Éditeur(s) : HAL CCSD
Résumé : International audience
The k-nearest neighbors (k-NN) classification rule is still an essential tool for computer vision applications, such as scene recognition. However, k-NN still features some major drawbacks, which mainly reside in the uniform voting among the nearest prototypes in the feature space. In this paper, we propose a new method that is able to learn the "relevance" of prototypes, thus classifying test data using a weighted k-NN rule. In particular, our algorithm, called Multi-class Leveraged k-nearest neighbor (MLNN), learns the prototype weights in a boosting framework, by minimizing a surrogate exponential risk over training data. We propose two main contributions for improving computational speed and accuracy. On the one hand, we implement learning in an inherently multiclass way, thus providing significant computation time reduction over one-versus-all approaches. Furthermore, the leveraging weights enable effective data selection, thus reducing the cost of k-NN search at classification time. On the other hand, we propose a kernel generalization of our approach to take into account real-valued similarities between data in the feature space, thus enabling more accurate estimation of the local class density. We tested MLNN on three datasets of natural images. Results show that MLNN significantly outperforms classic k-NN and weighted k-NN voting. Furthermore, using an adaptive Gaussian kernel provides significant performance improvement. Finally, the best results are obtained when using MLNN with an appropriate learned metric distance.
Proceedings of the 10th Asian Conference on Computer Vision, ACCV 2010, November 8-12, 2010, Queenstown, New Zealand
Queenstown, New Zealand

hal-00664606
https://hal.inria.fr/hal-00664606
https://hal.inria.fr/hal-00664606/document
https://hal.inria.fr/hal-00664606/file/mlnn_accvfinal.pdf

Éditeur(s) : HAL CCSD Elsevier
Résumé : International audience
Voting rules relying on k-nearest neighbors (k-NN) are an effective tool in countless many machine learning techniques. Thanks to its simplicity, k-NN classification is very attractive to practitioners, as it enables very good performances in several practical applications. However, it suffers from various drawbacks, like sensitivity to "noisy" instances and poor generalization properties when dealing with sparse high-dimensional data. In this paper, we tackle the k-NN classification problem at its core by providing a novel k-NN boosting approach. Namely, we propose a supervised learning algorithm, called Universal Nearest Neighbors (UNN), that induces a leveraged k-NN rule by globally minimizing a surrogate risk upper bounding the empirical misclassification rate over training data. Interestingly, this surrogate risk can be arbitrary chosen from a class of Bregman loss functions, including the familiar exponential, logistic and squared losses. Furthermore, we show that UNN allows to efficiently filter a dataset of instances by keeping only a small fraction of data. Experimental results on the synthetic Ripley's dataset show that such a filtering strategy is able to reject "noisy" examples, and yields a classification error close to the optimal Bayes error. Experiments on standard UCI datasets show significant improvements over the current state of the art.
ISSN: 0925-2312

hal-00664462
https://hal.inria.fr/hal-00664462

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 : International audience
In this paper, we focus on the very specificity of rumors as pieces of information for modeling their process of propagation. We consider a population of pedestrians walking in a city and we assume that a rumor is transmitted by word of mouth from one to another. Although the diffusion of a rumor is of course a multi-dimensional process driven by sociological, economical and psychological elements, in this first step, we emphasize one main dimension of this complex phenomenon only. This dimension is the neighborhood of individuals likely to spread the information. With a confrontation of two antagonistic properties of the neighborhood that are profusion and scarcity of spreaders, we highlight specific characteristics of rumors. This study could lead to the psychological mechanism involved in the decision for a person to become or not a spreader himself/herself. In summary, we study if scarcity could be the silver bullet explaining how a rumor spreads.
ASONAM 2015 : IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining
Paris, France

hal-01252936
https://hal.archives-ouvertes.fr/hal-01252936
DOI : 10.1145/2808797.2809299

