Obfuscation is a software protection technique that transforms a program's binary code to conceal its behavior and to hinder analysis.Conversely, deobfuscation is an adversarial process that seeks to partially or fully remove the applied obfuscation in order to recover the original, unobfuscated code.This work introduces the first deobfuscation framework based on Reinforcement Learning (RL).It models an obfuscated function's binary code using a novel graph representation integrating both data and control-flow.The graph is then progressively simplified through a sequence of graph-edit operations, selected iteratively by a Graph Neural Network (GNN)-based agent operating within a RL pipeline.Experiments demonstrate promising results on Mixed Boolean Arithmetic (MBA) obfuscation, where multiple variants of diverse expressions can successfully be simplified into valid deobfuscated variants.* This work was partially

When building dimension reduction methods tailored for Symmetric Positive Definite (SPD) matrices, it is crucial to account for their Riemannian geometry.In this work, we propose an SPDNet-based autoencoder, that we call SPDNet-AE, that learns low-dimensional SPD representations of high-dimensional SPD matrices while preserving the geometry throughout the network.The SPDNet-AE is built using the BiMap layer of the SPDNet, but we allow it to have multiple channels.We show that our SPDNet-AE is able to learn a useful low-dimensional representation of the data for classification (without any class information).Moreover, we show that with a comparable number of parameters, a classical Euclidean autoencoder is not able to learn and maintain the SPD constraint on the input matrices.

In this paper, we introduce Edges , a novel deep architecture that aims at predicting the Graph Edit Distance (GED). Edges reformulates the quadratic assignment problem (QAP) associated to the GED problem as an edge prediction task within a GED instance graph constructed from the input graph pair. It uses a 3-Weisfeiler-Lehman expressive GNN, enabling to embed structural information at the edge level on this GED instance graph . It bypasses the need for costly matching solvers by directly predicting a soft assignment matrix through an end-to-end architecture. Extensive experiments on benchmark datasets demonstrate that the method enhances prediction accuracy through structural awareness while maintaining computational efficiency.

Since the advent of Graph Neural Networks (GNNs), many works have computed distances between graphs by embedding them in vector spaces using Message Passing GNNs (MPNNs).However, MPNNs are known for their lack of expressiveness as they are bounded by the firstorder Weisfeiler-Lehman test.In this paper, we use higher-order GNNs to tackle the metric learning problem and show on benchmark datasets how they can improve performance by using a node-level strategy and the Wasserstein distance. IntroductionA key challenge in modeling structured information with graphs lies in computing the distances between them.The Graph Edit Distance(GED)[4] is a state-ofthe-art method for this purpose; however, it suffers from NP-hard complexity.Recently, several architectures have been proposed to address this limitation [9,7,8,11,12] in a learning framework.These architectures generally consist of two main components.The first is an embedding block that uses siamese Graph Neural Networks(GNNs) to embed graphs either at the graph level or at the node level.The second component is a metric block that takes the embeddings generated by the first block as input and computes the distance between graphs, taking into account the embedding level.The rationale behind these architectures is that the embedding block learns an optimal representation to facilitate the computations in the metric block.To the best of our knowledge, existing embedding blocks in the literature rely on simple yet effective Message Passing Neural Networks(MPNNs), such as GCN [3] or GIN [2].Consequently, they suffer from the well-known limitations of MPNNs, including over-smoothing, over-squashing, and limited expressive power.This last limitation is particularly significant for metric learning, as it affects the ability to generate distinct embeddings for different graphs.Yet, GCN and GIN models have been shown to be at most equivalent to the firstorder Weisfeiler-Lehman(WL) test in the WL hierarchy [1].Recently, more expressive GNNs such as PPGN [5] and G 2 N 2 [6] have been introduced in the literature, achieving a 3-WL expressivity level.To attain this level of expressivity, these architectures naturally incorporate edge embeddings, adding valuable information to the traditional node-and graph-level representations.These recent developments raise two research questions: how can 3-WL GNNs be integrated into a metric learning framework, and do they enable improved performance?283

Circular and non-flat data distribution are prevalent across diverse domains of data science, yet their specific geometric structures often remain underutilized in machine learning frameworks. A principled approach to accounting for the underlying geometry of such data is pivotal, particularly when extending statistical models, like the pervasive Gaussian distribution. In this work, we tackle those issue by focusing on the manifold of symmetric positive definite matrices, a key focus in information geometry. We introduced a non-isotropic wrapped Gaussian by leveraging the exponential map, we derive theoretical properties of this distribution and propose a maximum likelihood framework for parameter estimation. Furthermore, we reinterpret established classifiers on SPD through a probabilistic lens and introduce new classifiers based on the wrapped Gaussian model. Experiments on synthetic and real-world datasets demonstrate the robustness and flexibility of this geometry-aware distribution, underscoring its potential to advance manifold-based data analysis. This work lays the groundwork for extending classical machine learning and statistical methods to more complex and structured data.

