L’archive ouverte pluridisciplinaire HAL, est destinée au dépôt et à la diffusion de documents scientifiques de niveau recherche, publiés ou non, émanant des établissements d’enseignemen[r]

There are well-known standard ways of extracting optimal features according to a given criterion. For instance in unsupervised problems, **the** first k principal components of a dataset give **the** best linear approximation of **the** original data in
R k for **the** quadratic norm (see [13] for **functional** principal component analysis (PCA)). In regression problems, **the** partial least-squares approach extracts features with maximal correlation with a target **variable** (see also Sliced Inversion Regression methods [4]). **The** main drawback of those approaches is that they extract features that are not easy to interpret: while **the** link between **the** original features **and** **the** new ones is linear, it is seldom sparse; an extracted feature generally depends on many original features.

En savoir plus
our knowledge, **the** model has been first mentioned in **the** work of Cardot et al. (2007) under **the** name of multiple **functional** linear model. An estimator of β is defined with an iterative backfitting algorithm **and** applied to **the** ozone prediction dataset initially studied by Aneiros-P´erez et al. (2004). **Variable** selection is performed by testing all **the** possible models **and** selecting **the** one minimising **the** prediction error over a test sample. Let us also mention **the** work of Chiou et al. (2016) who consider a multivariate linear regression model with **functional** output. They define a consistent **and** asymptotically normal estimator based on **the** multivariate **functional** principal components initially proposed by Chiou et al. (2014).

En savoir plus
not possible if dim(H j ) = +∞ since b Γ j is a finite-rank operator. Without this condition,
Equation (14) does not admit a closed-form solution **and**, hence, is not calculable. We then propose **the** GPD (Groupwise-Majorization-Descent) algorithm, initially proposed by Yang **and** Zou (2015), to compute **the** solution paths of **the** multivariate Group-Lasso penalized learning problem, without imposing **the** group-wise orthonormality condition. **The** GPD algorithm is also based on **the** principle of coordinate-wise descent but **the** minimisation problem (14) is modified in order to relax **the** group-wise orthonormality condition. We denote by b β (k) **the** value of **the** parameter at **the** end of iteration k. During iteration k + 1, we update sequentially each coordinate. Suppose that we have changed **the** j − 1 first coordinates (j = 1, ..., p), **the** current value of our estimator is ( b β 1 (k+1) , ..., b β j−1 (k+1) , b β j (k) , ..., b β p (k) ). We want to update **the** j-th coefficient **and**, ideally, we

En savoir plus
We have taken an extrinsic approach [ Benton et al. 2012 ], as opposed to an intrinsic one, to **the** representation of **the** core language, **the** module language, **and** **the** target language, which keeps our implementation close to **the** approach presented in **the** paper. **The** extrinsic encoding has an advantage of being more suitable for code extraction to obtain a certified implementation. That is, we have implemented **the** abstract syntax as simple inductive data types **and** given separate inductive definitions for relations such as elaboration, typing, **and** so on. **The** semantic objects of Figure 4 have been implemented as mutually defined inductive types using Coq’s with clause. **The** same approach is used for definitions of relations on environments. As described in Section 4, semantic objects are represented using finite maps **and** sets **and** indeed, **the** implementation makes use of Coq’s standard library implementations of such objects. Specifically, we use **the** FMapList **and** FSetList implementations of **the** FMap **and** FSet interfaces, respectively. Both FMapList **and** FSetList make use of **the** list data type together with a property that **the** list is ordered according to a strict order on **the** underlying data structure. **The** strict order for **the** underlying list allows us to prove an extensionality property for environments **and** sets (assuming proof irrelevance). That is, for any two environments 𝐸 1 **and** 𝐸 2 we have (∀𝑘, 𝐸 1 (𝑘) = 𝐸 2 (𝑘)) → 𝐸 1 = 𝐸 2 . **The** equal sign = refers to **the** Coq propositional equality, which means that we can use all **the** standard rewriting machinery instead of using setoid equality.

En savoir plus
I. I NTRODUCTION
An old model of computation, recursion schemes were originally designed as a canonical **programming** calculus for studying program transformation **and** control structures. In recent years, higher-order recursion schemes (HORS) have received much attention as a method of constructing rich **and** robust classes of possibly infinite ranked trees (or sets of such trees) with strong algorithmic properties. **The** interest was sparked by **the** discovery of Knapik et al. [2] that HORSs which satisfy a syntactic constraint called safety generate **the** same class of trees as higher-order pushdown automata. Remarkably these trees have decidable monadic second-order (MSO) theories, subsuming earlier well-known MSO decidability results for regular (or order-0) trees [3] **and** algebraic (or order-1) trees [4]. We now know [5] that **the** modal µ-calculus (local) model checking problem for

