Showing papers by "Mines ParisTech published in 2005"

Standardizing the Performance Evaluation of Short-Term Wind Power Prediction Models:

Abstract: Short-term wind power prediction is a primary requirement for efficient large-scale integration of wind generation in power systems and electricity markets. The choice of an appropriate prediction model among the numerous available models is not trivial, and has to be based on an objective evaluation of model performance.This paper proposes a standardized protocol for the evaluation of short-term wind-power prediction systems. A number of reference prediction models are also described, and their use for performance comparison is analysed. The use of the protocol is demonstrated, using results from both on-shore and offshore wind farms. The work was developed in the frame of the Anemos project (EU R&D project) where the protocol has been used to evaluate more than 10 prediction systems.

Antileakage Fourier transform for seismic data regularization

Abstract: Seismic data regularization, which spatially transforms irregularly sampled acquired data to regularly sampled data, is a long-standing problem in seismic data processing. Data regularization can be implemented using Fourier theory by using a method that estimates the spatial frequency content on an irregularly sampled grid. The data can then be reconstructed on any desired grid. Difficulties arise from the nonorthogonality of the global Fourier basis functions on an irregular grid, which results in the problem of “spectral leakage”: energy from one Fourier coefficient leaks onto others. We investigate the nonorthogonality of the Fourier basis on an irregularly sampled grid and propose a technique called “antileakage Fourier transform” to overcome the spectral leakage. In the antileakage Fourier transform, we first solve for the most energetic Fourier coefficient, assuming that it causes the most severe leakage. To attenuate all aliases and the leakage of this component onto other Fourier coefficients, the data component corresponding to this most energetic Fourier coefficient is subtracted from the original input on the irregular grid. We then use this new input to solve for the next Fourier coefficient, repeating the procedure until all Fourier coefficients are estimated. This procedure is equivalent to “reorthogonalizing” the global Fourier basis on an irregularly sampled grid. We demonstrate the robustness and effectiveness of this technique with successful applications to both synthetic and real data examples.

Safe motion planning in dynamic environments

Abstract: This paper addresses the problem of motion planning (MP) in dynamic environments. It is first argued that dynamic environments impose a real-time constraint upon MP: it has a limited time only to compute a motion, the time available being a function of the dynamicity of the environment. Now, given the intrinsic complexity of MP, computing a complete motion to the goal within the time available is impossible to achieve in most real situations. Partial motion planning (PMP) is the answer to this problem proposed in this paper. PMP is a motion planning scheme with an anytime flavor: when the time available is over, PMP returns the best partial motion to the goal computed so far. Like reactive navigation scheme, PMP faces a safety issue: what guarantee is there that the system will never end up in a critical situation yielding an inevitable collision? The answer proposed in this paper to this safety issue relies upon the concept of inevitable collision states (ICS). ICS takes into account the dynamics of both the system and the moving obstacles. By computing ICS-free partial motion, the system safety can be guaranteed. Application of PMP to the case of a car-like system in a dynamic environment is presented.

Predictive low-rank decomposition for kernel methods

Abstract: Low-rank matrix decompositions are essential tools in the application of kernel methods to large-scale learning problems. These decompositions have generally been treated as black boxes---the decomposition of the kernel matrix that they deliver is independent of the specific learning task at hand---and this is a potentially significant source of inefficiency. In this paper, we present an algorithm that can exploit side information (e.g., classification labels, regression responses) in the computation of low-rank decompositions for kernel matrices. Our algorithm has the same favorable scaling as state-of-the-art methods such as incomplete Cholesky decomposition---it is linear in the number of data points and quadratic in the rank of the approximation. We present simulation results that show that our algorithm yields decompositions of significantly smaller rank than those found by incomplete Cholesky decomposition.

Graph Kernels for Molecular Structure−Activity Relationship Analysis with Support Vector Machines

Abstract: The support vector machine algorithm together with graph kernel functions has recently been introduced to model structure-activity relationships (SAR) of molecules from their 2D structure, without the need for explicit molecular descriptor computation. We propose two extensions to this approach with the double goal to reduce the computational burden associated with the model and to enhance its predictive accuracy: description of the molecules by a Morgan index process and definition of a second-order Markov model for random walks on 2D structures. Experiments on two mutagenicity data sets validate the proposed extensions, making this approach a possible complementary alternative to other modeling strategies.