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

Éditeur(s) : HAL CCSD
Résumé : International audience
Human spatial motions determine geographic social contacts that influence the way an information is spread on a population or a community. As mobility is a transverse dimension to social practices it is important to better understand its role. With the Eternal-Return model we propose, we simulate an artificial world populated by heterogeneous agents who differ in their mobility. We have chosen a multi-agent framework perspective for this simulation. We endow the agents with simple rules on how to move around the space and how to establish proximity-contacts. This allows to distinguish different kinds of mobile agents, from sedentary ones to travelers. To summarize the dynamics induced by mobility over time, we define the mobility-based Social Proximity Network as being the network of all distinct contacts between agents. Its properties give insight in the process of information spreading. We conduct simulations to understand how an information can be broadcast when agent-nodes are in motion.
IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining
Niagara falls, Canada

hal-00874344
https://hal.archives-ouvertes.fr/hal-00874344

Éditeur(s) : HAL CCSD Wiley
Résumé : International audience
Microalgae in the division Haptophyta play key roles in the marine ecosystem and in global biogeochemical processes. Despite their ecological importance, knowledge on seasonal dynamics, community composition and abundance at the species level is limited due to their small cell size and few morphological features visible under the light microscope. Here, we present unique data on haptophyte seasonal diversity and dynamics from two annual cycles, with the taxonomic resolution and sampling depth obtained with high-throughput sequencing. From outer Oslofjorden, S Norway, nano- and picoplanktonic samples were collected monthly for 2years, and the haptophytes targeted by amplification of RNA/cDNA with Haptophyta-specific 18S rDNA V4 primers. We obtained 156 operational taxonomic units (OTUs), from c. 400.000 454 pyrosequencing reads, after rigorous bioinformatic filtering and clustering at 99.5%. Most OTUs represented uncultured and/or not yet 18S rDNA-sequenced species. Haptophyte OTU richness and community composition exhibited high temporal variation and significant yearly periodicity. Richness was highest in September-October (autumn) and lowest in April-May (spring). Some taxa were detected all year, such as Chrysochromulina simplex, Emiliania huxleyi and Phaeocystis cordata, whereas most calcifying coccolithophores only appeared from summer to early winter. We also revealed the seasonal dynamics of OTUs representing putative novel classes (clades HAP-3-5) or orders (clades D, E, F). Season, light and temperature accounted for 29% of the variation in OTU composition. Residual variation may be related to biotic factors, such as competition and viral infection. This study provides new, in-depth knowledge on seasonal diversity and dynamics of haptophytes in North Atlantic coastal waters. See also the Perspective by Massana
ISSN: 0962-1083

hal-01545321
https://hal.archives-ouvertes.fr/hal-01545321
DOI : 10.1111/mec.13160

Éditeur(s) : HAL CCSD Wiley-Blackwell
Résumé : International audience
Marine protist diversity inventories have largely focused on planktonic environments, while benthic protists have received relatively little attention. We therefore hypothesize that current diversity surveys have only skimmed the surface of protist diversity in marine sediments, which may harbor greater diversity than planktonic environments. We tested this by analyzing sequences of the hypervariable V4 18S rRNA from benthic and planktonic protist communities sampled in European coastal regions. Despite a similar number of OTUs in both realms, richness estimations indicated that we recovered at least 70% of the diversity in planktonic protist communities, but only 33% in benthic communities. There was also little overlap of OTUs between planktonic and benthic communities, as well as between separate benthic communities. We argue that these patterns reflect the heterogeneity and diversity of benthic habitats. A comparison of all OTUs against the Protist Ribosomal Reference database showed that a higher proportion of benthic than planktonic protist diversity is missing from public databases; similar results were obtained by comparing all OTUs against environmental references from NCBI's Short Read Archive. We suggest that the benthic realm may therefore be the world's largest reservoir of marine protist diversity, with most taxa at present undescribed.
ISSN: 0168-6496

hal-01383722
http://hal.upmc.fr/hal-01383722
http://hal.upmc.fr/hal-01383722/document
http://hal.upmc.fr/hal-01383722/file/Forster_2016_Benthic_protists.pdf
DOI : 10.1093/femsec/fiw120

