1,720,967 research outputs found
Graphical Model approaches for Biclustering
Get PDFIn many scientific areas, it is crucial to group (cluster) a set of objects, based on a set of observed features. Such operation is widely known as Clustering and it has been exploited in the most different scenarios ranging from Economics to Biology passing through Psychology. Making a step forward, there exist contexts where it is crucial to group objects and simultaneously identify the features that allow to recognize such objects from the others. In gene expression analysis, for instance, the identification of subsets of genes showing a coherent pattern of expression in subsets of objects/samples can provide crucial information about active biological processes. Such information, which cannot be retrieved by classical clustering approaches, can be extracted with the so called Biclustering, a class of approaches which aim at simultaneously clustering both rows and columns of a given data matrix (where each row corresponds to a different object/sample and each column to a different feature). The problem of biclustering, also known as co-clustering, has been recently exploited in a wide range of scenarios such as Bioinformatics, market segmentation, data mining, text analysis and recommender systems. Many approaches have been proposed to address the biclustering problem, each one characterized by different properties such as interpretability, effectiveness or computational complexity. A recent trend involves the exploitation of sophisticated computational models (Graphical Models) to face the intrinsic complexity of biclustering, and to retrieve very accurate solutions. Graphical Models represent the decomposition of a global objective function to analyse in a set of smaller/local functions defined over a subset of variables. The advantages in using Graphical Models relies in the fact that the graphical representation can highlight useful hidden properties of the considered objective function, plus, the analysis of smaller local problems can be dealt with less computational effort. Due to the difficulties in obtaining a representative and solvable model, and since biclustering is a complex and challenging problem, there exist few promising approaches in literature based on Graphical models facing biclustering. 3 This thesis is inserted in the above mentioned scenario and it investigates the exploitation of Graphical Models to face the biclustering problem. We explored different type of Graphical Models, in particular: Factor Graphs and Bayesian Networks. We present three novel algorithms (with extensions) and evaluate such techniques using available benchmark datasets. All the models have been compared with the state-of-the-art competitors and the results show that Factor Graph approaches lead to solid and efficient solutions for dataset of contained dimensions, whereas Bayesian Networks can manage huge datasets, with the overcome that setting the parameters can be not trivial. As another contribution of the thesis, we widen the range of biclustering applications by studying the suitability of these approaches in some Computer Vision problems where biclustering has been never adopted before. Summarizing, with this thesis we provide evidence that Graphical Model techniques can have a significant impact in the biclustering scenario. Moreover, we demonstrate that biclustering techniques are ductile and can produce effective solutions in the most different fields of applications
Biclustering gene expressions using factor graphs and the max-sum algorithm
No full textBiclustering is an intrinsically challenging andhighly complex problem, particularly studied in thebiology field, where the goal is to simultaneouslycluster genes and samples of an expression data matrix.In this paper we present a novel approach togene expression biclustering by providing a binaryFactor Graph formulation to such problem. In moredetail, we reformulate biclustering as a sequentialsearch for single biclusters and use an efficient optimizationprocedure based on the Max Sum algorithm.Such approach, drastically alleviates thescaling issues of previous approaches for biclusteringbased on Factor Graphs obtaining significantlymore accurate results on synthetic datasets. A furtheranalysis on two real-world datasets confirmsthe potentials of the proposed methodology whencompared to alternative state of the art methods
A biclustering approach based on factor graphs and the max-sum algorithm
No full textBiclustering represents an intrinsically complex problem, where the aim is to perform a simultaneous row- and column-clustering of a given data matrix. Some recent approaches model this problem using factor graphs, so to exploit their ability to open the door to efficient optimization approaches for well designed function decompositions. However, while such models provide promising results, they do not scale to data matrices of reasonable size. In this paper, we take a step towards addressing this issue, by proposing a novel approach to biclustering based on factor graphs, which yields high quality solutions and scales more favorably than previous methods. Specifically, we cast biclustering as the sequential search for a single bicluster, and propose a binary and compact factor graph that can be solved efficiently using the max-sum algorithm. The proposed approach has been tested and compared with state-of-the-art methods on four datasets (two synthetic and two real world data), providing encouraging results with respect both to previous approaches based on factor graphs and to other state-of-the-art method
The multi-modal universe of fast-fashion: the Visuelle 2.0 benchmark
Get PDFWe present Visuelle 2.0, the first dataset useful for facing diverse
prediction problems that a fast-fashion company has to manage routinely.
Furthermore, we demonstrate how the use of computer vision is substantial in
this scenario. Visuelle 2.0 contains data for 6 seasons / 5355 clothing
products of Nuna Lie, a famous Italian company with hundreds of shops located
in different areas within the country. In particular, we focus on a specific
prediction problem, namely short-observation new product sale forecasting
(SO-fore). SO-fore assumes that the season has started and a set of new
products is on the shelves of the different stores. The goal is to forecast the
sales for a particular horizon, given a short, available past (few weeks),
since no earlier statistics are available. To be successful, SO-fore approaches
should capture this short past and exploit other modalities or exogenous data.
