A few weeks ago, I wrote about how and why I was learning Machine Learning, mainly through Andrew Ng's Coursera course. Machine Learning is built on prerequisites, so much so that learning by first principles seems overwhelming. Do you really need to spend a month learning linear algebra? You'll be okay if you have some math and programming experience. You really just have to be familiar with Sigma notation and be able to express it in a for loop. Sure, your assignments will take longer to complete and the first few times you see those giant equations your head will spin, but you can do this! Calculus is not even required.

Typed model counting expands model counting of propositional formulas by the ability to distinguish between certain types of models. Formally, we incorporate elements of a commutative monoid that represent these model types directly into the propositional formulas. An advantage of this approach is the ability of preserving information about which parts of a formula are satisfied by a certain type of model. We exploit this benefit when applying typed model counting to probabilistic conditional reasoning at maximum entropy. In particular, we address the task of determining the conditional structure induced by a reasoner’s probabilistic conditional knowledge base in order to draw nonmonotonic inferences based on the maximum entropy distribution.

Word Sense Disambiguation (WSD) is a classification task that consists of determining which of the senses of an ambiguous word is activated in a specific context. Research in this field has primarily concentrated on investigating English and a few other well-resourced languages. Recently, studies done on a corpus of Old English (Wunderlich 2015) showed that, even with limited resources, it is still possible to approach the problem of WSD. In this paper, a WSD system has been developed for the Low Resource Language (LRL) Venetan, which has recently received some attention from the Natural Language Processing (NLP) community. Our main contributions are twofold: first, we select and annotate a corpus for Venetan, considering two words (one abstract and one concrete term) and using two levels of annotation (fine- and coarse-grained), reporting on annotator agreement. Second, we report results of proof-of-concept experiments of supervised WSD performed with Support Vector Machines on this corpus. To our knowledge, our work is the first time that WSD for a European Dialect like Venetan has been studied.

In this paper, we investigate the authorship of several short historical texts that are written by ten ancient Arabic travelers: this Arabic dataset, which was collected by the authors in 2011, and called AAAT (Authorship attribution of Ancient Arabic Texts) corpus, is considered as a reference dataset in Arabic. Several experiments of authorship attribution are conducted by using different features namely: characters, character n-grams, and lexical features such as words, word n-grams, and rare words. On the other hand, different classifiers are employed, such as: statistical distances, Multi Layer Percep-tron (MLP), Support Vector Machines (SVM) and Linear Regression (LR). In this investigation, a new fusion technique is proposed to enhance the overall performances of the classifiers: it is called Score Based Fusion (SBF). Results show good attribution performances with an optimal score between 80% and 90% of good authorship attribution. The proposed fusion technique raised this score to 100% of good authorship attribution. Moreover, this comparative survey has revealed interesting results concerning the Arabic language and more particularly with short texts.

The multilabel learning problem with large number of labels, features, and data-points has generated a tremendous interest recently. A recurring theme of these problems is that only a few labels are active in any given data point as compared to the total number of labels. However, only a small number of existing work take direct advantage of this inherent extreme sparsity in the label space. By the virtue of Restricted Isometry Property (RIP), satisfied by many random ensembles, we propose a novel procedure for multilabel learning known as RIPML. During the training phase, in RIPML, labels are projected onto a random low-dimensional subspace followed by solving a least-square problem in this subspace. Inference is done by a k-nearest neighbor (kNN) based approach. We demonstrate the effectiveness of RIPML by conducting extensive simulations and comparing results with the state-of-the-art linear dimensionality reduction based approaches.

Demand forecasting is an important challenge for industries seeking to optimize service quality and expenditures. Generating accurate forecasts is difficult because it depends on the quality of the data available to train predictive models, as well as on the model chosen for the task. We evaluate the approach on two datasets of varying complexity and compare the results with three machine learning algorithms. Results show our approach can outperform these approaches.

Millions of users use online, open-ended blocks programming environments like App Inventor to learn how to program and to build personally meaningful programs and apps. As part of understanding the computational thinking concepts being learned by these users, we want to distinguish original projects that they create from unoriginal ones that arise from learning activities like tutorials and exercises. Given all the projects of students taking an App Inventor course, we describe how to automatically classify them as original vs. unoriginal using a hierarchical clustering technique. Although our current analysis focuses only on a small group of users (16 students taking a course in our institution) and their 902 projects, our findings establish a foundation for extending this analysis to larger groups of users.

Anomalies in healthcare claims data can be indicative of possible fraudulent activities, contributing to a significant portion of overall healthcare costs. Medicare is a large government run healthcare program that serves the needs of the elderly in the United States. The increasing elderly population and their reliance on the Medicare program create an environment with rising costs and increased risk of fraud. The detection of these potentially fraudulent activities can recover costs and lessen the overall impact of fraud on the Medicare program. In this paper, we propose a new method to detect fraud by discovering outliers, or anomalies, in payments made to Medicare providers. We employ a multivariate outlier detection method split into two parts. In the first part, we create a multivariate regression model and generate corresponding residuals. In the second part, these residuals are used as inputs into a generalizable univariate probability model. We create this Bayesian probability model using probabilistic programming. Our results indicate our model is robust and less dependent on underlying data distributions, versus Mahalanobis distance. Moreover, we are able to demonstrate successful anomaly detection, within Medicare specialties, providing meaningful results for further investigation.

Textual queries in networks such as Twitter can have more than one label resulting in a multi-label classification problem. To reduce computational costs a low-dimensional representation of a large network is learned that preserves proximity among nodes in the same community. Similar to sequence of words in a sentence, DeepWalk consider sequence of nodes in a shallow graph and clustering is done using hierarchical softmax in an unsupervised manner. In this paper, we generate network abstractions at different levels using deep convolutional neural networks. Since, class labels of connected nodes in a network keep changing we consider a fuzzy recurrent feedback controller to ensure robustness to noise.

Distributed Denial of Service (DDoS) attacks are a major threat to Internet security, with their use continuing to grow. Attackers are finding more sophisticated methods to attack servers. A lot of defense mechanisms have been proposed for DDoS attacks at IP and TCP layers. Those methods will not work well for application layer DDoS attacks that utilize legitimate application layer requests to overwhelm a webserver. These attacks look legitimate in both packets and protocol characteristics, which makes them harder to detect. In this paper, we propose an anomaly detection method to detect application layer DDoS attacks. We take a text mining approach to extract features which represent a user’s HTTP request sequence using bigrams. We apply the one class Support Vector Machine (SVM) algorithm on the extracted features from normal users’ HTTP request sequences. The one class SVM labels any newly seen instance that deviates from the normal, trained model as an application layer DDoS instance. We apply our experimental analysis on real web server logs collected from a student resource website. Three different variants of HTTP GET flood attacks are implemented on our server, generated via penetration testing. Our results show that the proposed method is able to detect application layer DDoS attacks with very good performance results.