Éditeur(s) : HAL CCSD
Résumé : International audience
The general context of this paper is the Bass model which presented a theory of the adoption of new products. We propose an agent based modelling to allow to model the respective grow of competing products. We assume that there is competition for the same market among two trademarks: each one has its own rate of spontaneous innovation and its own rate of imitation. This paper deals with the relative weight of these competing behaviors on the global dynamics; in particular, we ask the question of the equivalence between mass media influence and word-of-mouth effect.
IEEE RCIS 2014 Eighth International Conference on Research Challenges in Information Science
Marrakech, Morocco

hal-01288467
https://hal.archives-ouvertes.fr/hal-01288467
https://hal.archives-ouvertes.fr/hal-01288467/document
https://hal.archives-ouvertes.fr/hal-01288467/file/poster.pdf

Éditeur(s) : HAL CCSD Oldendorf/Luhe
Résumé : International audience
Large filamentous sulfur-oxidizing bacteria belonging to the Beggiatoacae family can cover large portions of shallow marine sediments surrounding mangroves in Guadeloupe (French West Indies). In order to assess the importance of Beggiatoa mats as an infaunal food source, observations were conducted of the area within mats and at increasing distances from mats. We used natural isotopic compositions and a C-13 enrichment study. Both revealed an ingestion of bacterial mats by associated meiofauna, dominated by rotifers and to a smaller extent by small polychaetes and nematodes. Compared to adjacent sites, sediment covered by bacterial mats presented a higher abundance of diatoms, whereas the total biomass of bacteria did not vary. This constant bacterial abundance suggests that the proportion of organic matter represented by sulfur bacteria is limited compared to the fraction of total bacteria. There was no significant difference in infaunal abundance in mats, suggesting that the availability of this chemosynthetic food resource had a limited local effect. Grazers presented a delta C-13 value increasing with distance from the mat. However, isotopic composition of phospholipid-derived fatty acids specific for diatoms and bacteria revealed that this change is related to modifications of delta C-13 dietary components rather than to changes in diet composition. These complementary methods revealed that the occurrence of sulfur-oxidizing bacterial mats does not necessarily affect grazer abundance and importance of bacteria in their diet. Despite its wide occurrence, Beggiatoa mats would consequently have a minor influence on the structure of the mangrove food web.
ISSN: 0171-8630

hal-01545912
https://hal.archives-ouvertes.fr/hal-01545912
DOI : 10.3354/meps11035

Éditeur(s) : HAL CCSD IEEE
Résumé : International audience
This paper proposes a novel automated approach for the categorization of cells in fluorescence microscopy images. Our supervised classification method aims at recognizing patterns of unlabeled cells based on an annotated dataset. First, the cell images need to be indexed by encoding them in a feature space. For this purpose, we propose tailored bio-inspired features relying on the distribution of contrast information. Then, a supervised learning algorithm is proposed for classifying the cells. We carried out experiments on cellular images related to the diagnosis of autoimmune diseases, testing our classification method on the HEp-2 Cells dataset of Foggia et al (CBMS 2010). Results show classification precision larger than 96% on average, thus confirming promising application of our approach to the challenging application of cellular image classification for computer-aided diagnosis.
ICPR 2012
Tsukuba, Japan

hal-00958856
https://hal.archives-ouvertes.fr/hal-00958856
https://hal.archives-ouvertes.fr/hal-00958856/document
https://hal.archives-ouvertes.fr/hal-00958856/file/icpr_2012.pdf

