On-line Learning

Top PDF On-line Learning:

Inverse Kinematics On-line Learning: a Kernel-Based Policy-Gradient approach

Inverse Kinematics On-line Learning: a Kernel-Based Policy-Gradient approach

September 24, 2010 Abstract In machine learning, “kernel methods” give a consistent framework for applying the perceptron algorithm to non-linear problems. In rein- forcement learning, an analog of the perceptron delta-rule can be derived from the ”policy-gradient” approach proposed by Williams in 1992 in the framework of stochastic neural networks. Despite its generality and straighforward applicability to continuous command problems, quite few developments of the method had been proposed since. Here we present an account of the use of a kernel transformation of the perception space for the on-line learning of a motor command, in the case of eye orienta- tion and multi-joint arm control. We show first that such a setting allows the system to solve non-linear problems, like the log-like resolution of a foveated retina, or the transformation from a cartesian perception space to the “angular” command of the multi-joint arm. More interestingly, the on-line recurrent learning we propose is simple and fully operant in chang- ing environments, and allows for constant improvements of the politics, on the basis of simple and measurables error terms.
En savoir plus

13 En savoir plus

CAFE: an automatic and on-line learning system to guide freshmen towards the meeting of Higher Education requirements

CAFE: an automatic and on-line learning system to guide freshmen towards the meeting of Higher Education requirements

This communication focuses on an original AfL based automatic and on-line learning system, called “CAFE”, implemented in the context of the Computer Programming Introduction course addressed to first year students. In a nutshell, the system proposes all over the semester multiple programming challenges. It leads the students to work, on a regular basis, on problems with increasing difficulties and cumulative expected learning outcomes, “ensuring that summative assessment has a positive impact on learning” (Nicol, 2009). CAFE also allows students for three submissions for each challenge, thus closing the feedback loop (Boud, 2000). In addition, for each submission, CAFE provides a high quality and automatic feedback that is parametrized according to the literature in order to maximize the likelihood of student self regulation. In particular, this feedback is
En savoir plus

2 En savoir plus

On-Line Learning of Lexical Items and Grammatical Constructions via Speech, Gaze and Action-Based Human-Robot Interaction

On-Line Learning of Lexical Items and Grammatical Constructions via Speech, Gaze and Action-Based Human-Robot Interaction

INSERM Stem Cell and Brain Research Institute, Lyon, France firstname.name@inserm.fr Abstract In order to be able to understand a conversation in interaction, a robot, has to first understand the language used by his inter- locutor. A central aspect of language learning is adaptability. Individuals can learn new words and new grammatical struc- tures. We have developed learning methods that allow the hu- manoid robot iCub to robot can learn new lexical items by in- teraction with the human and consolidation of its autobiograph- ical memory. Then, based on these open class words, the robot can bootstrap the acquisition of novel grammatical structures in real-time. Finally, we demonstrate how human gaze can be monitored, and could be used in order to reduce referential am- biguity inherent in such learning conditions. These learning ca- pabilities are demonstrated in a collection of videos.
En savoir plus

4 En savoir plus

Expert-based on-line learning and prediction in Content Delivery Networks

Expert-based on-line learning and prediction in Content Delivery Networks

In this paper, we focus on the use of machine learning techniques in CDNs. The remainder of this paper is organized as follows. In this section, we present the context of this study and some related work. In Section II, we define the theoretical framework including the problem statement. In the context of our paper, an expert refers to a software component that implements a prediction computation method. We define all the experts used (e.g. DES, Basic, polynomial or Savitzky-Golay regression, etc) coming from various fields (e.g. statistics, control theory) and explain how they compute their predictions. These predictions are used by a decision maker, called forecaster, to build its own prediction. We define two types of forecasters: Best Expert (BE) and K Best Experts (KBE). In Section III, we report our simulation results obtained from real traces of YouTube. We consider various video contents and evaluate the prediction accuracy of the forecasters and experts considered. We also compare the
En savoir plus

