The rise of powerful generative models has sparked concerns over data authenticity. While detection methods have been extensively developed for images and text, the case of tabular data, despite its ubiquity, has been largely overlooked. Yet, detecting synthetic tabular data is especially challenging due to its heterogeneous structure and unseen formats at test time. We address the underexplored task of detecting synthetic tabular data "in the wild", i.e. when the detector is deployed on tables with variable and previously unseen schemas. We introduce a novel datum-wise transformer architecture that significantly outperforms the only previously published baseline, improving both AUC and accuracy by 7 points. By incorporating a table-adaptation component, our model gains an additional 7 accuracy points, demonstrating enhanced robustness. This work provides the first strong evidence that detecting synthetic tabular data in real-world conditions is feasible, and demonstrates substantial improvements over previous approaches. The code will be made available in the extended version.

Multivariate time series (MTS) are prevalent but inherently complex, making them challenging to analyze due to strong temporal and inter-variable correlations. This complexity often results in the use of sophisticated and difficult-to-interpret machine learning models. In real-life scenarios where critical applications of these models are common, their acceptability is crucial. Counterfactual explanations have emerged as a valuable tool for understanding machine learning systems by providing post-hoc analyzes of classification models. We introduce CFE4MTS (CounterFactual Explanation for Multivariate Time Series), a conditional, generation-based, plausible counterfactual explanation method, specifically designed for multivariate time series classification. Our approach leverages advanced time series modeling techniques to generate interpretable counterfactuals that belong to a given target class distribution. To evaluate the effectiveness of our method, we apply it to various real datasets, demonstrating the superiority of our approach over the state of the art methods.

This paper introduces ORXESTRA, a context-aware execution time prediction model based on Transformers XL, specifically designed to accurately estimate performance in embedded system applications. Unlike traditional machine learning models that often overlook contextual information, resulting in biased predictions for individual isolated basic blocks, ORXESTRA overcomes this limitation by incorporating execution context awareness. By doing so, ORXESTRA effectively accounts for the processor micro-architecture without explicitly modeling micro-architectural elements such as caches, pipelines, and branch predictors. Our evaluations demonstrate ORXESTRA's ability to provide precise timing estimations for different ARM targets (Cortex M4, M7, A53, and A72), surpassing existing machine learning-based approaches in both prediction accuracy and prediction speed.

Analytical workflows in functional magnetic resonance imaging are highly flexible with limited best practices as to how to choose a pipeline. While it has been shown that the use of different pipelines might lead to different results, there is still a lack of understanding of the factors that drive these differences and of the stability of these differences across contexts. We use community detection algorithms to explore the pipeline space and assess the stability of pipeline relationships across different contexts. We show that there are subsets of pipelines that give similar results, especially those sharing specific parameters (e.g. number of motion regressors, software packages, etc.). Those pipeline-to-pipeline patterns are stable across groups of participants but not across different tasks. By visualizing the differences between communities, we show that the pipeline space is mainly driven by the size of the activation area in the brain and the scale of statistic values in statistic maps.

Videomicroscopy is a promising tool combined with machine learning for studying the early development of in vitro fertilized bovine embryos and assessing its transferability as soon as possible. We aim to predict the embryo transferability within four days at most, taking 2D time-lapse microscopy videos as input. We formulate this problem as a supervised binary classification problem for the classes transferable and not transferable. The challenges are three-fold: 1) poorly discriminating appearance and motion, 2) class ambiguity, 3) small amount of annotated data. We propose a 3D convolutional neural network involving three pathways, which makes it multi-scale in time and able to handle appearance and motion in different ways. For training, we retain the focal loss. Our model, named SFR, compares favorably to other methods. Experiments demonstrate its effectiveness and accuracy for our challenging biological task.

International audience

We propose a new visual servoing method that controls a robot's motion in a latent space. We aim to extract the best properties of two previously proposed servoing methods: we seek to obtain the accuracy of photometric methods such as Direct Visual Servoing (DVS), as well as the behavior and convergence of pose-based visual servoing (PBVS). Photometric methods suffer from limited convergence area due to a highly non-linear cost function, while PBVS requires estimating the pose of the camera which may introduce some noise and incurs a loss of accuracy. Our approach relies on shaping (with metric learning) a latent space, in which the representations of camera poses and the embeddings of their respective images are tied together. By leveraging the multimodal aspect of this shared space, our control law minimizes the difference between latent image representations thanks to information obtained from a set of pose embeddings. Experiments in simulation and on a robot validate the strength of our approach, showing that the sought out benefits are effectively found.