Éditeur(s) : HAL CCSD Elsevier
Résumé : International audience
Given a graph, its hyperbolicity is a measure of how close its distance distribution is to the one of a tree. This parameter has gained recent attention in the analysis of some graph algorithms and the classification of complex networks. We study on practical improvements for the computation of hyperbolicity in large graphs. Precisely, we investigate on relations between the hyperbolicity of a graph G and the hyperbolicity of its atoms, that are the subgraphs output by the clique-decomposition invented by Tarjan [51, 65]. We prove that the maximum hyperbolicity taken over the atoms is at most one unit off from the hyperbol-icity of G and the bound is sharp. We also give an algorithm to slightly modify the atoms, called the " substitution method " , which is at no extra cost than computing the clique-decomposition, and so that the maximum hyperbolicity taken over the resulting graphs is exactly the hyperbolicity of the input graph G. An experimental evaluation of our method for computing the hyperbolicity of a given graph from its atoms is provided for collaboration networks and biological networks. Finally, on a more theoretical side, we deduce from our results the first linear-time algorithm for computing the hyperbolicity of an outerplanar graph.
ISSN: 0304-3975

hal-01540756
https://hal.inria.fr/hal-01540756
https://hal.inria.fr/hal-01540756/document
https://hal.inria.fr/hal-01540756/file/CCDL-2017.pdf
DOI : 10.1016/j.tcs.2017.06.001

Éditeur(s) : HAL CCSD Springer Verlag
Résumé : International audience
A new genus and species of deep-sea eelpout (Zoarcidae), Barbapellis pterygalces, is herein described from a single adult female specimen carrying eggs and measuring 204 mm in total length. It was collected on the eastern Antarctic continental slope, off Terre Ad,lie, at ca. 1,200 m depth, on muddy bottom. Barbapellis is placed in the subfamily Gymnelinae and can be distinguished from all other genera by morphological characters unusual in zoarcids as well as by molecular barcoding. The genus is characterized by the presence of a complex system of cephalic skin crests and flaps, especially well developed around the mouth. A new terminology is proposed for the eleven crests and sevenfolds described. The skin folds may be used for foraging on muddy sea bottoms. The genus is also characterized by large, flexible, and highly palmate pectoral fins. The designation of the new genus is discussed.
ISSN: 0722-4060

hal-01544913
https://hal.archives-ouvertes.fr/hal-01544913
DOI : 10.1007/s00300-011-1057-3

Éditeur(s) : HAL CCSD Nature Publishing Group
Résumé : International audience
Forest ecosystems have integral roles in climate stability, biodiversity and economic development. Soil stewardship is essential for sustainable forest management. Organic matter (OM) removal and soil compaction are key disturbances associated with forest harvesting, but their impacts on forest ecosystems are not well understood. Because microbiological processes regulate soil ecology and biogeochemistry, microbial community structure might serve as indicator of forest ecosystem status, revealing changes in nutrient and energy flow patterns before they have irreversible effects on long-term soil productivity. We applied massively parallel pyrosequencing of over 4.6 million ribosomal marker sequences to assess the impact of OM removal and soil compaction on bacterial and fungal communities in a field experiment replicated at six forest sites in British Columbia, Canada. More than a decade after harvesting, diversity and structure of soil bacterial and fungal communities remained significantly altered by harvesting disturbances, with individual taxonomic groups responding differentially to varied levels of the disturbances. Plant symbionts, like ectomycorrhizal fungi, and saprobic taxa, such as ascomycetes and actinomycetes, were among the most sensitive to harvesting disturbances. Given their significant ecological roles in forest development, the fate of these taxa might be critical for sustainability of forest ecosystems. Although abundant bacterial populations were ubiquitous, abundant fungal populations often revealed a patchy distribution, consistent with their higher sensitivity to the examined soil disturbances. These results establish a comprehensive inventory of bacterial and fungal community composition in northern coniferous forests and demonstrate the long-term response of their structure to key disturbances associated with forest harvesting. The ISME Journal (2012) 6, 2199-2218; doi:10.1038/ismej.2012.84; published online 2 August 2012
ISSN: 1751-7362

hal-01546179
https://hal.archives-ouvertes.fr/hal-01546179
DOI : 10.1038/ismej.2012.84

