arXiv subjects

Luc Brun

Publications and source records attributed to Luc Brun.

17 recordsLinked to original sources

arXiv2024

Automatic Classification of Sleep Stages from EEG Signals Using Riemannian Metrics and Transformer Networks

Purpose: In sleep medicine, assessing the evolution of a subject's sleep often involves the costly manual scoring of electroencephalographic (EEG) signals. In recent years, a number of Deep Learning approaches have been proposed to automate this process, mainly by extracting features from said signals. However, despite some promising developments in related problems, such as Brain-Computer Interfaces, analyses of the covariances between brain regions remain underutilized in sleep stage scoring.Methods: Expanding upon our previous work, we investigate the capabilities of SPDTransNet, a Transformer-derived network designed to classify sleep stages from EEG data through timeseries of covariance matrices. Furthermore, we present a novel way of integrating learned signal-wise features into said matrices without sacrificing their Symmetric Definite Positive (SPD) nature.Results: Through comparison with other State-of-the-Art models within a methodology optimized for class-wise performance, we achieve a level of performance at or beyond various State-of-the-Art models, both in single-dataset and - particularly - multi-dataset experiments.Conclusion: In this article, we prove the capabilities of our SPDTransNet model, particularly its adaptability to multi-dataset tasks, within the context of EEG sleep stage scoring - though it could easily be adapted to any classification task involving timeseries of covariance matrices.

eess.SP↗

arXiv2024

When Quantization Affects Confidence of Large Language Models?

Recent studies introduced effective compression techniques for Large Language Models (LLMs) via post-training quantization or low-bit weight representation. Although quantized weights offer storage efficiency and allow for faster inference, existing works have indicated that quantization might compromise performance and exacerbate biases in LLMs. This study investigates the confidence and calibration of quantized models, considering factors such as language model type and scale as contributors to quantization loss. Firstly, we reveal that quantization with GPTQ to 4-bit results in a decrease in confidence regarding true labels, with varying impacts observed among different language models. Secondly, we observe fluctuations in the impact on confidence across different scales. Finally, we propose an explanation for quantization loss based on confidence levels, indicating that quantization disproportionately affects samples where the full model exhibited low confidence levels in the first place.

cs.CL↗

arXiv2023

Structure-Preserving Transformers for Sequences of SPD Matrices

In recent years, Transformer-based auto-attention mechanisms have been successfully applied to the analysis of a variety of context-reliant data types, from texts to images and beyond, including data from non-Euclidean geometries. In this paper, we present such a mechanism, designed to classify sequences of Symmetric Positive Definite matrices while preserving their Riemannian geometry throughout the analysis. We apply our method to automatic sleep staging on timeseries of EEG-derived covariance matrices from a standard dataset, obtaining high levels of stage-wise performance.

cs.LG↗

arXiv2023

Maximal Independent Sets for Pooling in Graph Neural Networks

Convolutional Neural Networks (CNNs) have enabled major advances in image classification through convolution and pooling. In particular, image pooling transforms a connected discrete lattice into a reduced lattice with the same connectivity and allows reduction functions to consider all pixels in an image. However, there is no pooling that satisfies these properties for graphs. In fact, traditional graph pooling methods suffer from at least one of the following drawbacks: Graph disconnection or overconnection, low decimation ratio, and deletion of large parts of graphs. In this paper, we present three pooling methods based on the notion of maximal independent sets that avoid these pitfalls. Our experimental results confirm the relevance of maximal independent set constraints for graph pooling.

cs.LG↗

arXiv2022

Maximal Independent Vertex Set applied to Graph Pooling

Convolutional neural networks (CNN) have enabled major advances in image classification through convolution and pooling. In particular, image pooling transforms a connected discrete grid into a reduced grid with the same connectivity and allows reduction functions to take into account all the pixels of an image. However, a pooling satisfying such properties does not exist for graphs. Indeed, some methods are based on a vertex selection step which induces an important loss of information. Other methods learn a fuzzy clustering of vertex sets which induces almost complete reduced graphs. We propose to overcome both problems using a new pooling method, named MIVSPool. This method is based on a selection of vertices called surviving vertices using a Maximal Independent Vertex Set (MIVS) and an assignment of the remaining vertices to the survivors. Consequently, our method does not discard any vertex information nor artificially increase the density of the graph. Experimental results show an increase in accuracy for graph classification on various standard datasets.

cs.LG↗

arXiv2021

A new Sinkhorn algorithm with Deletion and Insertion operations

This technical report is devoted to the continuous estimation of an epsilon-assignment. Roughly speaking, an epsilon assignment between two sets V1 and V2 may be understood as a bijective mapping between a sub part of V1 and a sub part of V2 . The remaining elements of V1 (not included in this mapping) are mapped onto an epsilon pseudo element of V2 . We say that such elements are deleted. Conversely, the remaining elements of V2 correspond to the image of the epsilon pseudo element of V1. We say that these elements are inserted. As a result our method provides a result similar to the one of the Sinkhorn algorithm with the additional ability to reject some elements which are either inserted or deleted. It thus naturally handles sets V1 and V2 of different sizes and decides mappings/insertions/deletions in a unified way. Our algorithms are iterative and differentiable and may thus be easily inserted within a backpropagation based learning framework such as artificial neural networks.