To these aims, Visuelle 2.0 is equipped with disaggregated data at the
item-shop level and multi-modal information for each clothing item, allowing
computer vision approaches to come into play. The main message that we deliver
is that the use of image data with deep networks boosts performances obtained
when using the time series in long-term forecasting scenarios, ameliorating the
WAPE and MAE by up to 5.48% and 7% respectively compared to competitive
baseline methods. The dataset is available at
https://humaticslab.github.io/forecasting/visuelleComment: Accepted at the 5th Workshop on Computer Vision for Fashion, Art, and
Design @ CVPR2
Biclustering of expression microarray data using Affinity Propagation
No full textBiclustering, namely simultaneous clustering of genes and samples, represents a challenging and important research line in the expression microarray data analysis. In this paper, we investigate the use of Affinity Propagation, a popular clustering method, to perform biclustering. Specifically, we cast Affinity Propagation into the Couple Two Way Clustering scheme, which allows to use a clustering technique to perform biclustering. We extend the CTWC approach, adapting it to Affinity Propagation, by introducing a stability criterion and by devising an approach to automatically assemble couples of stable clusters into biclusters. Empirical results, obtained in a synthetic benchmark for biclustering, show that our approach is extremely competitive with respect to the state of the art, achieving an accuracy of 91% in the worst case performance and 100% accuracy for all tested noise levels in the best case
Biclustering of time series data using factor graphs
No full textBiclustering regards the simultaneous clustering of both rows and columns of a given data matrix. A specific applica- tion scenario for biclustering techniques concerns the anal- ysis of gene expression time-series data, wherein columns dataset are temporally related. In this context, bicluster- ing solutions should involve subset of genes sharing ‘simi- lar’ behaviours among consecutive experimental conditions. Due to the intrinsic spatial constraint required by time-series dataset, current Factor Graph (FG) based approaches can- not be applied. In this paper we introduce Time-Series constraints forcing biclustering solution to have contiguous columns. We optimize the model by using the Max-Sum algorithm, whose message update rules have been derived exploiting The Higher Order Potentials (THOP). The pro- posed method has been assessed on a real world dataset and the retrieved biclusters show that it can provide accurate and biologically relevant solutions
Well Googled is Half Done: Multimodal Forecasting of New Fashion Product Sales with Image-based Google Trends
Get PDFNew fashion product sales forecasting is a challenging problem that involves
many business dynamics and cannot be solved by classical forecasting
approaches. In this paper, we investigate the effectiveness of systematically
probing exogenous knowledge in the form of Google Trends time series and
combining it with multi-modal information related to a brand-new fashion item,
in order to effectively forecast its sales despite the lack of past data. In
particular, we propose a neural network-based approach, where an encoder learns
a representation of the exogenous time series, while the decoder forecasts the
sales based on the Google Trends encoding and the available visual and metadata
information. Our model works in a non-autoregressive manner, avoiding the
compounding effect of large first-step errors. As a second contribution, we
present VISUELLE, a publicly available dataset for the task of new fashion
product sales forecasting, containing multimodal information for 5577 real, new
products sold between 2016-2019 from Nunalie, an Italian fast-fashion company.
The dataset is equipped with images of products, metadata, related sales, and
associated Google Trends. We use VISUELLE to compare our approach against
state-of-the-art alternatives and several baselines, showing that our neural
network-based approach is the most accurate in terms of both percentage and
absolute error. It is worth noting that the addition of exogenous knowledge
boosts the forecasting accuracy by 1.5% in terms of Weighted Absolute
Percentage Error (WAPE), revealing the importance of exploiting informative
external information. The code and dataset are both available at
https://github.com/HumaticsLAB/GTM-Transformer.Comment: Accepted by the Wiley Journal of Forecastin
Disentangled Latent Spaces Facilitate Data-Driven Auxiliary Learning
No full textLearning to solve auxiliary tasks concurrently with a principal task of interest can improve performance when data is scarce or the principal task is complex. This idea is inspired by the improved generalization capability induced by solving multiple tasks simultaneously, leading to a robust shared representation. However, selecting optimal auxiliary tasks typically requires manual design or costly meta-learning approaches. We propose Detaux, a framework that discovers an unrelated auxiliary classification task via weakly supervised disentanglement at the representation level. Isolating variations relevant to the principal task in one subspace while generating orthogonal subspaces with high separability allows us to discover auxiliary labels by clustering in these subspaces, allowing a transition from Single-Task Learning (STL) to Multi-Task Learning (MTL). In particular, the original labels associated with the principal task and the newly discovered ones can be fed into any MTL framework. Experiments and ablation studies highlight the effectiveness of Detaux and reveal an unexplored link between disentangled representations and MTL. The source code is available at https://github.com/intelligolabs/Detaux
A binary Factor Graph model for biclustering
No full textBiclustering, which can be defined as the simultaneous clustering of rows and columns in a data matrix, has received increasing attention in recent years, particularly in the field of Bioinformatics (e.g. for the analysis of microarray data). This paper proposes a novel biclustering approach, which extends the Affinity Propagation [Frey 07] clustering algorithm to the biclustering case. In particular, we propose a new exemplar based model, encoded as a binary factor graph, which allows to cluster rows and columns simultaneously. Moreover, we propose a linear formulation of such model to solve the optimization problem using Linear Programming techniques. The proposed approach has been tested by using a well known synthetic microarray benchmark, with encouraging results
- …
