We propose an approach to remodelling classical planning domains via the addition of macro operators and removal of original operators either for the domain as a whole or instance-by-instance. For the latter remodelling, we train a predictor to choose the best reformulation of the domain based on instance characteristic. In the domain level remodelling, we try find a fixed remodelling that works best on average over our training set. Operator removal does not generally preserve solubility and proving solubility preservation of domain models is PSPACE-complete. So we use an approach that uses training instances to empirically estimate the probability of solubility preservation and maintains a minimum value of that probability on the training instances. We show that the instance-specific approach outperforms the traditional best-on-average macro-only remodelling approach in 9 out of 14 cases of domain/macro-source combinations, and that it can outperform fixed domain-based models generated with existing macro learning tools.

Gaining insight in a web user's personality is very valuable for applications that rely on personalisation, such as recommender systems and personalised advertising. In this paper we explore the use of machine learning techniques for inferring a user's personality traits from their Facebook status updates. Even with a small set of training examples we can outperform the majority class baseline algorithm. Furthermore, the results are improved by adding training examples from another source. This is an interesting result because it indicates that personality trait recognition generalises across social media platforms.

Boosting methods are highly popular and effective supervised learning methods which combine weak learners into a single accurate model with good statistical performance. In this paper, we analyze two well-known boosting methods, AdaBoost and Incremental Forward Stagewise Regression (FS$_\varepsilon$), by establishing their precise connections to the Mirror Descent algorithm, which is a first-order method in convex optimization. As a consequence of these connections we obtain novel computational guarantees for these boosting methods. In particular, we characterize convergence bounds of AdaBoost, related to both the margin and log-exponential loss function, for any step-size sequence. Furthermore, this paper presents, for the first time, precise computational complexity results for FS$_\varepsilon$.

We propose a sampling scheme suitable for reducing a data set prior to selecting a hypothesis with minimum empirical risk. The sampling only considers a subset of the ultimate (unknown) hypothesis set, but can nonetheless guarantee that the final excess risk will compare favorably with utilizing the entire original data set. We demonstrate the practical benefits of our approach on a large dataset which we subsample and subsequently fit with boosted trees.

This work proposes a way to align statistical modeling with decision making. We provide a method that propagates the uncertainty in predictive modeling to the uncertainty in operational cost, where operational cost is the amount spent by the practitioner in solving the problem. The method allows us to explore the range of operational costs associated with the set of reasonable statistical models, so as to provide a useful way for practitioners to understand uncertainty. To do this, the operational cost is cast as a regularization term in a learning algorithm's objective function, allowing either an optimistic or pessimistic view of possible costs, depending on the regularization parameter. From another perspective, if we have prior knowledge about the operational cost, for instance that it should be low, this knowledge can help to restrict the hypothesis space, and can help with generalization. We provide a theoretical generalization bound for this scenario. We also show that learning with operational costs is related to robust optimization.

In domains like bioinformatics, information retrieval and social network analysis, one can find learning tasks where the goal consists of inferring a ranking of objects, conditioned on a particular target object. We present a general kernel framework for learning conditional rankings from various types of relational data, where rankings can be conditioned on unseen data objects. We propose efficient algorithms for conditional ranking by optimizing squared regression and ranking loss functions. We show theoretically, that learning with the ranking loss is likely to generalize better than with the regression loss. Further, we prove that symmetry or reciprocity properties of relations can be efficiently enforced in the learned models. Experiments on synthetic and real-world data illustrate that the proposed methods deliver state-of-the-art performance in terms of predictive power and computational efficiency. Moreover, we also show empirically that incorporating symmetry or reciprocity properties can improve the generalization performance.

Author identification is a subfield of Natural Language Processing (NLP) that uses machine learning techniques to identify the author of a text. Most previous research focused on long texts with the assumption that a minimum text length threshold exists under which author identification would no longer be effective. This paper examines author identification in short texts far below this threshold, focusing on messages retrieved from Twitter (maximum length: 140 characters) to determine the most effective feature set for author identification. Both Bag-of-Words (BOW) and Style Marker feature sets were extracted and evaluated through a series of 15 experiments involving up to 12 authors with large and small dataset sizes. Support Vector Machines (SVM) were used for all experiments. Our results achieve classification accuracies approaching that of longer texts, even for small dataset sizes of 60 training instances per author. Style Marker feature sets were found to be significantly more useful than BOW feature sets as well as orders of magnitude faster, and are therefore suggested for potential applications in future research.

We study high-dimensional asymptotic performance limits of binary supervised classification problems where the class conditional densities are Gaussian with unknown means and covariances and the number of signal dimensions scales faster than the number of labeled training samples. We show that the Bayes error, namely the minimum attainable error probability with complete distributional knowledge and equally likely classes, can be arbitrarily close to zero and yet the limiting minimax error probability of every supervised learning algorithm is no better than a random coin toss. In contrast to related studies where the classification difficulty (Bayes error) is made to vanish, we hold it constant when taking high-dimensional limits. In contrast to VC-dimension based minimax lower bounds that consider the worst case error probability over all distributions that have a fixed Bayes error, our worst case is over the family of Gaussian distributions with constant Bayes error. We also show that a nontrivial asymptotic minimax error probability can only be attained for parametric subsets of zero measure (in a suitable measure space). These results expose the fundamental importance of prior knowledge and suggest that unless we impose strong structural constraints, such as sparsity, on the parametric space, supervised learning may be ineffective in high dimensional small sample settings.

State-of-the-art link prediction utilizes combinations of complex features derived from network panel data. We here show that computationally less expensive features can achieve the same performance in the common scenario in which the data is available as a sequence of interactions. Our features are based on social vector clocks, an adaptation of the vector-clock concept introduced in distributed computing to social interaction networks. In fact, our experiments suggest that by taking into account the order and spacing of interactions, social vector clocks exploit different aspects of link formation so that their combination with previous approaches yields the most accurate predictor to date.

In machine learning, the choice of a learning algorithm that is suitable for the application domain is critical. The performance metric used to compare different algorithms must also reflect the concerns of users in the application domain under consideration. In this work, we propose a novel probability-based performance metric called Relevance Score for evaluating supervised learning algorithms. We evaluate the proposed metric through empirical analysis on a dataset gathered from an intelligent lighting pilot installation. In comparison to the commonly used Classification Accuracy metric, the Relevance Score proves to be more appropriate for a certain class of applications.