We tackle, in the multiple-play bandit setting, the online ranking problem of assigning L items to K predefined positions on a web page in order to maximize the number of user clicks. We propose a generic algorithm, UniRank, that tackles state-of-the-art click models. The regret bound of this algorithm is a direct consequence of the pseudo-unimodality property of the bandit setting with respect to a graph where nodes are ordered sets of indistinguishable items. The main contribution of UniRank is its O(L/$\Delta$ logT) regret for T consecutive assignments, where $\Delta$ relates to the reward-gap between two items. This regret bound is based on the usually implicit condition that two items may not have the same attractiveness. Experiments against state-of-the-art learning algorithms specialized or not for different click models, show that our method has better regret performance than other generic algorithms on real life and synthetic datasets.

Automatic prediction of the execution time of programs for a given architecture is crucial, both for performance analysis in general and for compiler designers in particular. In this paper, we present CATREEN, a recurrent neural network able to predict the steady-state execution time of each basic block in a program. Contrarily to other models, CATREEN can take into account the execution context formed by the previously executed basic blocks which allows accounting for the processor micro-architecture without explicit modeling of micro-architectural elements (caches, pipelines, branch predictors, etc.). The evaluations conducted with synthetic programs and real ones (programs from Mibench and Polybench) show that CATREEN can provide accurate prediction for execution time with 11.4% and 16.5% error on average, respectively and that we got an improvement of 18% and 27.6% respectively when comparing our tool estimations to the state-of-the-art LSTM-based model.

International audience

Despite its robust performance under various illumination conditions, multispectral scene analysis has not been widely deployed due to two strong practical limitations: 1) thermal cameras, especially high-resolution ones are much more expensive than conventional visible cameras; 2) the most commonly adopted multispectral architectures, two-stream neural networks, nearly double the inference time of a regular mono-spectral model which makes them impractical in embedded environments. In this work, we aim to tackle these two limitations by proposing a novel knowledge distillation framework named Modality Distillation (MD). The proposed framework distils the knowledge from a high thermal resolution two-stream network with feature-level fusion to a low thermal resolution one-stream network with image-level fusion. We show on different multispectral scene analysis benchmarks that our method can effectively allow the use of low-resolution thermal sensors with more compact one-stream networks.

Data from multiple sensors provide independent and complementary information, which may improve the robustness and reliability of scene analysis applications. While there exist many large-scale labelled benchmarks acquired by a single sensor, collecting labelled multi-sensor data is more expensive and time-consuming. In this work, we explore the construction of an accurate multispectral (here, visible & thermal cameras) scene analysis system with minimal annotation efforts via an active learning strategy based on the cross-modality prediction inconsistency. Experiments on multispectral datasets and vision tasks demonstrate the effectiveness of our method. In particular, with only 10% of labelled data on KAIST multispectral pedestrian detection dataset, we obtain comparable performance as other fully supervised State-of-the-Art methods.

We tackle the online ranking problem of assigning $L$ items to $K$ positions on a web page in order to maximize the number of user clicks. We propose an original algorithm, easy to implement and with strong theoretical guarantees to tackle this problem in the Position-Based Model (PBM) setting, well suited for applications where items are displayed on a grid. Besides learning to rank, our algorithm, GRAB (for parametric Graph for unimodal RAnking Bandit), also learns the parameter of a compact graph over permutations of $K$ items among $L$. The logarithmic regret bound of this algorithm is a direct consequence of the unimodality property of the bandit setting with respect to the learned graph. Experiments against state-of-the-art learning algorithms which also tackle the PBM setting, show that our method is more efficient while giving regret performance on par with the best known algorithms on simulated and real life datasets.

In this paper we propose a deep architecture and the associated learning strategy for end-to-end direct visual servoing. The considered approach allows to sequentially predict, in se(3), the velocity of a camera mounted on the robot’s end-effector for positioning tasks. Positioning is achieved with high precision despite large initial errors in both cartesian and image spaces. Training is fully done in simulation, alleviating the burden of data collection. We demonstrate the efficiency of our method in experiments in both simulated and real-world environments. We also show that the proposed approach is able to handle multiple scenes.

We are interested in understanding the underlying generation process for long sequences of symbolic events. To do so, we propose COSSU, an algorithm to mine small and meaningful sets of sequential rules. The rules are selected using an MDL-inspired criterion that favors compactness and relies on a novel rule-based encoding scheme for sequences. Our evaluation shows that COSSU can successfully retrieve relevant sets of closed sequential rules from a long sequence. Such rules constitute an interpretable model that exhibits competitive accuracy for the tasks of next-element prediction and classification.