7 En savoir plus

Effects of line-blocking on the non-LTE Fe I spectral line formation

Effects of line-blocking on the non-LTE Fe I spectral line formation

In our study, in order to include the e ffects of line-blocking in  we proceeded by sampling metal line opacities for about 9000 wavelength points in the spectral region between 1000 Å and 20 000 Å and adding them to the standard back- ground continuous opacities. In order not to count the contribu- tion of Fe line opacities twice, metal line opacities were only added to the background when computing intensities for the photo-ionisation rates and not for Fe  and Fe  line profiles. Molecular line opacities were not considered in our line forma- tion calculations which is a reasonable approach for our stars where the atomic contribution dominates overwhelmingly. We performed simple bench-mark test LTE radiative transfer calcu- lations with background line opacities sampled between 9000 and 70 000 wavelength points covering the same wavelength region. The tests indicate that photo-ionisation rates computed with our choice of opacity sampling for 9000 wavelengths are su fficiently accurate for our purposes. In addition, we per- formed non-LTE test calculations sampling the opacities at dif- ferent wavelengths and found that the overall e ffect on the final Fe non-LTE abundance corrections is less than 0.01 dex.
En savoir plus

9 En savoir plus

Dictionary-based Learning in MR Fingerprinting: Statistical Learning versus Deep Learning

Dictionary-based Learning in MR Fingerprinting: Statistical Learning versus Deep Learning

Synopsis In MR Fingerprinting, the exhaustive search in the dictionary may be bypassed by learning a mapping between fingerprints and parameter spaces. In general, the relationship between these spaces is particularly non-linear, which implies the use of advanced regression methods: deep learning frameworks but also methods based on statistical models have been proposed. In this study, we compare reconstruction time, accuracy and noise robustness of the conventional dictionary-matching method and two methods that handle the modelling of the non-linear relashionship with a neural network and a statistical inverse regression model.
En savoir plus

5 En savoir plus

Collab-Net on line: An on line platform for DSS research collaboration in EWG-DSS

Collab-Net on line: An on line platform for DSS research collaboration in EWG-DSS

ICDSST 2016 Conference on Decision Support Systems Addressing Sustainability & Societal Challenges Collab-Net on line: An on line platform for DSS-research collaboration in EWG-DSS Recently, the scientific research collaboration has been extended to social network analysis area, which concerns in evaluating the collaborative interaction among papers authors within publication databases. The Collab-Net platform aims to investigate these publication relationship in an automatic way by a Web-based platform. The present system was developed using free platforms for software development and database system purposes. The main goal of the Collab-Net is to allow researchers to analyze their own collaborative network, as well as possibilities for future collaboration among EWG-DSS members only using a Web-based platform, in anywhere at anytime. The Collab-Net system can be used as two end-users profiles: Administrator and Member profiles.
En savoir plus

3 En savoir plus

Collab-Net on line: An on line platform for DSS research collaboration in EWG-DSS

Collab-Net on line: An on line platform for DSS research collaboration in EWG-DSS

ICDSST 2016 Conference on Decision Support Systems Addressing Sustainability & Societal Challenges Collab-Net on line: An on line platform for DSS-research collaboration in EWG-DSS Recently, the scientific research collaboration has been extended to social network analysis area, which concerns in evaluating the collaborative interaction among papers authors within publication databases. The Collab-Net platform aims to investigate these publication relationship in an automatic way by a Web-based platform. The present system was developed using free platforms for software development and database system purposes. The main goal of the Collab-Net is to allow researchers to analyze their own collaborative network, as well as possibilities for future collaboration among EWG-DSS members only using a Web-based platform, in anywhere at anytime. The Collab-Net system can be used as two end-users profiles: Administrator and Member profiles.
En savoir plus

2 En savoir plus

Interactive Illustrative Line Styles and Line Style Transfer Functions for Flow Visualization

Interactive Illustrative Line Styles and Line Style Transfer Functions for Flow Visualization

Figure 1: A view-oriented strip subdivided into a number of bands mirrored around the centerline. that is necessary to allow us to define visually distinguishable styles, each of the bands increasing the number of visual variables that can be controlled. These bands run parallel to the centerline and together define the visualization line style. Specifically, we represent each line from the 3D dataset by a view-oriented line strip as done in many previous line- based rendering systems. This strip is subdivided into two mirrored sets of bands, one on each side of the line (Fig. 1). The three visual properties that we control per band are color, width, and distance offset w.r.t. the viewer, each of which can be controlled independently. While the distance offset is not actually a visual property, it has an effect when used as a depth-dependent halo [TCM06,EBRI09]. In that case the halo line band is folded back, away from the viewer. The effect is that the visible width of the halo depends on the difference in distance between two lines w.r.t. the viewer, improving the depth perception. Therefore, our extended linestyle model can be seen as a generalization of the depth-dependent line halo technique [EBRI09].
En savoir plus

10 En savoir plus

Learning Analytics : terminologie du Learning Analytics.

