Littérature scientifique sur le sujet « Post-clustering inference »

An Adaptive Neuro-Fuzzy Inference System, based on a jack-knife approach, is proposed for the post-calibration of weather radar rainfall estimation exploiting available raingauge observations. The methodology relies on the construction of a fuzzy inference system with three inputs (radar x coordinate, y coordinate and rainfall estimation at raingauge locations) and one output (raingauge observations). Subtractive clustering is used to generate the initial fuzzy inference system. Artificial neural network learning provides a fast way to automatically generate additional fuzzy rules and membership functions for the fuzzy inference system. Fuzzy logic enhances the generalisation of the artificial neural network system. In order to demonstrate the steps of the radar rainfall post-calibration using the Adaptive Neuro-Fuzzy Inference System, CAPPIs of one-hour rainfall accumulation and corresponding raingauge observations have been selected. Results show that the proposed approach looks for a response that is a compromise between radar rainfall estimations and raingauge observations and does not necessarily consider the raingauge observations as ground truth. The algorithm is very fast and can be implemented for real time post-calibration. This algorithm makes use of all available data—raingauge observations are usually scarce—for training and checking the neuro-fuzzy inference system. It also provides a degree of reliability of the post-calibration.

The random effect approach for meta-analysis was motivated by a lack of consistent assessment of homogeneity of treatment effect before pooling. The random effect model assumes that the distribution of the treatment effect is fully heterogenous across the experiments. However, other models arising by grouping some of the experiments are plausible. We illustrate on simulated binary experiments that the fully heterogenous model gives a poor meta-inference when fully heterogeneity is not the true model and that the knowledge of the true cluster model considerably improves the inference. We propose the use of a Bayesian model selection procedure for estimating the true cluster model, and Bayesian model averaging to incorporate into the meta-analysis the clustering estimation. A well-known meta-analysis for six major multicentre trials to assess the efficacy of a given dose of aspirin in post-myocardial infarction patients is reanalysed.

Background: Today, the usage of artificial intelligence systems and computational intelligence is increasing. This study aimed to determine the fuzzy system algorithms to model and predict the amount of air pollution based on the measured data in subway stations. Methods: In this study, first, the effective variables on the concentration of particulate matter were determined in metro stations. Then, PM2.5, PM10, and total size particle (TSP) concentrations were measured. Finally, the particles’ concentration was modeled using fuzzy systems, including the fuzzy inference system (FIS) and adaptive neuro-fuzzy inference system (ANFIS). Results: It was revealed that FIS with modes gradient segmentation (FIS-GS) could predict 76% and ANFIS-FCM with modes of clustering and post-diffusion training algorithm (CPDTA) could predict 85% of PM2.5, PM10, and TSP particle concentrations. Conclusion: According to the results, among the models studied in this work, ANFIS-FCM-CPDTA, due to its better ability to extract knowledge and ambiguous rules of the fuzzy system, was considered a suitable model.

A GIS-based approach is used in this study to obtain a better LCZ map of Berlin in comparison to the remote-sensing-based WUDAPT L0 approach. The LCZ classification of land use/cover can be used, among other applications, to characterize the urban heat island. An improved fuzzy logic method is employed for the purpose of classification of the zone properties to yield the GIS-LCZ map over 100 m × 100 m grid tiles covering the Berlin region. The zone properties are calculated from raster and vector datasets with the aids of the urban multi-scale environmental predictor (UMEP), QGIS and Python scripts. The standard framework is modified by reducing the threshold for the zone property impervious fraction for LCZ E to better detect paved surfaces in urban areas. Another modification is the reduction in the window size in the majority filter during post-processing, compared to the WUDAPT L0 method, to retain more details in the GIS-LCZ map. Moreover, new training areas are generated considering building height information. The result of the GIS-LCZ approach is compared to the new training areas for accuracy assessment, which shows better overall accuracy compared to that of the WUDAPT L0 method. The new training areas are also submitted to the LCZ generator and the resulting LCZ-map gives a better overall accuracy value compared to the previous (WUDAPT) submission. This study shows one shortcoming of the WUDAPT L0 method: it does not explicitly use building height information and that leads to misclassification of LCZs in several cases. The GIS-LCZ method addresses this shortcoming effectively. Finally, an unsupervised machine learning method, k-means clustering, is applied to cluster the grid tiles according to their zone properties into custom classes. The custom clusters are compared to the GIS-LCZ classes and the results indicate that k-means clustering can identify more complex city-specific classes or LCZ transition types, while the GIS-LCZ method always divides regions into the standard LCZ classes.

