Pattern sampling has emerged as a promising approach for information discovery in large databases, allowing analysts to focus on a manageable subset of patterns. In this approach, patterns are randomly drawn based on an interestingness measure, such as frequency or hyper-volume. This paper presents the first sampling approach designed to handle interval patterns in numerical databases. This approach, named Fips , samples interval patterns proportionally to their frequency. It uses a multi-step sampling procedure and addresses a key challenge in numerical data: accurately determining the number of interval patterns that cover each object. We extend this work with HFips , which samples interval patterns proportionally to both their frequency and hyper-volume. These methods efficiently tackle the well-known long-tail phenomenon in pattern sampling. We formally prove that Fips and HFips sample interval patterns in proportion to their frequency and the product of hyper-volume and frequency, respectively. Through experiments on several databases, we demonstrate the quality of the obtained patterns and their robustness against the long-tail phenomenon.

One of the objectives of the Hérelles project is to find new mechanisms to facilitate the labeling (or semantization) of clusters from time series of satellite images. To achieve this, a proposed solution is to associate textual elements of interest with satellite data. The first step in this process consists of an automatic extraction of the information in the form of rules from urban planning documents composed in the French language. To address this challenge, we propose a method which is based on the multi-label classification of textual segments. It includes a special format for representing segments, in which each segment has a title and a subtitle. In addition, we propose a cascade approach aiming to deal with hierarchy of class labels. Finally, we develop several text augmentation techniques for the texts in French, which are able to improve the prediction results. We demonstrate experimentally that the resulting framework correctly classifies each type of segment with more than 90% of accuracy.

Developing products for improving plant nutritional status, e.g. fertilizers or plant growth regulators, is an important topic to move towards sustainability in agriculture and to ensure to feed the world population. A key challenge is to identify when, what, and how much nutrients to add to plants' growth environment. In this paper, we study a use case on how to characterize rapeseed plant nutrient deficiencies during their growth. A promising approach consists of deriving data from spectroscopy of leaves, and using this representation to predict what kind of deficiency (if any) plants are undergoing. We are considering three research questions: 1) from which day after onset of a nutrient deficiency we can identify it, 2) whether leaves that have sprouted under nutrient-rich conditions can still help in identifying problems. Third, and most importantly, performing the spectroscopy on the full range of wavelengths is expensive, which under production conditions allows for only relatively few samples. We therefore explore how to perform dimensionality reduction and preprocessing to achieve good predictive accuracy. We show that 1) deficiencies can be identified early on, 2) leaf generations help to predict nutrient deficiencies, and 3) that preprocessing increases the accuracy and dimensionality reduction can be performed without loss of accuracy. Along the way, we find that our some of our industry partners' assumptions about the data do not seem to be borne out by our empirical results, and that the subset of data they initially selected turns out to be too easy to model. The full data leads to more informative insights.

Manufacturing enterprises are engaged in implementing new technologies to enhance their manufacturing lines in a smart way. These new technologies give manufacturing enterprises the knowledge, understanding, insight and foresight to improve products, processes and decisions, thereby creating a competitive advantage. In this paper, we explore the use of semantics to enhance deep learning models. We propose an ontology-based LSTM neural network, in which the deep architecture is designed with an ontology to extract high-level cognitive features and stacked LSTM layers for learning temporal dependencies. Our model is applied to a real manufacturing data set with multivariate time series for classification problems. The experiments show that our model can improve performance compared with conventional methods.

Pan Assays Interference Compounds (PAINS) are a significant problem in modern drug discovery: compounds showing non-target specific activity in high-throughput screening can mislead medicinal chemists during hit identification, wasting time and resources. Recent work has shown that existing structural alerts are not up to the task of identifying PAINS. To address this short-coming, we are in the process of developing a tool, PrePeP, that predicts PAINS, and allows experts to visually explore the reasons for the prediction. In the paper, we discuss the different aspects that are involved in developing a functional tool: systematically deriving structural descriptors, addressing the extreme imbalance of the data, offering visual information that pharmacological chemists are familiar with. We evaluate the quality of the approach using benchmark data sets from the literature and show that we correct several short-comings of existing PAINS alerts that have recently been pointed out.