En savoir plus
Algebraic types **and** pattern matching in Why 3
1 Motivation
This work was inspired by **the** recent experiments [1] with verification of floating- point computations in Why [2]. According to **the** IEEE Standard 754, which specifies **the** representation **and** operation for **the** floating-point numbers, at any point a programmer can choose: one of five different encodings (binary numbers of single, double, **and** quadruple precision **and** decimal numbers of double **and** quadruple precision); one of five rounding algorithms; a computation mode with or without overflows. Correspondingly, **the** **logical** annotations in a floating- point program must take into account **the** encoding of a particular **variable** or constant, as well as **the** current rounding algorithm **and** computation mode. This can be done, of course, with a number of appropriately chosen predicates **and** series of «if-then-else» expressions. However, a more elegant solution would be to use three enumerated types, namely:

En savoir plus
Colon Cancer GP-GEP (8 functions) 0.002874
FRPR 0.024015
It is impossible to represent a VR space on hard media. Snapshots of **the** visual spaces computed over **the** Breast Cancer data set using GP-GEP (Fig. 4(b), Fig. 5(a)) **and** FRPR (Fig. 5(b)) show **the** similarity structure **and** **the** orig- inal classes (benign **and** malignant tumors). In addition, **the** classes are wrapped with transparent membranes as an aid **and** Fig. 4(a) may be compared to Fig. 4(b) to appreciate **the** enhancement. Both algorithms, GP-GEP **and** FRPR, succeeded in showing **the** benign class (light objects) more densely packed **and** homogeneous than **the** malignant class. In addition, it can be observed that both classes have an important intersection making it difficult to expect perfect classification of this data with machine learning techniques. **The** same class structure is exhibited by **the** FRPR space (Fig. 5(b)) despite its lower mapping error.

En savoir plus
grated omics for **the** identification of key functionalities in biological wastewater treatment microbial communities,” Microbial Biotechnol- ogy, 2015.
[3] S. Widder, R. J. Allen, T. Pfeiffer, T. P. Curtis, C. Wiuf, W. T. Sloan, O. X. Cordero, S. P. Brown, B. Momeni, W. Shou, H. Kettle, H. J. Flint, A. F. Haas, B. Laroche, J. U. Kreft, P. B. Rainey, S. Freilich, S. Schuster, K. Milferstedt, J. R. Van Der Meer, T. Grobkopf, J. Huisman, A. Free, C. Picioreanu, C. Quince, I. Klapper, S. Labarthe, B. F. Smets, H. Wang, O. S. Soyer, S. D. Allison, J. Chong, M. C. Lagomarsino, O. A. Croze, J. Hamelin, J. Harmand, R. Hoyle, T. T. Hwa, Q. Jin, D. R. Johnson, V. de Lorenzo, M. Mobilia, B. Murphy, F. Peaudecerf, J. I. Prosser, R. A. Quinn, M. Ralser, A. G. Smith, J. P. Steyer, N. Swainston, C. E. Tarnita, E. Trably, P. B. Warren, **and** P. Wilmes, “Challenges in microbial ecology: Building predictive understanding of community function **and** dynamics,” 2016. [4] M. J. Wade, J. Harmand, B. Benyahia, T. Bouchez, S. Chaillou,

En savoir plus
Abstract. **Logical** relations are one among **the** most powerful tech- niques in **the** theory of **programming** languages, **and** have been used extensively for proving properties of a variety of higher-order calculi. However, there are properties that cannot be immediately proved by means of **logical** relations, for instance program continuity **and** differen- tiability in higher-order languages extended with real-valued functions. Informally, **the** problem stems from **the** fact that these properties are naturally expressed on terms of non-ground type (or, equivalently, on open terms of base type), **and** there is no apparent good definition for a base case (i.e. for closed terms of ground types). To overcome this is- sue, we study a generalization of **the** concept of a **logical** relation, called open **logical** relation, **and** prove that it can be fruitfully applied in sev- eral contexts in which **the** property of interest is about expressions of first-order type. Our setting is a simply-typed λ-calculus enriched with real numbers **and** real-valued first-order functions from a given set, such as **the** one of continuous or differentiable functions. We first prove a containment theorem stating that for any collection of real-valued first- order functions including projection functions **and** closed under function composition, any well-typed term of first-order type denotes a function belonging to that collection. Then, we show by way of open **logical** re- lations **the** correctness of **the** core of a recently published algorithm for forward automatic differentiation. Finally, we define a refinement-based type system for local continuity in an extension of our calculus with con- ditionals, **and** prove **the** soundness of **the** type system using open **logical** relations.

