Mélanie Ducoffe's research while affiliated with Géosciences Environnement Toulouse - Observatoire Midi-Pyrénées and other places

Publications (16)

Chapter
The use of ML technology to design safety-critical systems requires a complete understanding of the neural network’s properties. Among the relevant properties in an industrial context, the verification of partial monotony may become mandatory. This paper proposes a method to evaluate the monotony property using a Mixed Integer Linear Programming (M...
Preprint
Full-text available
A variety of methods have been proposed to try to explain how deep neural networks make their decisions. Key to those approaches is the need to sample the pixel space efficiently in order to derive importance maps. However, it has been shown that the sampling methods used to date introduce biases and other artifacts, leading to inaccurate estimates...
Article
The ability to detect anomalies in time series is considered highly valuable in numerous application domains. The sequential nature of time series objects is responsible for an additional feature complexity, ultimately requiring specialized approaches in order to solve the task. Essential characteristics of time series, situated outside the time do...
Preprint
We describe a complete method that learns a neural network which is guaranteed to overestimate a reference function on a given domain. The neural network can then be used as a surrogate for the reference function. The method involves two steps. In the first step, we construct an adaptive set of Majoring Points. In the second step, we optimize a wel...
Article
Full-text available
Deep learning applications have been thriving over the last decade in many different domains, including computer vision and natural language understanding. The drivers for the vibrant development of deep learning have been the availability of abundant data, breakthroughs of algorithms and the advancements in hardware. Despite the fact that complex...
Preprint
Full-text available
The ability to detect anomalies in time series is considered as highly valuable within plenty of application domains. The sequential nature of time series objects is responsible for an additional feature complexity, ultimately requiring specialized approaches for solving the task. Essential characteristics of time series, laying outside the time do...
Preprint
Deep learning applications have been thriving over the last decade in many different domains, including computer vision and natural language understanding. The drivers for the vibrant development of deep learning have been the availability of abundant data, breakthroughs of algorithms and the advancements in hardware. Despite the fact that complex...
Conference Paper
Full-text available
Embedding simulation models developed during the design of a platform opens a lot of potential new functionalities but requires additional certification. Usually, these models require too much computing power, take too much time to run so we need to build an approximation of these models that can be compatible with operational constraints, hardware...
Article
Full-text available
Modern vehicles are more and more connected. For instance, in the aerospace industry, newer aircraft are already equipped with data concentrators and enough wireless connectivity to transmit sensor data collected during the whole flight to the ground, usually when the airplane is at the gate. Moreover, platforms that were not designed with such cap...
Thesis
Our work is presented in three separate parts which can be read independently. Firstly we propose three active learning heuristics that scale to deep neural networks: We scale query by committee, an ensemble active learning methods. We speed up the computation time by sampling a committee of deep networks by applying dropout on the trained model. A...
Article
We propose a new active learning strategy designed for deep neural networks. The goal is to minimize the number of data annotation queried from an oracle during training. Previous active learning strategies scalable for deep networks were mostly based on uncertain sample selection. In this work, we focus on examples lying close to the decision boun...
Article
Full-text available
The Wasserstein distance received a lot of attention recently in the community of machine learning, especially for its principled way of comparing distributions. It has found numerous applications in several hard problems, such as domain adaptation, dimensionality reduction or generative models. However, its use is still limited by a heavy computat...
Article
Full-text available
Theano is a Python library that allows to define, optimize, and evaluate mathematical expressions involving multi-dimensional arrays efficiently. Since its introduction, it has been one of the most used CPU and GPU mathematical compilers - especially in the machine learning community - and has shown steady performance improvements. Theano is being...
Article
While the current trend is to increase the depth of neural networks to increase their performance, the size of their training database has to grow accordingly. We notice an emergence of tremendous databases, although providing labels to build a training set still remains a very expensive task. We tackle the problem of selecting the samples to be la...

Citations

... To address this problem, the prognostic metrics could act as a supervisory function in the HI design process instead of simply being a way of measuring HI's efficiency. The Artificial Neural Network (ANN) is a helpful approach in the case of PHM [6,7] and is a powerful mathematical method for entering this challenge. Notwithstanding, explicitly implementing the prognostic metrics (Mo, Tr, and Pr) into an ANN is difficult because the backpropagation algorithm involves the metrics' derivatives, which are difficult to calculate. ...
... However, when machine learning is applied for critical tasks such has in the transportation or the medical domain, empirical and theoretical guarantees are required. Some recent papers [7] propose theoretical guarantees for particular neural networks, but this remains an open problem for Deep Neural Networks. Empirically, weakness of deep models with respect to adversarial attack was first shown in [27], and is an active research topic. ...
... The authors (Ducoffe et al. 2019) propose the usage of the Wasserstein GAN for anomaly detection on MNIST datasets time series data. For the MNIST dataset, they trained the W-GAN with encoders and performed anomaly detection for each digit considered abnormal. ...
... In the Tf-idf BOW Regression the BOW is expanded to the whole vocabulary of a training sample instead of only the explicit terms. Furthermore, the model TDS Deconvolution is a deconvolutional neural network (Vanni et al., 2018) that estimates the importance of each word of the input for the classifier decision. In our experiments, we worked with 179k lyrics that carry gold labels provided by Deezer (17k tagged as explicit) and obtained the results shown in Fig. 3. ...
... CEAL first uses the underlying AL strategy to extract the samples from the unlabeled dataset, and then extracts additional samples by assigning pseudo-labels to those samples that are confidently predicted by the model. Enhanced adversarial approaches such as Deep-Fool Active Learning (DFAL) [24] and Adversarial Basic Interactive Method (AdvBIM) [25] have also recently become popular, which seek adversarial examples in unlabeled datasets to increase the diversity of the samples being queried. Sinha et al. [12] proposed a hybrid adversarial approach that combines variational auto-encoders with an adversarial discriminator to increase batch diversity. ...
... Il y a donc bien l'idée d'un antagonisme entre le peuple et ses dirigeants dans ce slogan. Mayaffre et al. (2017) attribuent également une variante du « dégagisme » à Macron, du fait de la distance qu'il prend avec les partis traditionnels. Ils le qualifient alors de « dégagisme poli ». ...
... To achieve that, we specify and adapt an architecture designed for and trained with two input measures and show that we can use this modified network, without retraining, to compute barycenters of more than two measures. Directly predicting Wasserstein barycenters avoids the need to compute a Wasserstein embedding [13], and our experiments suggest that this results in better Wasserstein barycenters approximations. Our implementation is publicly available. ...
... Deep learning applications in healthcare address a wide range of concerns, including individualized therapy recommendations, infection monitoring, and cancer detection [37]. Physicians now have access to vast amounts of data from many sources, including radiological imaging, genetic sequencing, and pathological imaging [38]. ...
... Active learning algorithms address the issue by finding the most informative instances for annotation [3][4][5] and have been benchmarked on natural image datasets [6][7][8][9][10][11][12][13] . Pre-training methods such as transfer learning and self-supervised learning have shown a great potential for improving the network performance on classification tasks with only a small number of labeled images [14][15][16][17] . ...