Learning-based image hashing consists in turning high-dimensional image features into compact binary codes, while preserving their semantic similarity (i.e., if two images are close in terms of content, their codes should be close as well). In this context, many existing hashing techniques rely on supervision for preserving these semantic properties. In this paper, we aim at learning such binary codes by exploiting the underlying structure of unlabeled data, using deep learning. The proposed deep network is based on a stacked convolutional autoencoder which hierarchically maps input images into a low-dimensional space. A binary relaxation constraint applied to the middle layer of the network - the one containing the code - makes the codes sparse and binary. To demonstrate the competitiveness of the proposed architecture, we evaluate the so produced hash codes on image retrieval and image classification tasks on the MNIST dataset, and compare its performance with state-of-the-art approaches.

International audience

We introduce the notion of Optimal Patterns (OPs), defined as the best patterns according to a given user preference, and show that OPs encompass many data mining problems. Then, we propose a generic method based on a Dynamic Constraint Satisfaction Problem to mine OPs, and we show that any OP is characterized by a basic constraint and a set of constraints to be dynamically added. Finally, we perform an experimental study comparing our approach vs adhoc methods on several types of OPs.

This paper introduces a novel image representation capturing feature dependencies through the mining of meaningful combinations of visual features. This representation leads to a compact and discriminative encoding of images that can be used for image classification, object detection or object recognition. The method relies on (i) multiple random projections of the input space followed by local binarization of projected histograms encoded as sets of items, and (ii) the representation of images as Histograms of Pattern Sets (HoPS). The approach is validated on four publicly available datasets (Daimler Pedestrian, Oxford Flowers, KTH Texture and PASCAL VOC2007), allowing comparisons with many recent approaches. The proposed image representation reaches state-of-the-art performance on each one of these datasets.

We introduce skypattern cubes and propose an efficient bottom-up approach to compute them. Our approach relies on derivation rules collecting skypatterns of a parent node from its child nodes without any dominance test. Non-derivable skypatterns are computed on the fly thanks to Dynamic CSP. The bottom-up principle enables to provide a concise representation of the cube based on skypattern equivalence classes without any supplementary effort. Experiments show the effectiveness of our proposal.

Sequential pattern mining is one of the most studied and challenging tasks in data mining. However, the extension of well-known methods from many other classical patterns to sequences is not a trivial task. In this paper we study the notion of δ-freeness for sequences. While this notion has extensively been discussed for itemsets, this work is the first to extend it to sequences. We define an efficient algorithm devoted to the extraction of δ-free sequential patterns. Furthermore, we show the advantage of the δ-free sequences and highlight their importance when building sequence classifiers, and we show how they can be used to address the feature selection problem in statistical classifiers, as well as to build symbolic classifiers which optimizes both accuracy and earliness of predictions.

Sequential pattern mining under various constraints is a challenging data mining task. The paper provides a generic framework based on constraint programming to discover sequence patterns defined by constraints on local patterns (e.g., Gap, regular expressions) or constraints on patterns involving combination of local patterns such as relevant subgroups and top-k patterns. This framework enables the user to mine in a declarative way both kinds of patterns. The solving step is done by exploiting the machinery of Constraint Programming. For complex patterns involving combination of local patterns, we improve the mining step by using dynamic CSP. Finally, we present two case studies in biomedical information extraction and stylistic analysis in linguistics.

We propose an effective method to compute the sky pattern cubes thanks to a relaxation strategy in the pattern mining process. Our approach is based on the fact that each node of the cube can be approximated by the set of edge-sky patterns (a relaxed form of sky patterns) w.r.t. The whole set of measures M. Then we transform the problem into a skyline cube mining in M dimensions. The set of edge-sky patterns can be efficiently mined by using either a dynamic CSP method or an extended version of a static method based on the theoretical relationships between patterns and condensed representations of sky patterns. Experiments conducted on UCI datasets and on a real-life dataset (Mutagen city) show the relevance and performance of our approach.

Text-based image retrieval is a popular and simple framework consisting in using text annotations (e.g. image names, tags) to perform image retrieval, allowing to handle efficiently very large image collections. Even if the set of images retrieved using text annotations is noisy, it constitutes a reasonable initial set of images that can be considered as a bootstrap and improved further by analyzing image content. In this context, this paper introduces an approach for improving this initial set by re-ranking the so-obtained images, assuming that non-relevant images are scattered (i.e. they do not form clusters), unlike the relevant ones. More specifically, the approach consists in computing efficiently and on the fly frequent closed patterns, and in re-ranking images based on the number of patterns they contain. To do this, the paper introduces a simple but powerful new scoring function. The approach is validated on three different datasets for which state-of-the-art results are obtained.