En savoir plus
4 Discussion **and** Conclusion
**The** contribution of this paper is not about a new query language, but about an interactive navigation process that guides users from query to query, along **the** LIS principles. **The** query language proposed here has been driven by LIS constraints (a query denotes a complex class), **and** **the** wish to have queries as concise **and** natural as possible. However, it is interesting to discuss further its expressivity compared to SPARQL [PAG06]. **The** missing graph patterns are **the** OPTIONAL, UNION, **and** FILTER patterns. In our case, **the** OPTIONAL pattern is useless because **the** SELECT clause has only one **variable**. Our prototype has already UNION patterns through an or operator but we do not have consistency **and** completeness results yet about their navigation. It also has negation, **and** some limited forms of FILTER patterns as predefined classes of literals. For instance, **the** class match "regexp" denotes **the** set of all strings that match a regular expression. Similarly, we have classes for intervals **and** inequalities over numbers **and** dates. **The** most important restriction in our queries is **the** one-**variable** SELECT clause. However, **the** index alleviates this restriction to some extent. Suppose **the** SPARQL query SELECT ?x ?y WHERE { ?x rdf:type gen:man . ?x gen:mother ?y }. By setting **the** query to a man, **and** by expanding mother : ?, **the** index gives for each mother, how many male children she has. A highlighting mechanism allows to select a man in **the** extension to discover who is his mother; **and** alternately, to select a mother in **the** index to discover which are her children. **The** index is an inverted view over **the** table of SPARQL results. Each subtree of **the** index (with count annotation) is a histogram of **the** values from a column of **the** table. **The** highlighting mechanism enables to retrieve **the** associations between **the** values of **the** different columns, i.e., **the** result tuples.

En savoir plus
At first **the** calculation of **the** pressure angles in **the** joints along **the** trajectory for all possible structures of **the** parallel mechanism with **variable** architecture must be accomplished, then **the** best structure must be chosen for which **the** maximum value of **the** pressure angle along **the** trajectory is always less than **the** limit value. If there is no structure satisfying this condition, **the** given trajectory must be decomposed in several parts **and** **the** generation of **the** motion must be carried out by different structures. It is obvious that in this case it would be desirable that **the** trajectory can be realized by minimal structural changes.

En savoir plus
Abstract Thanks to **the** rise of new wearable **and** non-intrusive sensor tech- nology, Internet of Things (IoT) contributes in human daily life improve- ment. In **the** context of smart vehicles, human affective monitoring should be based on a context-aware system in order to consider **the** interactions be- tween **the** driver, his vehicle **and** his ambient environment. In this chapter, we propose AffectiveROAD platform, that sense **the** human physiological changes, **the** ambient environment inside **the** vehicle, **and** **the** vehicle speed. **The** proposed sensor-based solutions are not only providing real-time phys- iological monitoring, but also enriching **the** tools for human affective **and** cognitive states tracking. Thanks to this platform, several driver’s state in- dicators such as stress **and** arousal may be developed **and** validated. Two types of wireless physiological sensors are used to monitor **the** electrodermal activity, **the** heart rate, **the** skin temperature, **the** respiration, **and** **the** motion of **the** driver. Moreover, we developed a sensor network allowing to capture **the** ambient temperature, humidity, pressure, **and** luminosity. **The** vehicle speed is extracted from **the** Global Position System (GPS) data captured using a smartphone. Two GoPro devices are used to capture **the** internal **and** external scenes. **The** purpose of this chapter is to describe a real-world driving protocol to collect data using **the** proposed IoT-based materials **and** to announce **the** publication of a database for driver’s state monitoring re- search. We propose 13 datasets related to drives in different road types: city **and** highway. A part of **the** database concerning **the** physiological **and** **the** environmental data is released for public use.

En savoir plus
141 En savoir plus

of exponentials, which are responsible for size increases in message passing. In order to obtain space bounds, **the** exponentials must therefore be restricted. **The** difficulty is to do so without **the** **programming** language becoming too weak. In this paper we argue that intml with its subexponentials represents a good solution to this problem.
Subexponentials make intml an expressive higher-order language. For example, we show that it can type **the** Kierstead terms, which cannot be typed in similar linear type systems, such as [14]. Moreover, **the** proof of flogspace- completeness in this paper is completely straightforward. In earlier work, such as [36, 42], **the** completeness proofs were more involved. **The** expressiveness of intml is further illustrated by **the** **programming** examples, such as in Section 4. Subexponentials not only give us control over space usage, they also afford an efficient treatment of controlled duplication. For example, Ghica **and** Smith [15] treat copying in **the** language by actual duplication of terms. Here we show how to allow copying by sharing without **the** need to duplicate parts of **the** program. Finally, **the** language intml presented in this paper is **the** first higher-order language with logarithmic space bounds that allows one to use (**and** define!) higher-order combinators such as for tail recursion **and** call with current continuation.

