Deep learning has made significant advances in creating efficient representations of time series data by automatically identifying complex patterns. However, these approaches lack interpretability, as the time series is transformed into a latent vector that is not easily interpretable. On the other hand, Symbolic Aggregate approximation (SAX) methods allow the creation of symbolic representations that can be interpreted but do not capture complex patterns effectively. In this work, we propose a set of requirements for a neural representation of univariate time series to be interpretable. We propose a new unsupervised neural architecture that meets these requirements. The proposed model produces consistent, discrete, interpretable, and visualizable representations. The model is learned independently of any downstream tasks in an unsupervised setting to ensure robustness. As a demonstration of the effectiveness of the proposed model, we propose experiments on classification tasks using UCR archive datasets. The obtained results are extensively compared to other interpretable models and state-of the-art neural representation learning models. The experiments show that the proposed model yields, on average better results than other interpretable approaches on multiple datasets. We also present qualitative experiments to asses the interpretability of the approach.

Data-driven approaches to modeling physical systems fail to generalize to unseen systems that share the same general dynamics with the learning domain, but correspond to different physical contexts. We propose a new framework for this key problem, context-informed dynamics adaptation (CoDA), which takes into account the distributional shift across systems for fast and efficient adaptation to new dynamics. CoDA leverages multiple environments, each associated to a different dynamic, and learns to condition the dynamics model on contextual parameters, specific to each environment. The conditioning is performed via a hypernetwork, learned jointly with a context vector from observed data. The proposed formulation constrains the search hypothesis space for fast adaptation and better generalization across environments with few samples. We theoretically motivate our approach and show state-of-the-art generalization results on a set of nonlinear dynamics, representative of a variety of application domains. We also show, on these systems, that new system parameters can be inferred from context vectors with minimal supervision.

When modeling dynamical systems from real-world data samples, the distribution of data often changes according to the environment in which they are captured, and the dynamics of the system itself vary from one environment to another. Generalizing across environments thus challenges the conventional frameworks. The classical settings suggest either considering data as i.i.d and learning a single model to cover all situations or learning environment-specific models. Both are sub-optimal: the former disregards the discrepancies between environments leading to biased solutions, while the latter does not exploit their potential commonalities and is prone to scarcity problems. We propose LEADS, a novel framework that leverages the commonalities and discrepancies among known environments to improve model generalization. This is achieved with a tailored training formulation aiming at capturing common dynamics within a shared model while additional terms capture environment-specific dynamics. We ground our approach in theory, exhibiting a decrease in sample complexity w.r.t classical alternatives. We show how theory and practice coincides on the simplified case of linear dynamics. Moreover, we instantiate this framework for neural networks and evaluate it experimentally on representative families of nonlinear dynamics. We show that this new setting can exploit knowledge extracted from environment-dependent data and improves generalization for both known and novel environments.

Extreme classification task where the number of classes is very large has received important focus over the last decade. Usual efficient multi-class classification approaches have not been designed to deal with such large number of classes. A particular issue in the context of large scale problems concerns the computational classification complexity : best multi-class approaches have generally a linear complexity with respect to the number of classes which does not allow these approaches to scale up. Recent works have put their focus on using hierarchical classification process in order to speed-up the classification of new instances. Using a priori information on labels such as a label hierarchy allows to build an efficient hierarchical structure over the labels in order to decrease logarithmically the classification time. However such information on labels is not always available nor useful. Finding a suitable hierarchical organization of the labels is thus a crucial issue as the accuracy of the model depends highly on the label assignment through the label tree. We propose in this work a new algorithm to build iteratively a hierarchical label structure by proposing a partitioning algorithm which optimizes simultaneously the structure in terms of classification complexity and the label partitioning problem in order to achieve high classification performances. Beginning from a flat tree structure, our algorithm selects iteratively a node to expand by adding a new level of nodes between the considered node and its children. This operation increases the speed-up of the classification process. Once the node is selected, best partitioning of the classes has to be computed. We propose to consider a measure based on the maximization of the expected loss of the sub-levels in order to minimize the global error of the structure. This choice enforces hardly separable classes to be group together in same partitions at the first levels of the tree structure and it delays errors at a deep level of the structure where there is no incidence on the accuracy of other classes. Experiments on real big text data from recent challenge assess the performances of our model.