Éditeur(s) : HAL CCSD Oldendorf/Luhe
Résumé : International audience
Non-calcifying photosynthetic anthozoans have emerged as a group that may thrive under high carbon dioxide partial pressure (pCO(2)) conditions via increased productivity. However, the physiological mechanisms underlying this potential success are unclear. Here we investigated the impact of high pCO(2) on the dissolved inorganic carbon (DIC) use in the temperate sea anemone Anemonia viridis. We assessed the impacts of long-term exposure to high pCO(2), i.e. sampling in situ natural CO2 vents (Vulcano, Italy), and short-term exposure, i.e. during a 3 wk controlled laboratory experiment. We focused on photo-physiological parameters (net photosynthesis rates, chlorophyll a content and Symbiodinium density) and on carbonic anhydrase (CA) activity, an enzyme involved in the energy-demanding process of DIC absorption. Long-term exposure to high pCO(2) had no impact on Symbiodinium density and chlorophyll a content. In contrst, short-term exposure to high pCO(2) induced a significant reduction in Symbiodinium density, which together with unchanged net photosynthesis resulted in the increase of Symbiodinium productivity per cell. Finally, in both in situ long-term and laboratory short-term exposure to high pCO(2), we observed a significant decrease in the CA activity of sea anemones, suggesting a change in DIC use (i.e. from an HCO3- to a CO2 user). This change could enable a shift in the energy budget that may increase the ability of non-calcifying photosynthetic anthozoans to cope with ocean acidification.
ISSN: 0171-8630

hal-01546163
https://hal.archives-ouvertes.fr/hal-01546163
DOI : 10.3354/meps11916

Éditeur(s) : HAL CCSD Public Library of Science
Résumé : International audience
The effects of biodiversity on the transmission of infectious diseases now stand as a cornerstone of many public health policies. The upper Amazonia and Guyana shield are hot-spots of biodiversity that offer genuine opportunities to explore the relationship between the risk of transmission of Chagas disease and the diversity of its triatomine vectors. Over 730 triato-mines were light-trapped in four geomorphological landscapes shaping French-Guiana, and we determined their taxonomic status and infection by Trypanosoma cruzi. We used a model selection approach to unravel the spatial and temporal variations in species abundance , diversity and infection. The vector community in French-Guiana is typically made of one key species (Panstrongylus geniculatus) that is more abundant than three secondary species combined (Rhodnius pictipes, Panstrongylus lignarius and Eratyrus mucronatus), and four other species that complete the assemblage. Although the overall abundance of adult triatomines does not vary across French-Guiana, their diversity increases along a coastal-inland gradient. These variations unravelled a non-monotonic relationship between vector biodiversity and the risk of transmission of Chagas disease, so that intermediate biodiversity levels are associated with the lowest risks. We also observed biannual variations in triatomine abundance, representing the first report of a biannual pattern in the risk of Cha-gas disease transmission. Those variations were highly and negatively correlated with the average monthly rainfall. We discuss the implications of these patterns for the transmission of T. cruzi by assemblages of triatomine species, and for the dual challenge of controlling Amazonian vector communities that are made of both highly diverse and mostly intrusive species.
ISSN: 1935-2727

Droits : http://creativecommons.org/licenses/by/
hal-01298091
https://hal-univ-perp.archives-ouvertes.fr/hal-01298091
https://hal-univ-perp.archives-ouvertes.fr/hal-01298091/document
https://hal-univ-perp.archives-ouvertes.fr/hal-01298091/file/journal.pntd.0004427.PDF
DOI : 10.1371/journal.pntd.0004427