In response to the current issues of poor real-time performance, high computational costs, and excessive memory usage of object detection algorithms based on deep convolutional neural networks in embedded devices, a method for improving deep convolutional neural networks based on model compression and knowledge distillation is proposed. Firstly, data augmentation is employed in the preprocessing stage to increase the diversity of training samples, thereby improving the model’s robustness and generalization capability. The K-means++ clustering algorithm generates candidate bounding boxes, adapting to defects of different sizes and selecting finer features earlier. Secondly, the cross stage partial (CSP) Darknet53 network and spatial pyramid pooling (SPP) module extract features from the input raw images, enhancing the accuracy of defect location detection and recognition in YOLO. Finally, the concept of model compression is integrated, utilizing scaling factors in the batch normalization (BN) layer, and introducing sparse factors to perform sparse training on the network. Channel pruning and layer pruning are applied to the sparse model, and post-processing methods using knowledge distillation are used to effectively reduce the model size and forward inference time while maintaining model accuracy. The improved model size decreases from 244 M to 4.19 M, the detection speed increases from 32.8 f/s to 68 f/s, and mAP reaches 97.41. Experimental results demonstrate that this method is conducive to deploying network models on embedded devices with limited GPU computing and storage resources. It can be applied in distributed service architectures for edge computing, providing new technological references for deploying deep learning models in the industrial sector.

Although hedgehogs are well-known examples of postglacial recolonisation, the specific processes that shape their population structures have not been examined by detailed sampling and fast-evolving genetic markers in combination with model based clustering methods. This study aims to analyse the impacts of isolation within glacial refugia and of postglacial expansion on the population structure of the Northern White-breasted hedgehog (Erinaceus roumanicus). It also discusses the role of the processes at edges of species distribution in its evolutionary history. The maternally inherited mitochondrial control region and the bi-parentally inherited nuclear microsatellites were used to examine samples within the Central Europe, Balkan Peninsula and adjacent islands. Bayesian coalescent inference and neutrality tests proposed a recent increase in the population size. The most pronounced pattern of population structure involved differentiation of the insular populations in the Mediterranean Sea and the population within the contact zone with E. europaeus in Central Europe. An interspecies hybrid was detected for the first time in Central Europe. A low genetic diversity was observed in Crete, while the highest genetic distances among individuals were found in Romania. The recent population in the post-refugial area related to the Balkan Peninsula shows a complex pattern with pronounced subpopulations located mainly in the Pannonian Basin and at the Adriatic and Pontic coasts. Detailed analyses indicate that parapatry and peripatry may not be the only factors that limit range expansion, but also strong microevolutionary forces that may change the genetic structure of the species. Here we present evidence showing that population differentiation may occur not only during the glacial restriction of the range into the refugia, but also during the interglacial range expansion. Population differentiation at the Balkan Peninsula and adjacent regions could be ascribed to diversification in steppe/forest biomes and complicated geomorphology, including pronounced geographic barriers as Carpathians.

Generalized Category Discovery (GCD) aims to recognize both known and novel categories in an unlabeled dataset by leveraging another labeled dataset with only known categories. Without considering knowledge transfer from known to novel categories, current methods usually perform poorly on novel categories due to the lack of corresponding supervision. To mitigate this issue, we propose a unified Knowledge Transfer Network (KTN), which solves two obstacles to knowledge transfer in GCD. First, the mixture of known and novel categories in unlabeled data makes it difficult to identify transfer candidates (i.e., samples with novel categories). For this, we propose an entropy-based method that leverages knowledge in the pre-trained classifier to differentiate known and novel categories without requiring extra data or parameters. Second, the lack of prior knowledge of novel categories presents challenges in quantifying semantic relationships between categories to decide the transfer weights. For this, we model different categories with prototypes and treat their similarities as transfer weights to measure the semantic similarities between categories. On the basis of two treatments, we transfer knowledge from known to novel categories by conducting pre-adjustment of logits and post-adjustment of labels for transfer candidates based on the transfer weights between different categories. With the weighted adjustment, KTN can generate more accurate pseudo-labels for unlabeled data, which helps to learn more discriminative features and boost model performance on novel categories. Extensive experiments show that our method outperforms state-of-the-art models on all evaluation metrics across multiple benchmark datasets. Furthermore, different from previous clustering-based methods that can only work offline with abundant data, KTN can be deployed online conveniently with faster inference speed. Code and data are available at https://github.com/yibai-shi/KTN.

