Éditeur(s) : HAL CCSD IEEE
Résumé : International audience
Recent works display that large scale image classification problems rule out computationally demanding methods. On such problems, simple approaches like k-NN are affordable contenders, with still room space for statistical improvements under the algorithmic constraints. A recent work showed how to leverage k-NN to yield a formal boosting algorithm. This method, however, has numerical issues that make it not suited for large scale problems. We propose here an Adaptive Newton-Raphson scheme to leverage k-NN, N³, which does not suffer these issues. We show that it is a boosting algorithm, with several key algorithmic and statistical properties. In particular, it may be sufficient to boost a subsample to reach desired bounds for the loss at hand in the boosting framework. Experiments are provided on the SUN, and Caltech databases. They confirm that boosting a subsample -- sometimes containing few examples only -- is sufficient to reach the convergence regime of N³. Under such conditions, N³ challenges the accuracy of contenders with lower computational cost and lower memory requirement.
IEEE International Workshop on Machine Learning for Signal Processing
Southampton, United Kingdom

hal-00959125
https://hal.archives-ouvertes.fr/hal-00959125
https://hal.archives-ouvertes.fr/hal-00959125/document
https://hal.archives-ouvertes.fr/hal-00959125/file/BNNB_mlsp2013.pdf

Éditeur(s) : HAL CCSD
Résumé : A standard approach for large scale image classification involves high dimensional features and Stochastic Gradient Descent algorithm (SGD) for the minimization of classical Hinge Loss in the primal space. Although complexity of Stochastic Gradient Descent is linear with the number of samples these method suffers from slow convergence. In order to cope with this issue, we propose here a Stochastic Low-Rank Newton Descent SLND for minimization of any calibrated loss in the primal space. SLND approximates the inverse Hessian by the best low-rank approximation according to squared Frobenius norm. We provide core optimization for fast convergence. Theoretically speaking, we show explicit convergence rates of the algorithm using these calibrated losses, which in addition provide working sets of parameters for experiments. Experiments are provided on the SUN, Caltech256 and ImageNet databases, with simple, uniform and efficient ways to tune remaining SLND parameters. On each of these databases, SLND challenges the accuracy of SGD with a speed of convergence faster by order of magnitude.
https://hal.archives-ouvertes.fr/hal-00825414

hal-00825414
https://hal.archives-ouvertes.fr/hal-00825414
https://hal.archives-ouvertes.fr/hal-00825414/document
https://hal.archives-ouvertes.fr/hal-00825414/file/TechnicalReport.pdf