cs.LG↗

arXiv2019

Improved local search for graph edit distance

The graph edit distance (GED) measures the dissimilarity between two graphs as the minimal cost of a sequence of elementary operations transforming one graph into another. This measure is fundamental in many areas such as structural pattern recognition or classification. However, exactly computing GED is NP-hard. Among different classes of heuristic algorithms that were proposed to compute approximate solutions, local search based algorithms provide the tightest upper bounds for GED. In this paper, we present K-REFINE and RANDPOST. K-REFINE generalizes and improves an existing local search algorithm and performs particularly well on small graphs. RANDPOST is a general warm start framework that stochastically generates promising initial solutions to be used by any local search based GED algorithm. It is particularly efficient on large graphs. An extensive empirical evaluation demonstrates that both K-REFINE and RANDPOST perform excellently in practice.

cs.DS↗

arXiv2019

Upper Bounding the Graph Edit Distance Based on Rings and Machine Learning

The graph edit distance (GED) is a flexible distance measure which is widely used for inexact graph matching. Since its exact computation is NP-hard, heuristics are used in practice. A popular approach is to obtain upper bounds for GED via transformations to the linear sum assignment problem with error-correction (LSAPE). Typically, local structures and distances between them are employed for carrying out this transformation, but recently also machine learning techniques have been used. In this paper, we formally define a unifying framework LSAPE-GED for transformations from GED to LSAPE. We also introduce rings, a new kind of local structures designed for graphs where most information resides in the topology rather than in the node labels. Furthermore, we propose two new ring based heuristics RING and RING-ML, which instantiate LSAPE-GED using the traditional and the machine learning based approach for transforming GED to LSAPE, respectively. Extensive experiments show that using rings for upper bounding GED significantly improves the state of the art on datasets where most information resides in the graphs' topologies. This closes the gap between fast but rather inaccurate LSAPE based heuristics and more accurate but significantly slower GED algorithms based on local search.

cs.DS↗

arXiv2019

Generalized Median Graph via Iterative Alternate Minimizations

Computing a graph prototype may constitute a core element for clustering or classification tasks. However, its computation is an NP-Hard problem, even for simple classes of graphs. In this paper, we propose an efficient approach based on block coordinate descent to compute a generalized median graph from a set of graphs. This approach relies on a clear definition of the optimization process and handles labeling on both edges and nodes. This iterative process optimizes the edit operations to perform on a graph alternatively on nodes and edges. Several experiments on different datasets show the efficiency of our approach.

cs.CV↗

arXiv2019

Skeleton-Based Hand Gesture Recognition by Learning SPD Matrices with Neural Networks

In this paper, we propose a new hand gesture recognition method based on skeletal data by learning SPD matrices with neural networks. We model the hand skeleton as a graph and introduce a neural network for SPD matrix learning, taking as input the 3D coordinates of hand joints. The proposed network is based on two newly designed layers that transform a set of SPD matrices into a SPD matrix. For gesture recognition, we train a linear SVM classifier using features extracted from our network. Experimental results on a challenging dataset (Dynamic Hand Gesture dataset from the SHREC 2017 3D Shape Retrieval Contest) show that the proposed method outperforms state-of-the-art methods.

cs.CV↗

arXiv2019

A neural network based on SPD manifold learning for skeleton-based hand gesture recognition

This paper proposes a new neural network based on SPD manifold learning for skeleton-based hand gesture recognition. Given the stream of hand's joint positions, our approach combines two aggregation processes on respectively spatial and temporal domains. The pipeline of our network architecture consists in three main stages. The first stage is based on a convolutional layer to increase the discriminative power of learned features. The second stage relies on different architectures for spatial and temporal Gaussian aggregation of joint features. The third stage learns a final SPD matrix from skeletal data. A new type of layer is proposed for the third stage, based on a variant of stochastic gradient descent on Stiefel manifolds. The proposed network is validated on two challenging datasets and shows state-of-the-art accuracies on both datasets.