The success of a free-floating car-sharing service depends on a good allocation of the vehicles across the city, i.e. where and when they are needed by citizens. This requires predicting the demand across the geographical regions and across time, which is challenging due to the sparsity and variability of the data. Furthermore, the purpose of these predictions is to help computing the best possible car positions for the next day, hence the need to model both the prediction task and the optimisation task in a compatible way. As the allocation optimisation involves reasoning about the number of cars to assign to geographical regions, we propose to predict the expected utilisation of a car when added to a region. We discuss the challenges in modeling both the machine learning and the relocation problem, and we propose a integer linear programming method that solves the relocation problem while taking into account the model predictions and relocation distances. We experiment with the dataset from a citywide car sharing company and show how our method can increase the allocation strategies and hence profitability of the service.

Multispectral image pairs can provide complementary visual information, making pedestrian detection systems more robust and reliable. To benefit from both RGB and thermal IR modalities, we introduce a novel attentive multispectral feature fusion approach. Under the guidance of the inter- and intra-modality attention modules, our deep learning architecture learns to dynamically weigh and fuse the multispectral features. Experiments on two public multi-spectral object detection datasets demonstrate that the proposed approach significantly improves the detection accuracy at a low computation cost.

Online handwriting recognition is challenging but an already well-studied topic. However, recent advances in the development of convolutional neural networks (CNN) make us believe that these networks could still improve the state of the art especially in the much more challenging context of online children handwritten letters recognition. This is because, children handwriting is, at an early stage of learning, approximate and includes deformed letters. To evaluate the potential of these networks, we study the early and late fusions of different input channels that can provide a CNN with information about the handwriting dynamics in addition to the static image of the characters. The experiments on a real children handwriting dataset with 27 000 characters acquired in primary schools, show that using multiple channels with CNN, improves the accuracy performance of different CNN architectures and different fusion settings for character recognition.

Multispectral images (e.g. visible and infrared) may be particularly useful when detecting objects with the same model in different environments (e.g. day/night outdoor scenes). To effectively use the different spectra, the main technical problem resides in the information fusion process. In this paper, we propose a new halfway feature fusion method for neural networks that leverages the complementary/consistency balance existing in multispectral features by adding to the network architecture, a particular module that cyclically fuses and refines each spectral feature. We evaluate the effectiveness of our fusion method on two challenging multispectral datasets for object detection. Our results show that implementing our Cyclic Fuse-and-Refine module in any network improves the performance on both datasets compared to other state-of-the-art multispectral object detection methods.

Times series classification can be successfully tackled by jointly learning a shapelet-based representation of the series in the dataset and classifying the series according to this representation. This shapelet-based classification is both accurate and explainable since the shapelets are time series themselves and thus can be visualized and be provided as a classification explanation. In this paper, we claim that not all shapelets are good visual explanations and we propose a simple, yet also accurate, adversarily regularized EXplainable Convolutional Neural Network, XCNN, that can learn shapelets that are, by design, suited for explanations. We validate our method on the usual univariate time series benchmarks of the UCR repository.

Domain adaptation consists in learning from a source data distribution a model that will be used on a different target data distribution. The domain adaptation procedure is usually unsuccessful if the source domain is too different from the target one. In this paper, we study domain adaptation for image classification with deep learning in the context of multiple available source domains. We propose a multisource domain adaptation method that selects and weights the sources based on inter-domain distances. We provide encouraging results on both classical benchmarks and a new real world application with 21 domains.

Our research tackles the challenge of milk production resource use efficiency in dairy farms with machine learning methods. Reproduction is a key factor for dairy farm performance since cows milk production begin with the birth of a calf. Therefore, detecting estrus, the only period when the cow is susceptible to pregnancy, is crucial for farm efficiency. Our goal is to enhance estrus detection (performance, interpretability), especially on the currently undetected silent estrus (35% of total estrus), and allow farmers to rely on automatic estrus detection solutions based on affordable data (activity, temperature). In this paper, we first propose a novel approach with real-world data analysis to address both behavioral and silent estrus detection through machine learning methods. Second, we present LCE, a local cascade based algorithm that significantly outperforms a typical commercial solution for estrus detection, driven by its ability to detect silent estrus. Then, our study reveals the pivotal role of activity sensors deployment in estrus detection. Finally, we propose an approach relying on global and local (behavioral versus silent) algorithm interpretability (SHAP) to reduce the mistrust in estrus detection solutions.

Color constancy is the ability of the human visual system to perceive constant colors for a surface despite changes in the spectrum of the illumination. In computer vision, the main approach consists in estimating the illuminant color and then to remove its impact on the color of the objects. Many image processing algorithms have been proposed to tackle this problem automatically. However, most of these approaches are handcrafted and mostly rely on strong empirical assumptions, e.g., that the average reflectance in a scene is gray. State-of-the-art approaches can perform very well on some given datasets but poorly adapt on some others. In this paper, we have investigated how neural networks-based approaches can be used to deal with the color constancy problem. We have proposed a new network architecture based on existing successful hand-crafted approaches and a large number of improvements to tackle this problem by learning a suitable deep model. We show our results on most of the standard benchmarks used in the color constancy domain.