Learning Analytics : terminologie du Learning Analytics.

L’incorporation de terme « traces » se justifient par le fait que le terme trace n'est pas perçu par tout le monde de la même façon, dans le cadre des Learning Analytics, on prend en compte les traces au sens d'éléments laissés par l'utilisateur durant sa session (logs, productions, ...), mais aussi toutes sortes de données liées à la session d'apprentissage comme le scénario pédagogique qui préexiste à la session, les données de paramétrages de la plateforme lorsqu'il y en a une, des listes d'apprenants éventuellement organisée en groupe, ..
En savoir plus

11 En savoir plus

Characterizing N+-perfect line graphs

Characterizing N+-perfect line graphs

coincide and that ASTAB ∗ (G) shall be the studied polyhedral relaxation P (G) of STAB(G) in (2). In particular, in this paper we have proved that every N + -perfect line graph is h-perfect. Then, combining these results, we obtain that a line graph is joined a-perfect if and only if it is h-perfect. However, it seems nat- ural to look for a proof of the latter result independent of the N + -operator.

18 En savoir plus

Data Prefetching via Off-line Learning

Data Prefetching via Off-line Learning

Data Prefetching via Off-line Learning Wong Weng Fai Abstract—The widely acknowledged performance gap between processors and memory has been the subject of much research. In the Explicitly Parallel Instruction Computing (EPIC) paradigm, the combination of in-order issue and the presence of a large number of parallel function units has further worsen the problem. Prefetching, by hardware, software or a combination of both, has been one of the primary mechanisms to alleviate this problem. In this talk, we will discuss two prefetching mechanisms, one hardware and other software, suitable for implementation in EPIC processors. Both methods rely on the off-line learning of Markovian predictors. In the hardware mechanism, the predictors are loaded into a table that is used by a prefetch engine. We have shown that the method is particularly effective for prefetching into the L2 cache. Our software mechanism which we called predicated prefetch leverages on informing loads. This is used in conjunction with data remapping and offline learning of Markovian predictors. This distinguishes our approach from early software prefetching techniques that only involves static program analysis. Our experiments show that this framework, together with the algorithms used in it, can effectively remove, in the best instance, 30% of the stall cycles due to cache misses. The results also show that the framework performs better than pure hardware stride predictors and has lower bandwidth and instruction overheads than that of pure software approaches.
En savoir plus

1 En savoir plus

Forecasting for dynamic line rating

Forecasting for dynamic line rating

Generally, the discussion on whether DLR presents an economically feasible and rational solution focuses on two dimensions that mainly represent the viewpoints of different network stakeholders (utilities and consumers). Switching to a DLR operation mode requires installing new equipment for conductor monitoring and adopting new technologies for ambient conditions measurement/forecasting. In addition, it may require upgrade of some other transmission line components, but the conductors, in order to allow higher loading with DLR. For a utility company, this amounts to launching a new, possibly riskier project whose benefits must be weighed against the obvious choice of upgrading an otherwise seasonally rated grid. The relative merits of each alternative can be evaluated on the grounds of several investment performance metrics (capital intensity, project lifetime, payback period, etc.) provided, of course, that all inputs into the decision-making process (costs/benefits) can be adequately expressed in financial terms. This can be a tedious task when taking into account the complexity of modern networks and the great number of parameters involved, although flow-based approaches presently being developed in central Western Europe may be significantly helpful. Furthermore, cost estimates are typically uncertain and can significantly vary across countries or regions.
En savoir plus