cs.CV↗

arXiv2016

Linear Sum Assignment with Edition

We consider the problem of transforming a set of elements into another by a sequence of elementary edit operations, namely substitutions, removals and insertions of elements. Each possible edit operation is penalized by a non-negative cost and the cost of a transformation is measured by summing the costs of its operations. A solution to this problem consists in defining a transformation having a minimal cost, among all possible transformations. To compute such a solution, the classical approach consists in representing removal and insertion operations by augmenting the two sets so that they get the same size. This allows to express the problem as a linear sum assignment problem (LSAP), which thus finds an optimal bijection (or permutation, perfect matching) between the two augmented sets. While the LSAP is known to be efficiently solvable in polynomial time complexity, for instance with the Hungarian algorithm, useless time and memory are spent to treat the elements which have been added to the initial sets. In this report, we show that the problem can be formalized as an extension of the LSAP which considers only one additional element in each set to represent removal and insertion operations. A solution to the problem is no longer represented as a bijection between the two augmented sets. We show that the considered problem is a binary linear program (BLP) very close to the LSAP. While it can be solved by any BLP solver, we propose an adaptation of the Hungarian algorithm which improves the time and memory complexities previously obtained by the approach based on the LSAP. The importance of the improvement increases as the size of the two sets and their absolute difference increase. Based on the analysis of the problem presented in this report, other classical algorithms can be adapted.

cs.DS↗

arXiv2015

A Quadratic Assignment Formulation of the Graph Edit Distance

Computing efficiently a robust measure of similarity or dissimilarity between graphs is a major challenge in Pattern Recognition. The Graph Edit Distance (GED) is a flexible measure of dissimilarity between graphs which arises in error-tolerant graph matching. It is defined from an optimal sequence of edit operations (edit path) transforming one graph into an other. Unfortunately, the exact computation of this measure is NP-hard. In the last decade, several approaches have been proposed to approximate the GED in polynomial time, mainly by solving linear programming problems. Among them, the bipartite GED has received much attention. It is deduced from a linear sum assignment of the nodes of the two graphs, which can be efficiently computed by Hungarian-type algorithms. However, edit operations on nodes and edges are not handled simultaneously, which limits the accuracy of the approximation. To overcome this limitation, we propose to extend the linear assignment model to a quadratic one, for directed or undirected graphs having labelized nodes and edges. This is realized through the definition of a family of edit paths induced by assignments between nodes. We formally show that the GED, restricted to the paths in this family, is equivalent to a quadratic assignment problem. Since this problem is NP-hard, we propose to compute an approximate solution by an adaptation of the Integer Projected Fixed Point method. Experiments show that the proposed approach is generally able to reach a more accurate approximation of the optimal GED than the bipartite GED, with a computational cost that is still affordable for graphs of non trivial sizes.

cs.DS↗

arXiv2009

Combinatorial pyramids and discrete geometry for energy-minimizing segmentation

This paper defines the basis of a new hierarchical framework for segmentation algorithms based on energy minimization schemes. This new framework is based on two formal tools. First, a combinatorial pyramid encode efficiently a hierarchy of partitions. Secondly, discrete geometric estimators measure precisely some important geometric parameters of the regions. These measures combined with photometrical and topological features of the partition allows to design energy terms based on discrete measures. Our segmentation framework exploits these energies to build a pyramid of image partitions with a minimization scheme. Some experiments illustrating our framework are shown and discussed.

cs.CV↗

arXiv2008

Hierarchical Bag of Paths for Kernel Based Shape Classification

Graph kernels methods are based on an implicit embedding of graphs within a vector space of large dimension. This implicit embedding allows to apply to graphs methods which where until recently solely reserved to numerical data. Within the shape classification framework, graphs are often produced by a skeletonization step which is sensitive to noise. We propose in this paper to integrate the robustness to structural noise by using a kernel based on a bag of path where each path is associated to a hierarchy encoding successive simplifications of the path. Several experiments prove the robustness and the flexibility of our approach compared to alternative shape classification methods.

cs.CV↗

arXiv2007

Hierarchy construction schemes within the Scale set framework

Segmentation algorithms based on an energy minimisation framework often depend on a scale parameter which balances a fit to data and a regularising term. Irregular pyramids are defined as a stack of graphs successively reduced. Within this framework, the scale is often defined implicitly as the height in the pyramid. However, each level of an irregular pyramid can not usually be readily associated to the global optimum of an energy or a global criterion on the base level graph. This last drawback is addressed by the scale set framework designed by Guigues. The methods designed by this author allow to build a hierarchy and to design cuts within this hierarchy which globally minimise an energy. This paper studies the influence of the construction scheme of the initial hierarchy on the resulting optimal cuts. We propose one sequential and one parallel method with two variations within both. Our sequential methods provide partitions near the global optima while parallel methods require less execution times than the sequential method of Guigues even on sequential machines.

cs.CV↗

arXiv2007

Contains and Inside relationships within combinatorial Pyramids

Irregular pyramids are made of a stack of successively reduced graphs embedded in the plane. Such pyramids are used within the segmentation framework to encode a hierarchy of partitions. The different graph models used within the irregular pyramid framework encode different types of relationships between regions. This paper compares different graph models used within the irregular pyramid framework according to a set of relationships between regions. We also define a new algorithm based on a pyramid of combinatorial maps which allows to determine if one region contains the other using only local calculus.

cs.CV↗