BACKGROUND: This article provides an overview of the first BIOASQ challenge, a competition on large-scale biomedical semantic indexing and question answering (QA), which took place between March and September 2013. BIOASQ assesses the ability of systems to semantically index very large numbers of biomedical scientific articles, and to return concise and user-understandable answers to given natural language questions by combining information from biomedical articles and ontologies. RESULTS: The 2013 BIOASQ competition comprised two tasks, Task 1a and Task 1b. In Task 1a participants were asked to automatically annotate new PUBMED documents with MESH headings. Twelve teams participated in Task 1a, with a total of 46 system runs submitted, and one of the teams performing consistently better than the MTI indexer used by NLM to suggest MESH headings to curators. Task 1b used benchmark datasets containing 29 development and 282 test English questions, along with gold standard (reference) answers, prepared by a team of biomedical experts from around Europe and participants had to automatically produce answers. Three teams participated in Task 1b, with 11 system runs. The BIOASQ infrastructure, including benchmark datasets, evaluation mechanisms, and the results of the participants and baseline methods, is publicly available. CONCLUSIONS: A publicly available evaluation infrastructure for biomedical semantic indexing and QA has been developed, which includes benchmark datasets, and can be used to evaluate systems that: assign MESH headings to published articles or to English questions; retrieve relevant RDF triples from ontologies, relevant articles and snippets from PUBMED Central; produce "exact" and paragraph-sized "ideal" answers (summaries). The results of the systems that participated in the 2013 BIOASQ competition are promising. In Task 1a one of the systems performed consistently better from the NLM's MTI indexer. In Task 1b the systems received high scores in the manual evaluation of the "ideal" answers; hence, they produced high quality summaries as answers. Overall, BIOASQ helped obtain a unified view of how techniques from text classification, semantic indexing, document and passage retrieval, question answering, and text summarization can be combined to allow biomedical experts to obtain concise, user-understandable answers to questions reflecting their real information needs.

Behavior Model Inference techniques aim at mining behavior models from execution traces. While most of approaches usually ground on local similarities in traces, recent work, referred to as behavior mining with temporal steering, propose to include long term dependencies in the mining process. Such dependencies correspond to temporal implications between events in execution traces, whose consideration allows to ensure a better consistency of the extracted model. Nevertheless, the existing approaches are usually limited by their high computational complexity and the approximations to reduce the cost of temporal rules checking. This paper revisits behavior mining with temporal steering by defining an efficient algorithm that performs an exact consideration of the observed dependencies: in our experiments, greatly reduced processing times (from exponential to quasi-linear) for exact mining with temporal steering have been observed. Furthermore, beyond highlighting the great benefits of considering temporal dependencies, this paper also proposes new key extensions to the existing work that allow to include more complex dependencies in the mining process. Intensive evaluation finally demonstrates the great performances of the proposed approach.

Reverse engineering of behavior models has received a lot of attention over the last few years. However, no standard benchmark exists for the comparison and analysis of published miners. Evaluation is usually performed on few case studies, which fails to demonstrate effectiveness in a broad context. This paper proposes a general, approach-independent, platform for the intensive evaluation of behavior miners. Its goals are essentially: provide a benchmark mechanism for reverse engineering; allow analysis of miners w.r.t. a class of programs and/or behaviors; help users in choosing the best suited approach for their objective.

We consider the problem of discovering links of an evolving undirected graph given a series of past snapshots of that graph. The graph is observed through the time sequence of its adjacency matrix and only the presence of edges is observed. The absence of an edge on a certain snapshot cannot be distinguished from a missing entry in the adjacency matrix. Additional information can be provided by examining the dynamics of the graph through a set of topological features, such as the degrees of the vertices. We develop a novel methodology by building on both static matrix completion methods and the estimation of the future state of relevant graph features. Our procedure relies on the formulation of an optimization problem which can be approximately solved by a fast alternating linearized algorithm whose properties are examined. We show experiments with both simulated and real data which reveal the interest of our methodology.

How to determine a priori whether a learning algorithm is suited to a learning problem instance is a major scientific and technological challenge. A first step toward this goal, inspired by the Phase Transition (PT) paradigm developed in the Constraint Satisfaction domain, is presented in this paper.Based on the PT paradigm, extensive and principled experiments allow for constructing the Competence Map associated to a learning algorithm, describing the regions where this algorithm on average fails or succeeds. The approach is illustrated on the long and widely used C4.5 algorithm. A non trivial failure region in the landscape of k-term DNF languages is observed and some interpretations are offered for the experimental results.