Motivation-driven educational game design: applying best practices to music education

Abstract: Building upon research on motivation theory, we provide insights on how video games can be framed as expert tools that naturally reconcile learning and fun, a worthy goal since students are forced to where players volunteer, namely learning. Our main contribution is to suggest best practices rooted in psychology and motivation theory that we believe should be used when designing educational games.As a case study, we illustrate this approach by looking at the design of our ongoing Cha-Luva Swing Festival project, a video game dedicated to music education. Its overall pedagogical goal is to spur the players' curiosity and creativity by enabling them to easily play music using gamepads as musical instruments. This example perfectly illustrates our results, since motivational lapses due to instrumental limitations often lead to resignation and surrender.

Vehicle detection combining gradient analysis and AdaBoost classification

Abstract: This paper presents a real-time vision-based vehicle's rear detection system using gradient based methods and Adaboost classification, for ACC applications. Our detection algorithm consists of two main steps: gradient driven hypothesis generation and appearance based hypothesis verification. In the hypothesis generation step, possible target locations are hypothesized. This step uses an adaptive range-dependant threshold and symmetry for gradient maxima localization. Appearance-based hypothesis validation verifies those hypothesis using AdaBoost for classification with illumination independent classifiers. The monocular system was tested under different traffic scenarios (e.g., simply structured highway, complex urban environments, varying lightening conditions), illustrating good performance.

A novel representation of protein sequences for prediction of subcellular location using support vector machines

Abstract: As the number of complete genomes rapidly increases, accurate methods to automatically predict the subcellular location of proteins are increasingly useful to help their functional annotation. In order to improve the predictive accuracy of the many prediction methods developed to date, a novel representation of protein sequences is proposed. This representation involves local compositions of amino acids and twin amino acids, and local frequencies of distance between successive (basic, hydrophobic, and other) amino acids. For calculating the local features, each sequence is split into three parts: N-terminal, middle, and C-terminal. The N-terminal part is further divided into four regions to consider ambiguity in the length and position of signal sequences. We tested this representation with support vector machines on two data sets extracted from the SWISS-PROT database. Through fivefold cross-validation tests, overall accuracies of more than 87% and 91% were obtained for eukaryotic and prokaryotic proteins, respectively. It is concluded that considering the respective features in the N-terminal, middle, and C-terminal parts is helpful to predict the subcellular location.

The inference of protein–protein interactions by co-evolutionary analysis is improved by excluding the information about the phylogenetic relationships

Abstract: Motivation: The prediction of protein--protein interactions is currently an important issue in bioinformatics. The mirror tree method uses evolutionary information to predict protein--protein interactions. However, it has been recognized that predictions by the mirror tree method lead to many false positives. The incentive of our study was to solve this problem by improving the method of extracting the co-evolutionary information regarding the protein pairs. Results: We developed a novel method to predict protein--protein interactions from co-evolutionary information in the framework of the mirror tree method. The originality is the use of the projection operator to exclude the information about the phylogenetic relationships among the source organisms from the distance matrix. Each distance matrix was transformed into a vector for the operation. The vector is referred to as a 'phylogenetic vector'. We have proposed three ways to extract the phylogenetic information: (1) using the 16S rRNA from the same source organisms as the proteins under consideration, (2) averaging the phylogenetic vectors and (3) analyzing the principal components of the phylogenetic vectors. We examined the performance of the proposed methods to predict interacting protein pairs from Escherichia coli, using experimentally verified data. Our method was successful, and it drastically reduced the number of false positives in the prediction. Availability: The R script for the prediction of protein--protein interactions reported in this manuscript is available at http://timpani.genome.ad.jp/~proj/ Contact: sato@kuicr.kyoto-u.ac.jp Supplementary information: The information is also available at the same site as the R script.

Towards an epistemology of collective action: management research as a responsive and actionable discipline

Abstract: Critics simultaneously question the relevance of academic knowledge about management, the scientific unity of the field and the performance of business schools. They signal the lack of an adequate epistemology for management research. This paper explores the possible nature of such an epistemology. Classic epistemology looked for universal truth but ended up accepting a model of action: the experimental method. Contemporary critiques of classic epistemology offer relativist views about truth but focus on metaphysics of action, that is, false universals of collective action. Truth being dependant of models of action, a useful epistemology can be defined as the research-based revision of these models. It is quite different from pragmatism and relativism: applied to the ‘knowing observer’ it led to important discoveries in physics and mathematics. Management research should similarly recognize that each knowing method gives access to different truths and corresponds to a different conception of responsiveness and actionability. The identity and value of management research is also clarified as opposing metaphysics of action and developing a research process for the identification and revision of new models of action. Four examples of management research leading to such revision are discussed: the theory of management instruments vs the metaphysics of social control; the theory of prescribers vs the metaphysics of markets and hierarchies; the theory of collective knowledge production vs the metaphysics of expertise; the theory of innovative design vs the metaphysics of R&D projects. Within an epistemology of collective action, management research can be defined as a basic discipline and not as an applied one.