Abstract BACKGROUND Diffuse Intrinsic Pontine Glioma (DIPG) is a pediatric brain tumor (glioma), resistant to chemotherapy, with only a temporary response to radiotherapy and an extremely bad prognosis. Genomic abnormalities have been found, indicating abnormal activation of certain growth factor signaling pathways, while expression analysis suggests involvement of developmental signaling pathways.10–15 signal transduction pathways can drive cancer growth and metastasis. We have developed, and biologically validated, a method which enables quantitative measurements of functional activity of signal transduction pathways in individual cell/tissue samples, based on Bayesian computational model inference of pathway activity from measurements of mRNA levels of target genes of the transcription factor associated with the respective signalling pathway. A major envisioned clinical utility is prediction of therapy response. MATERIAL AND METHODS For signaling pathway analysis, Affymetrix expression microarray data were available (GEO dataset GSE26576) from 2 normal brain stem samples and from 6 low grade glioma and 26 DIPG samples (post-mortem after therapy). Of one DIPG patient samples were available before and after therapy. Signaling pathway activity scores were calculated for estrogen and androgen receptor, PI3K-FOXO, MAPK-AP1, JAK-STAT, NFκB, Hedgehog (HH), TGFβ, NOTCH and Wnt pathways. PI3K pathway activity is the reverse of FOXO activity, in the absence of oxidative stress (measured by SOD2 expression). Pathway activity scores were compared between normal tissue and low grade glioma samples and DIPG, and k-means cluster analysis was performed on the DIPG pathway activity scores. RESULTS After treatment, HH pathway activity was increased in DIPG compared to low grade glioma (p=0.0003), PI3K pathway activity scores showed large variations in activity in the DIPG group. Tumors with cell cycle (CDK4/6, CCND1-3) or Receptor Tyrosine Kinase-related gene amplifications had higher PI3K and HH pathway activity compared to tumors without identified amplifications (p<0.05) which, in contrast, had higher MAPK-AP1 pathway activity (p<0.002). Pathway-based clustering analysis revealed two DIPG clusters, C1: high TGFβ/MAPK-AP1 and low PI3K/HH pathway activity; C2: low TGFβ/MAPK-AP1, high PI3K/HH pathway activity. C1 best resembled low grade glioma. In the patient with pre/post treatment samples, a C1 pathway profile switched to a C2 profile after treatment. CONCLUSION Using our quantitative analysis of signaling pathway activity in post-treatment DIPG, two pathway activity subtypes were identified, of which the HH/PI3K high, TGFβ low activity subtype was associated with defined gene amplifications, and may have been induced by chemoradiation therapy. Clusters are supported by a clear biological rationale. Identified signaling pathways are potentially drug targetable.

Dans le domaine de la transcriptomique, les avancées technologiques, telles que les puces à ADN et le séquençage à haut-débit, ont permis de quantifier l'expression génique à grande échelle. Ces progrès ont soulevé des défis statistiques, notamment pour l'analyse d'expression différentielle, visant à identifier les gènes différenciant significativement deux populations. Cependant, les procédures classiques d'inférence perdent leurs garanties de contrôle du taux de faux positifs lorsque les biologistes sélectionnent un sous-ensemble de gènes. Les méthodes d'inférence post hoc surmontent cette limitation en garantissant un contrôle sur le nombre de faux positifs, même pour des ensembles de gènes sélectionnés de manière arbitraire. La première contribution de ce manuscrit démontre l'efficacité de ces méthodes pour les données transcriptomiques de deux conditions biologiques, notamment grâce à l'introduction d'un algorithme de calcul des bornes post hoc à complexité linéaire, adapté à la grande dimension des données. Une application interactive a également été développée, facilitant la sélection et l'évaluation simultanée des bornes post hoc pour des ensembles de gènes d'intérêt. Ces contributions sont présentées dans la première partie du manuscrit. L'évolution technologique vers le séquençage en cellule unique a soulevé de nouvelles questions, notamment l'identification des gènes dont l'expression se distingue d'un groupe cellulaire à un (des) autre(s). Cette problématique est complexe car les groupes cellulaires doivent d'abord être estimés par une méthode de clustering, avant d'effectuer un test comparatif, menant ainsi à une analyse circulaire. Dans la seconde partie de ce manuscrit, nous présentons une revue des méthodes d'inférence post-clustering résolvant ce problème ainsi qu'une comparaison numérique des approches multivariées et marginales de comparaison de classes. Enfin, nous explorons comment l'utilisation des modèles de mélange dans l'étape de clustering peut être exploitée dans les tests post-clustering, et nous discutons de perspectives pour l'application de ces tests aux données transcriptomiques
In the field of transcriptomics, technological advances, such as microarrays and high-throughput sequencing, have enabled large-scale quantification of gene expression. These advances have raised statistical challenges, particularly in differential expression analysis, which aims to identify genes that significantly differentiate between two populations. However, traditional inference procedures lose their ability to control the false positive rate when biologists select a subset of genes. Post-hoc inference methods address this limitation by providing control over the number of false positives, even for arbitrary gene sets. The first contribution of this manuscript demonstrates the effectiveness of these methods for the differential analysis of transcriptomic data between two biological conditions, notably through the introduction of a linear-time algorithm for computing post-hoc bounds, adapted to the high dimensionality of the data. An interactive application was also developed to facilitate the selection and simultaneous evaluation of post-hoc bounds for sets of genes of interest. These contributions are presented in the first part of the manuscript. The technological evolution towards single-cell sequencing has raised new questions, particularly regarding the identification of genes whose expression distinguishes one cellular group from another. This issue is complex because cell groups must first be estimated using clustering method before performing a comparative test, leading to a circular analysis. In the second part of this manuscript, we present a review of post-clustering inference methods addressing this problem, as well as a numerical comparison of multivariate and marginal approaches for cluster comparison. Finally, we explore how the use of mixture models in the clustering step can be exploited in post-clustering tests, and discuss perspectives for applying these tests to transcriptomic data