose an integrated resampling approach to handle class imbalance by combining the Synthetic Minority Over-sampling TEchnique (SMOTE) and Particle Swarm Optimisation (PSO), a population-based meta-heuristic algorithm. We use the SMOTE for over-sampling and PSO for under-sampling. Findings: By applying eight well-known machine learning classifiers, the proposed integrated resampling approach is comprehensively examined using several imbalanced web domain datasets with different imbalance ratios. Com-pared to five other well-known resampling approaches, experimental results confirm that the proposed approach is highly effective. Practical implications: This study not only inspires the practical use of online credibility and performance data for identifying malicious web domains, but also provides an effective resampling approach for handling the class imbal-ance issue in the area of malicious web domain identification. Originality/value: Online credibility and performance data is applied to build malicious web domain identification models using machine learning techniques. An integrated resampling approach is proposed to address the class im-balance issue. The performance of the proposed approach is confirmed based on real-world datasets with different imbalance ratios.

Unnormalised latent variable models are a broad and flexible class of statistical models. However, learning their parameters from data is intractable, and few estimation techniques are currently available for such models. To increase the number of techniques in our arsenal, we propose variational noise-contrastive estimation (VNCE), building on NCE which is a method that only applies to unnormalised models. The core idea is to use a variational lower bound to the NCE objective function, which can be optimised in the same fashion as the evidence lower bound (ELBO) in standard variational inference (VI). We prove that VNCE can be used for both parameter estimation of unnormalised models and posterior inference of latent variables. The developed theory shows that VNCE has the same level of generality as standard VI, meaning that advances made there can be directly imported to the unnormalised setting. We validate VNCE on toy models and apply it to a realistic problem of estimating an undirected graphical model from incomplete data.

Instance embeddings are an efficient and versatile image representation that facilitates applications like recognition, verification, retrieval, and clustering. Many metric learning methods represent the input as a single point in the embedding space. Often the distance between points is used as a proxy for match confidence. However, this can fail to represent uncertainty which can arise when the input is ambiguous, e.g., due to occlusion or blurriness. We introduce the hedged instance embedding (HIB) in which embeddings are modeled as random variables and the model is trained under the variational information bottleneck principle (Alemi et al., 2016; Achille & Soatto, 2018). Empirical results on our new N-digit MNIST dataset show that our method leads to the desired behavior of "hedging its bets" across the embedding space upon encountering ambiguous inputs. This results in improved performance for image matching and classification tasks, more structure in the learned embedding space, and an ability to compute a per-exemplar uncertainty measure which is correlated with downstream performance. Embeddings are a versatile representation that support various downstream tasks, including image retrieval (Babenko et al., 2014) and face recognition (Schroff et al., 2015). Instance embeddings are often treated deterministically, i.e., z f(x) is a point in R One drawback of this representation is the difficulty of modeling aleatoric uncertainty (Kendall & Gal, 2017), i.e. uncertainty induced by the input.

We show that there are several regularized loss minimization problems that can use locally perturbed data with theoretical guarantees of generalization, i.e., loss consistency. Our results quantitatively connect the convergence rates of the learning problems to the impossibility for any adversary for recovering the original data from perturbed observations. To this end, we introduce a new concept of data irrecoverability, and show that the well-studied concept of data privacy implies data irrecoverability.

Sigmoid: tend to vanish gradient (cause there is a mechanism to reduce the gradient as "a" increases, where "a" is the input of a sigmoid function. When "a" grows to infinite large, S′(a) S(a)(1 S(a)) 1 (1 1) 0.

Sigmoid: tend to vanish gradient (cause there is a mechanism to reduce the gradient as "a" increases, where "a" is the input of a sigmoid function. When "a" grows to infinite large, S′(a) S(a)(1 S(a)) 1 (1 1) 0.

In this paper, we provide an overview of first-order and second-order variants of the gradient descent methods commonly used in machine learning. We propose a general framework in which 6 of these methods can be interpreted as different instances of the same approach. These methods are the vanilla gradient descent, the classical and generalized Gauss-Newton methods, the natural gradient descent method, the gradient covariance matrix approach, and Newton's method. Besides interpreting these methods within a single framework, we explain their specificities and show under which conditions some of them coincide. Machine learning generally amounts to solving an optimization problem where a loss function has to be minimized.

Principal component analysis (PCA) and singular value decomposition (SVD) are widely used in statistics, machine learning, and applied mathematics. It has been well studied in the case of homoskedastic noise, where the noise levels of the contamination are homogeneous. In this paper, we consider PCA and SVD in the presence of heteroskedastic noise, which arises naturally in a range of applications. We introduce a general framework for heteroskedastic PCA and propose an algorithm called HeteroPCA, which involves iteratively imputing the diagonal entries to remove the bias due to heteroskedasticity. This procedure is computationally efficient and provably optimal under the generalized spiked covariance model. A key technical step is a deterministic robust perturbation analysis on the singular subspace, which can be of independent interest. The effectiveness of the proposed algorithm is demonstrated in a suite of applications, including heteroskedastic low-rank matrix denoising, Poisson PCA, and SVD based on heteroskedastic and incomplete data.

Large-scale machine learning training, in particular distributed stochastic gradient descent, needs to be robust to inherent system variability such as node straggling and random communication delays. This work considers a distributed training framework where each worker node is allowed to perform local model updates and the resulting models are averaged periodically. We analyze the true speed of error convergence with respect to wall-clock time (instead of the number of iterations), and analyze how it is affected by the frequency of averaging. Stochastic gradient descent (SGD) is the backbone of stateof-the-art supervised learning, which is revolutionizing inference and decision-making in many diverse applications. Classical SGD was designed to be run on a single computing node, and its error-convergence with respect to the number of iterations has been extensively analyzed and improved via accelerated SGD methods. Due to the massive training data-sets and neural network architectures used today, it has became imperative to design distributed SGD implementations, where gradient computation and aggregation is parallelized across multiple worker nodes. Although parallelism boosts the amount of data processed per iteration, it exposes SGD to unpredictable node slowdown and communication delays stemming from variability in the computing infrastructure. Thus, there is a critical need to make distributed SGD fast, yet robust to system variability.

Computer algorithms are written with the intent that when run they perform a useful function. Typically any information obtained is unknown until the algorithm is run. However, if the behavior of an algorithm can be fully described by precomputing just once how this algorithm will respond when executed on any input, this precomputed result provides a complete specification for all solutions in the problem domain. We apply this idea to a previous anomaly detection algorithm, and in doing so transform it from one that merely detects individual anomalies when asked to discover potentially anomalous values, into an algorithm also capable of generating a complete specification for those values it would deem to be anomalous. This specification is derived by examining no more than a small training data, can be obtained in very small constant time, and is inherently far more useful than results obtained by repeated execution of this tool. For example, armed with such a specification one can ask how close an anomaly is to being deemed normal, and can validate this answer not by exhaustively testing the algorithm but by examining if the specification so generated is indeed correct. This powerful idea can be applied to any algorithm whose runtime behavior can be recovered from its construction and so has wide applicability.