This paper shows how data mining and in particular graph mining and clustering can help to tackle difficult tracking problems such as tracking possibly multiple objects in a video with a moving camera and without any contextual information on the objects to track. Starting from different segmentations of the video frames (dynamic and non dynamic ones), we extract frequent sub graph patterns to create spatio-temporal patterns that may correspond to interesting objects to track. We then cluster the obtained spatio-temporal patterns to get longer and more robust tracks along the video. We compare our tracking method called TRAP to two state-of-the-art tracking ones and show on four synthetic and real videos that our method is effective in this difficult context.

Dynamic graph mining is the task of searching for subgraph patterns that capture the evolution of a dynamic graph. In this paper, we are interested in mining dynamic graphs in videos. A video can be regarded as a dynamic graph, whose evolution over t

Bag-of-words-based image classification approaches mostly rely on low level local shape features. However, it has been shown that combining multiple cues such as color, texture, or shape is a challenging and promising task which can improve the classification accuracy. Most of the state-of-the-art feature fusion methods usually aim to weight the cues without considering their statistical dependence in the application at hand. In this paper, we present a new logistic regression-based fusion method, called LRFF, which takes advantage of the different cues without being tied to any of them. We also design a new marginalized kernel by making use of the output of the regression model. We show that such kernels, surprisingly ignored so far by the computer vision community, are particularly well suited to achieve image classification tasks. We compare our approach with existing methods that combine color and shape on three datasets. The proposed learning-based feature fusion process clearly outperforms the state-of-the art fusion methods for image classification.

In high dimensional data, the general performance of traditional clustering algorithms decreases.This is partly because the similarity criterion used by these algorithms becomes inadequate in high dimensional space.Another reason is that some dimensions are likely to be irrelevant or contain noisy data, thus hiding a possible clustering.To overcome these problems, subspace clustering techniques, which can automatically find clusters in relevant subsets of dimensions, have been developed.However, due to the huge number of subspaces to consider, these techniques often lack efficiency.In this paper we propose to extend the framework of bottomup subspace clustering algorithms by integrating background knowledge and, in particular, instance-level constraints to speed up the enumeration of subspaces.We show how this new framework can be applied to both density and distancebased bottom-up subspace clustering techniques.Our experiments on real datasets show that instance-level constraints cannot only increase the efficiency of the clustering process but also the accuracy of the resultant clustering.

Monitoring patients in intensive care units is a critical task. Simple condition detection is generally insufficient to diagnose a patient and may generate many false alarms to the clinician operator. Deeper knowledge is needed to discriminate among alarms those that necessitate urgent therapeutic action. We propose an intelligent monitoring system that makes use of many artificial intelligence techniques: artificial neural networks for temporal abstraction, temporal reasoning, model based diagnosis, decision rule based system for adaptivity and machine learning for knowledge acquisition. To tackle the difficulty of taking context change into account, we introduce a pilot aiming at adapting the system behavior by reconfiguring or tuning the parameters of the system modules. A prototype has been implemented and is currently experimented and evaluated. Some results, showing the benefits of the approach, are given.

We present a system towards the integration of data mining into relational databases. To this end, a relational database model is proposed, based on the so called virtual mining views. We show that several types of patterns and models over the data, such as itemsets, association rules and decision trees, can be represented and queried using a unifying framework.

We present a prototype of an inductive database. Our system enables the user to query not only the data stored in the database but also generalizations (e.g. rules or trees) over these data through the use of virtual mining views. The mining views are relational tables that virtually contain the complete output of data mining algorithms executed over a given dataset. The prototype implemented into PostgreSQL currently integrates frequent itemset, association rule and decision tree mining. We illustrate the interactive and iterative capabilities of our system with a description of a complete data mining scenario.

We present DL8, an exact algorithm for finding a decision tree that optimizes a ranking function under size, depth, accuracy and leaf constraints. Because the discovery of optimal trees has high theoretical complexity, until now few efforts have been made to compute such trees for real-world datasets. An exact algorithm is of both scientific and practical interest. From a scientific point of view, it can be used as a gold standard to evaluate the performance of heuristic constraint-based decision tree learners and to gain new insight in traditional decision tree learners. From the application point of view, it can be used to discover trees that cannot be found by heuristic decision tree learners. The key idea behind our algorithm is that there is a relation between constraints on decision trees and constraints on itemsets. We show that optimal decision trees can be extracted from lattices of itemsets in linear time. We give several strategies to efficiently build these lattices. Experiments show that under the same constraints, DL8 obtains better results than C4.5, which confirms that exhaustive search does not always imply overfitting. The results also show that DL8 is a useful and interesting tool to learn decision trees under constraints.