49 En savoir plus

Projective groups and line arrangements

Projective groups and line arrangements

mathematics to study the Minor fiber of an arrangement. For a line arrangement A in P 2 , its Milnor fiber F is on the one hand is a quasi-projective manifold, hence we may use mixed Hodge theory to study its cohomology jump loci; on the other hand, F is also a surface, for which the well-known surface theory may help. Moreover, a line arrangement is a quite special object in topology and much has been well understood about the complement M = M (A) for the arrangement A, see [OT]. Since F is clearly a Galois cover of M , it is a natural idea to study F using the known results on M and studying the monodromy associated to the Galois
En savoir plus

138 En savoir plus

MPLM -- MaTeLo Product Line Manager

MPLM -- MaTeLo Product Line Manager

4. CONCLUSION MPLM is a novel approach that extends model-based test- ing to product line engineering. The approach was applied in an industrial study from the aerospace domain. One ma- jor outcome of this case study is that the presented approach allows to significantly reduce the test efforts for variants of a product line. The product line usage model covers all features of the product line and hence allows to effectively reuse the test artefacts (e.g. generated test cases) across dif- ferent variants. In the past, separate projects were setup for different customers and consequently requirements analysis and test case design was done independently for all variants without systematic reuse.
En savoir plus

6 En savoir plus

Le PICC Line : intérêt pour le maintien à domicile

Le PICC Line : intérêt pour le maintien à domicile

Lors de mon stage à l’hôpital de Beauvais, j’ai pu observer, dans différents services, l’utilisation de matériel médical identique pour des indications différentes. Dans le service d’hémato-oncologie, j’ai observé l’utilisation de chambres implantables, nécessitant une opération chirurgicale et plusieurs jours d’attente avant utilisation ; puis dans le service d’infectiologie j’ai constaté l’utilisation du Peripherally Inserted Central Catheter (PICC line) sur un patient nécessitant un traitement antifongique parentéral. Chez ce patient, l’abord jugulaire était contre indiqué. L’usage de ce dispositif médical par un médecin informé des techniques utilisées dans les pays anglo-saxons pour un patient présentant des caractéristiques communes à celles des patients d’hémato-oncologie m’a poussé à m’y intéresser.
En savoir plus

62 En savoir plus

Pictorial analysis of line-drawings

Pictorial analysis of line-drawings

T. Hurtut, Y. Gousseau, F. Cheriet, & F. Schmitt / Pictorial analysis of line-drawings This approach is closely related to cognitive studies such as the seminal work of Attneave for instance [ Att54 ]. Studying the line content in artworks has been rarely in- vestigated in computer vision. Berezhnoy et al. [ BPvdH05 ] proposed a method using polynomial interpolation of paint- ing brush-strokes on Van Gogh artworks. This approach is adapted to very small strokes such as Van Gogh’s. Li and Wang studied ancient black and white ink drawings using wavelets and hidden Markov models [ LW04 ]. Their method is yet close to texture analysis in paintings. Onkarappa and Guru studied the spatial mutual arrangement of strokes in line-drawing images to achieve similarity retrieval [ OG07 ]. This approach aims at describing the artistic composition of an image.
En savoir plus

9 En savoir plus

On-line recognition of interval orders

On-line recognition of interval orders

Unit´e de recherche INRIA Lorraine, Technˆopole de Nancy-Brabois, Campus scientifique, ` NANCY 615 rue de Jardin Botanique, BP 101, 54600 VILLERS LES Unit´e de recherche INRIA Rennes, IR[r]

24 En savoir plus

Synthetic pulsational line profile variations

Synthetic pulsational line profile variations

Fig. 5 Discriminant value as function of possible mode identifica- tions, obtained using the moment method. Each panel includes the re- sults for spectral lines of the same ion, from up to down and left to right: SiIII, OII, FeIII, CII and NII lines. The correct mode identifica- tion ( = 0) is indicated with a dash-dotted vertical line. The lower the discriminant is, the higher the probability for this mode identification

6 En savoir plus

Transmission line fault locator

Transmission line fault locator

L’accès à ce site Web et l’utilisation de son contenu sont assujettis aux conditions présentées dans le site LISEZ CES CONDITIONS ATTENTIVEMENT AVANT D’UTILISER CE SITE WEB. NRC Publicat[r]

15 En savoir plus

Show all 4013 documents...