HAL - Lille 3
Not a member yet
5967 research outputs found
Sort by
AN EMBEDDING MECHANISM FOR NATURAL STEGANOGRAPHY AFTER DOWN-SAMPLING
No full textInternational audienceNatural Steganography (NS) uses the concept of cover-source switching to provide good undetectability performances [1]. The sensor noise of the source (camera) for a given ISO sensitivity ISO1 is first modeled as an independent Gaussian distribution for each photo-site, then the embedding mimics a switch to another sensitivity ISO2(> ISO1). Because the embedding has to be performed on developed images, we investigate in this paper how to generate a stego-signal once the image is down-sampled. By studying different down-sampling mechanisms (sub-sampling, box down-sampling, tent down-sampling) applied on RAW images generated from a monochrome sensor, we show that the use of a convolution kernel with overlapping boundaries implies a synchronization mechanism similar to the one used by the CMD and Synch embedding schemes, but motivated here by statistical foundations. For each mechanism and scaling factor, we also compute the associated embedding rates and show that our results are in-line with experimental results previously highlighted for other steganographic schemes
Proving Partial Correctness Beyond Programs
No full textInternational audiencePartial correctness is perhaps the most important functional property of algo-rithmic programs. It can be broadly stated as: on all terminating executions, a given relation holds between a program's inputs and outputs. It has been formalised in several logics, from, e.g, Hoare logics [1] to temporal logics [2]. Partial correctness is also a relevant property for any class of specification that has a notion of terminating execution. For example, communication protocols have both nonterminating executions (all messages are forever lost and resent) and terminating executions (all messages sent are properly received). Here, partial correctness may, for instance, require that on all terminating executions , the set of messages corresponding to the transmission of a given file are received and the reception of every message is acknowledged. How can one naturally specify such generic partial-correctness properties, and how can one verify them in a maximally trustworthy manner? One possibility would be to use Hoare logics, but that solution is not optimal because Hoare logics intrinsically require programs (as their deduction rules focus on how program-instructions modify logical predicates), and we are targeting systems specified in formalisms that are not programs but more abstract models, e.g., one more naturally specifies communication protocols in some version of state-transition systems. Another possibility is to state partial-correctness properties in temporal logic and to use a model checker to prove the temporal formula on a state-transition system specification. This is a better solution, since it stays at a model-abstraction level; however, as we are aiming at trustworthy verification, this is not satisfactory as even in case of a successful verification, one's trust in the result is limited as one does not get independently-checkable verification certificates. Moreover, model checkers are limited to essentially finite-state systems (perhaps up to some data abstraction), a limitation we want to avoid
Memory Bandits: a Bayesian approach for the Switching Bandit Problem
No full textInternational audienceThe Thompson Sampling exhibits excellent results in practice and it has been shown to be asymptotically optimal. The extension of Thompson Sampling algorithm to the Switching Multi-Armed Bandit problem, proposed in [13], is a Thompson Sampling equiped with a Bayesian online change point detector [1]. In this paper, we propose another extension of this approach based on a Bayesian aggregation framework. Experiments provide some evidences that in practice, the proposed algorithm compares favorably with the previous version of Thompson Sampling for the Switching Multi-Armed Bandit Problem, while it outperforms clearly other algorithms of the state-of-the-art
Quantifying the dynamics of emotional expressions in family therapy of patients with anorexia nervosa
No full textInternational audienc
Analyse de données pharmacogénomiques et moléculaires pour comprendre la résistance aux traitements des cancers du sein triple négatif
No full textGiven the large number of treatment-resistant triple-negative breast cancers, it is essential to understand the mechanisms of resistance and to find new effective molecules. First, we analyze two large-scale pharmacogenomic datasets. We propose a novel classification based on transcriptomic profiles of cell lines, according to a biological network-driven gene selection process. Our molecular classification shows greater homogeneity in drug response than when cell lines are grouped according to their original tissue. It also helps identify similar patterns of treatment response. In a second analysis, we study a cohort of patients with triple-negative breast cancer who have resisted to neoadjuvant chemotherapy. We perform complete molecular analyzes based on RNAseq and WES. We observe a high molecular heterogeneity of tumors before and after treatment. Although we highlighted clonal evolution under treatment, no recurrent mechanism of resistance could be identified Our results strongly suggest that each tumor has a unique molecular profile and that that it is increasingly important to have large series of tumors. Finally, we are improving a method for testing the overrepresentation of known RNA binding protein motifs in a given set of regulated sequences. This tool uses an innovative approach to control the proportion of false positives that is not realized by the existing algorithm. We show the effectiveness of our approach using two different datasets.Devant le grand nombre de tumeurs du sein triple négatif résistant