We develop a functional learning approach to modelling systems of time series which preserves the ability of standard linear time-series models (VARs) to uncover the Granger-causality links in between the series of the system while allowing for richer functional relationships. We propose a framework for learning multiple output-kernels associated with multiple input-kernels over a structured input space and outline an algorithm for simultaneous learning of the kernels with the model parameters with various forms of regularization including non-smooth sparsity inducing norms. We present results of synthetic experiments illustrating the benefits of the described approach.
This study aims to examine the impact of an Integrated Financial System (ERP system) implementation on accountant profiles in Swiss public administrations. ERP systems are widely researched by authors and are described as a form of change driver. Conversely, the types of changes and, more specifically, the consequences on accountants’ profiles are not studied and are therefore the core of our investigation. The methodology used in this study was based on the completion of a large survey allowing for statistical analysis and is focused on the necessary knowledge and skill sets of accountants working with an ERP system. In this survey, we brought to light new information about the current skill sets needed by accountants using an ERP system to improve the work of this profession and, therefore, to enhance the performance of finance and accounting staff and the quality of information supplied by public administrators. The results allowed us to design the profile of an accountant working with an ERP system in the public sector. In particular, the study examined knowledge and skill sets, as well as educational background and professional experience. Moreover, the criteria that impacted the ERP system users’ satisfaction were identified, and these findings especially provided practical implications for public sector CFOs. Finally, we highlighted the crucial need for continuous education in accounting and the necessity to reconsider and adapt the job descriptions of accounting and finance staff when they work with an ERP system.
We investigated two strategies for improving Information Retrieval thanks to incoming and outgoing citations. We first started from settings that worked last year and established a baseline. Then, we tried to rerank this run. The incoming citations’ strategy was to compute the number of incoming citations in PubMed Central, and to boost the score of the articles that were the most cited. The outgoing citations’ strategy was to promote the references of the retrieved documents. Unfortunately, no significant improvement from the baseline was observed.
The need for domain knowledge representation for program comprehension is now widely accepted in the program comprehension community. The so-called "concept assignment problem" represents the challenge to locate domain concepts in the source code of programs. The vast majority of attempts to solve it are based on static source code search for clues to domain concepts. In contrast, our approach is based on dynamic analysis using information retrieval (IR) metrics. First we explain how we modeled the domain concepts and their role in program comprehension. Next we present how some of the popular IR metrics could be adapted to the "concept assignment problem" and the way we implemented the search engine. Then we present our own metric and the performance of these metrics to retrieve domain concepts in source code. The contribution of the paper is to show how the IR metrics could be applied to the "concept assignment problem" when the "documents" to retrieve are domain concepts structured in an ontology.
The CLEF RepLab 2014 Track was the occasion to investigate the robustness of instance-based learning in a complete system for tweet monitoring and categorization based. The algorithm we implemented was a k-Nearest Neighbors. Dealing with the domain (automotive or banking) and the language (English or Spanish), the experiments showed that the categorizer was not affected by the choice of representation: even with all learning tweets merged into one single Knowledge Base (KB), the observed performances were close to those with dedicated KBs. Interestingly, English training data in addition to the sparse Spanish data were useful for Spanish categorization (+14% for accuracy for automotive, +26% for banking). Yet, performances suffered from an overprediction of the most prevalent category. The algorithm showed the defects of its virtues: it was very robust, but not easy to improve. BiTeM/SIBtex tools for tweet monitoring are available within the DrugsListener Project page of the BiTeM website (http://bitem.hesge.ch/).
We consider the problem of forecasting multiple time series across multiple cross-sections based solely on the past observations of the series. We propose to use panel vector autoregressive model to capture the inter-dependencies on the past values of the multiple series. We restrict the panel vector autoregressive model to exclude the cross-sectional relationships and propose a method to learn models with sparse Granger-causality structures coherent across the panel sections. The method extends the concepts of group variable selection and support union recovery into the panel setting by extending the group lasso penalty (Yuan & Lin, 2006) into matrix output regression setting with 3d-tensor of model parameters.
We study parametric unsupervised mixture learning. We measure the loss of intrinsic information from the observations to complex mixture models, and then to simple mixture models. We present a geometric picture, where all these representations are regarded as free points in the space of probability distributions. Based on minimum description length, we derive a simple geometric principle to learn all these models together. We present a new learning machine with theories, algorithms, and simulations.
In this study, eye tracking and mouse tracking data collected from two Swiss digital library web sites are compared, with respect to their specific areas of interest in order to answer two questions: Firstly, to know, how far the perception of the corresponding areas of interests differed from site to site and how far general recommendations can be inferred from this comparison. Secondly, the dispersion on the gaze and the mouse click plots were compared with the results of the two methods with each other to see if one method can be replaced by each other or if both methods should rather be considered as complementary. The results show that especially the choice of color and the use of contrast strongly influence gazes and clicks and that some areas of interest mainly attract views, but not clicks and vice versa, which leads to a complementary distribution pattern, and makes the question of replacing one method by the other obsolete.
In this paper we present reflections and recommendations concerning the conversion of library metadata into Linked Data. We will briefly describe the different data models that exist for this purpose and argue their strengths and weaknesses with a special focus on the entification issue, before illustrating this problem with the description of three different ongoing projects. As will be outlined afterwards, it is essential to distinguish at the start of a project between data-driven or user-driven design approaches. As an alternative, the realization of a Linked Data project for bibliographical data might also lead to a hybrid approach were the design process shifts reciprocally between the analysis of the data and the user needs.