En savoir plus
In **the** context of logic-based models, **the** inference of Boolean (genetic) networks from time-series gene expression data has been addressed by several authors under dif- ferent hypotheses **and** methods [LIA 98, AKU 00, IDE 00, LÄH 03]. Recently, a brief review **and** evaluation of these methods has been published in [BER 13]. Importantly, methods for reverse engineering of biological systems are highly dependent on avail- able (amount of) data, prior knowledge **and** modeling hypotheses. In particular, re- verse engineering of Boolean logic models by confronting prior knowledge on causal interactions with phosphorylation activities has been first described in [SAE 09]. A genetic algorithm implementation was proposed to solve **the** underlying optimization problem, **and** a software was provided, CellNOpt [TER 12]. Nonetheless, stochastic search methods cannot characterize **the** models precisely: they are intrinsically unable not just to provide a complete set of solutions, but also to guarantee that an optimal so- lution is found. To overcome some of this shortcomings, mathematical **programming** approaches were presented in [MIT 09, SHA 12]. Notably, authors in [SAE 09] have shown that **the** model is very likely to be non-identifiable when we consider **the** experi- mental error from measurements. Hence, rather than looking for **the** optimum Boolean model, one is interested in finding (nearly) optimal models within certain tolerance. Interestingly, in **the** context of quantitative modeling, authors in [CHE 09] have elabo- rated upon **the** same argument. Clearly, an exhaustive enumeration of (nearly) optimal solutions would allow for identifying admissible Boolean logic models without any methodological bias. Importantly, previous methods, namely stochastic search **and** mathematical **programming**, are not able to cope with this question. Moreover, all sub- sequent analysis will certainly profit from having such a complete characterization of feasible models. For example, finding “key-players” in

En savoir plus
Chapter 1. Introduction
obtain **the** most plausible models for certain environmental conditions or specific cell type. This is normally achieved by defining an objective fitness function to be optimized [Banga, 2008]. Optimization over quantitative modeling leads to continuous optimization problems. On **the** other hand, reverse engineering considering qualitative models typically give rise to combinatorial (discrete) optimization problems. Notably, this subject represents a very active area of research as illustrated by **the** successive “DREAM” challenges [Stolovitzky et al., 2007]. Importantly, methods for reverse engineering of biological systems are highly depen- dent on available (amount of) data, prior knowledge **and** modeling hypotheses. For instance, an inference method from gene expression data collected by DNA microarrays, may not be applicable to biochemical data like phosphorylation assays collected using xMAP Luminex technology. In particular, reverse engineering of **logical** models for signaling networks by confronting prior knowledge on causal interactions with phosphorylation activities has been first addressed in [Saez-Rodriguez et al., 2009]. Therein, authors have shown that **the** model is non-identifiable as soon as we consider **the** experimental error from measurements. Hence, rather than looking for **the** optimum **logical** model, one aims at finding (nearly) optimal mod- els within certain tolerance. Interestingly, in **the** context of mathematical modeling, authors in [Chen et al., 2009] have elaborated upon **the** same argument. Clearly, an exhaustive enu- meration of (nearly) optimal solutions would allow for identifying admissible **logical** models without any methodological bias. Furthermore, all subsequent analysis will certainly profit from having such a complete characterization of feasible models. That is, being able to ad- dress a given problem but considering an ensemble of **logical** models may lead to more robust solutions. In fact, this is in line with recent work showing that an ensemble of models often yields more robust predictions than each model in isolation [Kuepfer et al., 2007, Marbach et al., 2012]. Importantly, existing approaches, namely stochastic search **and** mathematical **programming**, are not well-suited to cope with this question in an exhaustive manner. Hence, there is an increasing demand of more powerful computational methods in order to achieve robust discoveries in **the** context of logic-based modeling.

En savoir plus
155 En savoir plus

In this paper, we present a system that allows users of a Virtual Learning Environment to seamlessly work with web-based laboratories consisting of real robots or 2D/3D simulators. User programs consist of fully-**functional** source code written on any of **the** supported **programming** languages (Python, Lisp, Matlab). **The** code is executed in **the** remote laboratory, thus it can access all **the** available information **and** services, without any additional remote communication overhead during execution. Upon finishing, **the** output of **the** process is returned back to **the** user’s browser, **and** **the** generated data is readily available to download for further analysis.

En savoir plus
Issues Raised by the Use of LP at Peerless Procurement, Provisioning, and Production at Peerless What the Provisioner Needs General Philosophy The Ideal System: The Ideal System: Data Ge[r]

280 En savoir plus