This paper addresses the problem of detecting groups of duplicates in large-scale unstructured image datasets such as the Internet. Leveraging the recent progress in data mining, we propose an efficient approach based on the search of closed patterns. Moreover, we present a novel way to encode the bag-of-words image representation into data mining transactions. We validate our approach on a new dataset of one million Internet images obtained with random searches on Google image search. Using the proposed method, we find more than 80 thousands groups of duplicates among the one million images in less than three minutes while using only 150 Megabytes of memory. Unlike other existing approaches, our method can scale gracefully to larger datasets as it has linear time and space (memory) complexities. Furthermore, the approach does not need (to build or use) any precomputed indexing structure.

Orphanet provides an international web-based knowledge portal for rare diseases including a collection of review articles. However, reviews and literature monitoring are manual. Thus, new documentation about a rare disease is a time-consuming process and automatically discovering knowledge from a large collection of texts is a crucial issue. This context represents a strong motivation to address the problem of extracting gene-rare diseases relationships from texts. In this paper, we tackle this issue with a cross-fertilization of information extraction and data mining techniques (sequential pattern mining under constraints). Experiments show the interest of the method for the documentation of rare diseases.

Pattern discovery is at the core of numerous data mining tasks. Although many methods focus on efficiency in pattern mining, they still suffer from the problem of choosing a threshold that influences the final extraction result. The goal of our study is to make the results of pattern mining useful from a user-preference point of view. To this end, we integrate into the pattern discovery process the idea of skyline queries in order to mine skyline patterns in a threshold-free manner. Because the skyline patterns satisfy a formal property of dominations, they not only have a global interest but also have semantics that are easily understood by the user. In this work, we first establish theoretical relationships between pattern condensed representations and skyline pattern mining. We also show that it is possible to compute automatically a subset of measures involved in the user query which allows the patterns to be condensed and thus facilitates the computation of the skyline patterns. This forms the basis for a novel approach to mining skyline patterns. We illustrate the efficiency of our approach over several data sets including a use case from chemo informatics and show that small sets of dominant patterns are produced under various measures.

Constraint-based mining is an active field of research which is a necessary step to achieve interactive and successful KDD processes. The limitations of the task lies in languages being limited to describe the mined patterns and the ability to expres

In this paper, we propose a new approach based on two original principles to categorize biomedical articles. On the one hand, we combine linguistic, structural and metric descriptors to build patterns stemming from data mining techniques. On the other hand, we take into account the importance of the absence of patterns to the categorization task by using an exclusion-inclusion method. To avoid a crisp effect between the absence and the presence of a pattern, the exclusion-inclusion method uses two regret measures to quantify the interest of a weak pattern according to the other classes and among patterns from a same class. The global decision is based on the generalization of the local patterns, firstly by using patterns excluding classes, then according to the regret ratios. Experiments show the effectiveness of the approach.

The discovery of biologically interpretable knowledge from gene expression data is one of the largest contemporary genomic challenges. As large volumes of expression data are being generated, there is a great need for automated tools that provide the means to analyze them. However, the same tools can provide an overwhelming number of candidate hypotheses which can hardly be manually exploited by an expert. An additional knowledge helping to focus automatically on the most plausible candidates only can up-value the experiment significantly. Background knowledge available in literature databases, biological ontologies and other sources can be used for this purpose. In this paper we propose and verify a methodology that enables to effectively mine and represent meaningful over-expression patterns. Each pattern represents a bi-set of a gene group over-expressed in a set of biological situations. The originality of the framework consists in its constraint-based nature and an effective cross-fertilization of constraints based on expression data and background knowledge. The result is a limited set of candidate patterns that are most likely interpretable by biologists. Supplemental automatic interpretations serve to ease this process. Various constraints can generate plausible pattern sets of different characteristics.

This work proposes an original combination of linguistic and structural descriptors to represent the content of biomedical papers. The objective is to show the effectiveness of descriptors taking into account the structure of documents to characterise three kinds of biomedical texts (reviews, research and clinical papers). The description of text is made at various levels, from the global level to the local one. The contexts makes it possible to characterise the three classes. The characterisation of the textual resources is carried out quantitatively by using the discriminating capacity of techniques of data mining based on emerging patterns.

In constraint-based mining, the monotone and anti-monotone properties are exploited to reduce the search space. Even if a constraint has not such suitable properties, existing algorithms can be re-used thanks to an approximation, called relaxation. In this paper, we automatically compute monotone relaxations of primitive-based constraints. First, we show that the latter are a superclass of combinations of both kinds of monotone constraints. Second, we add two operators to detect the properties of monotonicity of such constraints. Finally, we define relaxing operators to obtain monotone relaxations of them.