Résumé : After a period of rapid development of the sites (an increase in both reared surface area and production), tropical shrimp aquaculture is currently being faced with critical problems due to economical and ecological constraints. In many countries with favourable conditions for shrimp aquaculture such as indonesia or Vietnam, sites are often badly selected and/or over-exploited. This conducts to sorne extent to decreases or collapses in the production due to the difficulty in predicting the maximum production capacity of the sites. Furthermore, it appears that production sustainability depends on many factors among which socio-economy and ecolo gy are of prime importance. The STD3 project titled « Assessment of tropical shrimp aquaculture impact on the environment in tropical countries using hydrobiology, ecology and remote sensing as helping tools for diagnosis )) was initiated in 1994, with the aim to study the sustainability of marine shrimp aquaculture in tropical areas, main! y Indonesia and Vietnam, through the following tasks: (i) improve site selection and study the impact of aquaculture on marine environments presenting a variable sensitivity to organic sewage, (ii) analyse the socio-economical aspects and profitability of the aquaculture industry and of common resources (iii) use remote sensing and geographical data bases for diagnosis and monitoring of site degradation (iv) reinforce the capacity of Asian scientists through training. Severa! aquaculture sites have been investigated in the Lampung region (South Sumatra, Indonesia), the Mekong delta (Vietnam) and on the West coast of New-Caledonia, providing a large range of typical ecosystems encountered in Asia and in the Pacifie, from coralline sites to coastal plains with mangrove and deltaic areas. The study concerned the spatial structure and seasonal influence of these ecosystems, the temporal variation during the course of the project (3 years) and the relation between the ecological structure and farm productions. Remote sensing provides a synoptic vision over large land expanses. Severa! scenes were processed for landuse mapping using conventional classification techniques. Concerning water quality assessment, a general relation for the Java sea was found between image and field data in terms of total suspended matter. Applying this relation to a new site in Sumatra has provided an initial approach to water type and, together with land use mapping, a preliminary assessment of the suitability of the area to shrimp aquaculture development. Socio-economic research has been focused on the identification ofneeds for collective action, including public policy, in the perspective of shrimp farming sustainability. The main concern is in the regulation of shrimp farming intensification and extensification at the scale of coastal ecosystemic entities in a common property resource management perspective. The common considered here is coastal water quality. Comparing the local development profiles and the institutional grounds for the design and implementation of collective management rules shows that variables such as land tenure system, farm owner socio-economic profile, social homogeneity or heterogeneity of the farmers are key factors to analyse the potential for sustainability. The defmition of water quality and the possible means to ensure its collective management are discussed. The main conclusion is that the economie incentives to farming development are strong but there is no significant difference among traditional, semi-intensive and intensive systems in terms of economie efficiency or wealth distribution. The description of the functioning of the different kinds of ecosystems make it possible to give an advice concerning the positioning of the activity inside the ecosystem, and furthermore to have a reference state in order to determine the impact of the activity on the coastal environnement. This impact can be summarized as an increase in the concentration of total suspended matt€?r, of particulate organic matter and of sulfate reducing bacterias. This organic ma~.er is either issued from rearing activities or from telluric origine, due to mangrove eradication when building the ponds. An evaluation of the relationship existing between productivity and characteristics of each ecosystem bas been conducted by correlating ecological indicators data (concentration of total suspended matter, particulate organic matter, chlorophyll, percentage of pheopigments in total pigments and cyanobacteria and sulfatereducing bacteria in water and sediments) with the average production of the farms. This comparison lead to the compilation of a scale of observed production integrating the range of these environmental parameters with shrimp production levels, bound to facilitate the positioning of shrimp farming in relation with the confmement leve! of each ecosystem. Lastly, ali geographie data originated from the various compartments of the study have been geo-referenced and loaded into a geographical information system. This allows to display any query made on spatial variables and their related statistical data, including the ir variations over the last few years and to reveal patterns and phenomena otherwise not obvious.
Le programme mis en oeuvre par la DRV (Paris, CREMA-L'Houmeau, DELIAO Brest, COP Tahiti, GIE/RA Nouvelle Calédonie), dans le cadre d'un projet européen STD3 «(Sciences et Technique au Service du Développement», en collaboration avec l'Université de Montpellier II, l'Ecole Nationale d'Agronomie de Rennes, l'ITC (Hollande), l'Institut Océanographique de Nha Trang (Vietnam), le BADC de Jepara et le BPPT de Jakarta (Indonésie), avait pour but la mise au point de méthodes permettant d'améliorer la sélection des sites ainsi que le suivi du développement et de l'impact de l'aquaculture des crevettes Péneides en milieu tropical pour éviter la surexploitation. Quatre axes de recherche ont été défmis : a) identifier la nature et quantifier les rejets de déchets issus de ce type d'aquaculture, b) définir les caractéristiques écologiques des écosystèmes susceptibles d'accueillir l'activité aquacole et suivre leur évolution sous l'influence des rejets, c) déterminer l'influence de la gestion des fermes et des ressources (aspects économiques) sur les performances de productivité, d) déterminer la capacité de la télédétection comme outil de diagnostic d'impact à l'échelle de la région, e) mener un programme de formation d'étudiants et jeunes chercheurs Vietnamiens et Indonésiens.

Droits : 1998 Ifremer
http://archimer.ifremer.fr/doc/00132/24357/22357.pdf
http://archimer.ifremer.fr/doc/00132/24357/