aux traitements, il est essentiel de comprendre les mécanismes de résistance et de trouver de nouvelles molécules efficaces. En premier lieu, nous analysons deux ensembles de données pharmacogénomiques à grande échelle. Nous proposons une nouvelle classification basée sur des profils transcriptomiques de lignées cellulaires, selon un processus de sélection de gènes basé sur des réseaux biologiques. Notre classification moléculaire montre une plus grande homogénéité dans la réponse aux médicaments que lorsque l’on regroupe les lignées cellulaires en fonction de leur tissu d'origine. Elle permet également d’identifier des profils similaires de réponse aux traitements. Dans un second travail, nous étudions une cohorte de patients atteints d’un cancer du sein triple négatif ayant résisté à la chimiothérapie néoadjuvante. Nous effectuons des analyses moléculaires complètes basées sur du RNAseq et WES. Nous constatons une forte hétérogénéité moléculaire des tumeurs avant et après traitement. Bien que nous observons une évolution clonale sous traitement, aucun mécanisme récurrent de résistance n’a pu être identifié. Nos résultats suggèrent fortement que chaque tumeur a un profil moléculaire unique et qu'il est important d'étudier de grandes séries de tumeurs. Enfin, nous améliorons une méthode pour tester la surreprésentation de motifs connus de protéines de liaison à l'ARN, dans un ensemble donné de séquences régulées. Cet outil utilise une approche innovante pour contrôler la proportion de faux positifs qui n'est pas réalisé par l'algorithme existant. Nous montrons l'efficacité de notre approche en utilisant deux séries de données différentes
Importance des catégories socioprofessionnelles dans le rapport cognitif et émotionnel au cancer du sein et à sa prise en charge
No full textInternational audienc
Discrepancy Between Clinical and Pathologic Nodal Status of Esophageal Cancer and Impact on Prognosis and Therapeutic Strategy
No full textInternational audienceBACKGROUND: The impact of discrepancies between clinical (c) and pathologic (p) stages of esophageal cancer remains a poorly understood issue. This study aimed to compare the prognosis of patient groups treated by primary surgery including clinical N0/pathologic N0 (cN0pN0), clinical N0/pathologic N+ (cN0pN+), clinical N+/pathologic N0 (cN+pN0), and clinical N+/pathologic N+ (cN+pN+).METHODS: Data were collected from 30 European centers during the years 2000 to 2010. Among 2944 recruited patients, 1554 patients receiving primary surgery met the inclusion criteria including 613 cN0pN0, 403 cN0pN+, 220 cN+pN0, and 318 cN+pN+ patients. Analyses with adjustment of the propensity score were used to compensate for differences in baseline characteristics.RESULTS: Clinical T stages 3 and 4 were increased in cN+pN+ (73.0%), cN0pN+ (49.6%), and cN+pN0 (51.8%) compared with cN0pN0 (32.8%). Compared with cN0pN0, cN+pN+ and cN0pN+ showed an increase in the proportion of adenocarcinoma histologic subtype, poor tumor differentiation, pathologic T3 and T4 stages, and R1/2 resection margin. Adjusted 5-year overall survival (hazard ratio [HR] 3.12; 95% confidence interval [CI] 2.57-3.78; P < 0.001) and event-free survival (HR 2.87; 95% CI 2.39-3.45; P < 0.001) were significantly reduced in cN0pN+ compared with cN0pN0. No significant differences in 5-year overall survival or event-free survival between cN0pN+ and cN+pN+ were observed. Regression analysis identified an association of distal tumor location, advanced clinical T stage, and poor tumor differentiation with pN+ disease.CONCLUSIONS: This large multicenter study showed that cN0pN+ has a prognosis similar to that of cN+pN+ and worse than that of cN0pN0. Patients with clinical N0 disease but risk factors for pathologic N+ disease may benefit from neoadjuvant therapy before surgery
Survival benefit of neoadjuvant treatment in clinical T3N0M0 esophageal cancer: results from a retrospective multicenter European study
No full textInternational audienc
Multi-Objective Extremal Optimization in Processor Load Balancing for Distributed Programs
No full textInternational audienc
Patient navigation to reduce social inequalities in colorectal cancer screening participation: A cluster randomized controlled trial
No full textInternational audienceDespite free colorectal cancer screening in France, participation remains low and low socioeconomic status is associated with a low participation. Our aim was to assess the effect of a screening navigation program on participation and the reduction in social inequalities in a national-level organized mass screening program for colorectal cancer by fecal-occult blood test (FOBT). A multicenter (3 French departments) cluster randomized controlled trial was conducted over two years. The cluster was a small geographical unit stratified according to a deprivation index and the place of residence. A total of 14,556 subjects (72 clusters) were included in the control arm where the FOBT program involved the usual postal reminders, and 14,373 subjects (66 clusters) were included in the intervention arm. Intervention concerned only non-attended subjects with a phone number available defined as the navigable population. A screening navigator was added to the usual screening organization to identify and eliminate barriers to CRC screening with personalized contact. The participation rate by strata increased in the intervention arm. The increase was greater in affluent strata than in deprived ones. Multivariate analyses demonstrated that the intervention mainly with phone navigation increased individual participation (OR=1.19 [1.10, 1.29]) in the navigable population. For such interventions to reduce social inequalities in a country with a national level organized mass screening program, they should first be administered to deprived populations, in accordance with the principle of proportionate universalism. ClinicalTrials.gov Identifier: NCT01555450