Éditeur(s) : HAL CCSD Institute of Electrical and Electronics Engineers
Résumé : International audience
Tailoring nearest neighbors algorithms to boosting is an important problem. Recent papers study an approach, UNN, which provably minimizes particular convex surrogates under weak assumptions. However, numerical issues make it necessary to experimentally tweak parts of the UNN algorithm, at the possible expense of the algorithm's convergence and performance. In this paper, we propose a lightweight alternative algorithm optimizing proper scoring rules from a very broad set, and establish formal convergence rates under the boosting framework that surprisingly compete with those known for UNN. It is an adaptive Newton-Raphson algorithm, which belongs to the same lineage as the popular Gentle Adaboost. To the best of our knowledge, no such boosting-compliant convergence rates were previously known for these algorithms. We provide experiments on a dozen domains, including the challenging Caltech and SUN computer vision databases. They display that GNNB significantly outperforms UNN, both in terms of convergence rate and quality of the solution obtained, and GNNB provides a simple and efficient contender to techniques that can be used on very large domains, like stochastic gradient descent -- for which little is known to date. Experiments include a divide-and-conquer improvement of GNNB which exploits the link with proper scoring rules optimization.
ISSN: 0162-8828

hal-00958809
https://hal.archives-ouvertes.fr/hal-00958809
https://hal.archives-ouvertes.fr/hal-00958809/document
https://hal.archives-ouvertes.fr/hal-00958809/file/PAMI.pdf
DOI : 10.1109/TPAMI.2014.2307877

Éditeur(s) : HAL CCSD
Résumé : International audience
It is an admitted fact that mainstream boosting algorithms like AdaBoost do not perform well to estimate class conditional probabilities. In this paper, we analyze, in the light of this problem, a recent algorithm, unn, which leverages nearest neighbors while minimizing a convex loss. Our contribution is threefold. First, we show that there exists a subclass of surrogate losses, elsewhere called balanced, whose minimization brings simple and statistically efficient estimators for Bayes posteriors. Second, we show explicit convergence rates towards these estimators for unn, for any such surrogate loss, under a Weak Learning Assumption which parallels that of classical boosting results. Third and last, we provide experiments and comparisons on synthetic and real datasets, including the challenging SUN computer vision database. Results clearly display that boosting nearest neighbors may provide highly accurate estimators, sometimes more than a hundred times more accurate than those of other contenders like support vector machines.
European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases (ECML PKDD)
Bristol, United Kingdom

hal-00702771
https://hal.archives-ouvertes.fr/hal-00702771
https://hal.archives-ouvertes.fr/hal-00702771/document
https://hal.archives-ouvertes.fr/hal-00702771/file/ecml12-dnbnb-sub.pdf

Éditeur(s) : HAL CCSD Nature Publishing Group
Résumé : International audience
ISSN: 1752-0894

hal-01667978
https://hal.archives-ouvertes.fr/hal-01667978
DOI : 10.1038/s41561-017-0028-x

Éditeur(s) : HAL CCSD Springer
Résumé : Third International Workshop, MLMI 2012, Held in Conjunction with MICCAI 2012
International audience
Universal Nearest Neighbours (unn) is a classifier recently proposed, which can also effectively estimates the posterior probability of each classification act. This algorithm, intrinsically binary, requires the use of a decomposition method to cope with multiclass problems, thus reducing their complexity in less complex binary subtasks. Then, a reconstruction rule provides the final classification. In this paper we show that the application of unn algorithm in conjunction with a reconstruction rule based on the posterior probabilities provides a classification scheme robust among different biomedical image datasets. To this aim, we compare unn performance with those achieved by Support Vector Machine with two different kernels and by a k Nearest Neighbours classifier, and applying two different reconstruction rules for each of the aforementioned classification paradigms. The results on one private and five public biomedical datasets show satisfactory performance.
Machine Learning in Medical Imaging
Nice, France

hal-00958909
https://hal.archives-ouvertes.fr/hal-00958909
https://hal.archives-ouvertes.fr/hal-00958909/document
https://hal.archives-ouvertes.fr/hal-00958909/file/Miccai.pdf
DOI : 10.1007/978-3-642-35428-1_15