In mediated uncoupled learning (MU-learning), the goal is to predict an output variable $Y$ given an input variable $X$ as in ordinary supervised learning while the training dataset has no joint samples of $(X, Y)$ but only independent samples of $(X, U)$ and $(U, Y)$ each observed with a mediating variable $U$. The existing MU-learning methods can only handle the squared loss, which prohibited the use of other popular loss functions such as the cross-entropy loss. We propose a general MU-learning framework that allows for the problems with Bregman divergences, which cover a wide range of loss functions useful for various types of tasks, in a unified manner. This loss family has maximal generality among those whose minimizers characterize the conditional expectation. We prove that the proposed objective function is a tighter approximation to the oracle loss that one would minimize if ordinary supervised samples of $(X, Y)$ were available. We also propose an estimator of an interval containing the expected test loss of predictions of a trained model only using $(X, U)$- and $(U, Y)$-data. We provide a theoretical analysis on the excess risk for the proposed method and confirm its practical usefulness with regression experiments with synthetic data and low-quality image classification experiments with benchmark datasets.

Epistemic social choice aims at unveiling a hidden ground truth given votes, which are interpreted as noisy signals about it. We consider here a simple setting where votes consist of approval ballots: each voter approves a set of alternatives which they believe can possibly be the ground truth. Based on the intuitive idea that more reliable votes contain fewer alternatives, we define several noise models that are approval voting variants of the Mallows model. The likelihood-maximizing alternative is then characterized as the winner of a weighted approval rule, where the weight of a ballot decreases with its cardinality. We have conducted an experiment on three image annotation datasets; they conclude that rules based on our noise model outperform standard approval voting; the best performance is obtained by a variant of the Condorcet noise model.

Auto-Adaptive Laplacian Pyramids (ALP) is an iterative kernel-based regression model.It constructs a multi-scale representation of the train data, where the multi-scale modes are average residuals.In this work, we propose two extensions of the model.The first is a hybrid approach that combines ALP with Empirical Mode Decomposition to provide localization in the frequency domain.The second modifies ALP to fit datasets with non-uniform noise, which is achieved by computing the optimal stopping criterion in a point-dependent manner.Experimental results demonstrate these models for solar energy prediction and for forecasting epidemiology infections.

CNNs are often assumed to be capable of using contextual information about distinct objects (such as their directional relations) inside their receptive field. However, the nature and limits of this capacity has never been explored in full. We explore a specific type of relationship – directional – using a standard U-Net trained to optimize a cross-entropy loss function for segmentation. We train this network on a pretext segmentation task requiring directional relation reasoning for success and state that, with enough data and a sufficiently large receptive field, it succeeds to learn the proposed task. We further explore what the network has learned by analysing scenarios where the directional relationships are perturbed, and show that the network has learned to reason using these relationships.

Linear Sum Assignment Problem (LSAP) consists in mapping two sets of points of equal sizes according to a matrix encoding the cost of mapping each pair of points. The Linear Sum Assignment Problem with Edition (LSAPE) extends this problem by allowing the mapping of sets of different sizes and adding the possibility to reject some matchings. This problem is set up by a rectangular cost matrix whose last column and last line encode the costs of rejecting the match of an element of respectively the first and the second sets. LSAPE has been the workhorse of many fundamental graph problems such as graph edit distance, median graph computation or sub graph matching. LSAP may be solved using the Hungarian algorithm while an equivalent efficient discrete algorithm has been designed for LSAPE. However, while the Sinkhorn algorithm constitutes a continuous solver for LSAP, no such algorithm yet exists for LSAPE. This lack of solvers forbids the integration of LSAPE in Neural networks requiring continuous operations from the input to the final loss. This paper aims at providing such a solver, hence paving the way to an integration of LSAPE solvers in Neural Networks.

Epistemic voting interprets votes as noisy signals about a ground truth. We consider contexts where the truth consists of a set of objective winners, knowing a lower and upper bound on its cardinality. A prototypical problem for this setting is the aggregation of multi-label annotations with prior knowledge on the size of the ground truth. We posit noise models, for which we define rules that output an optimal set of winners. We report on experiments on multi-label annotations (which we collected).

Abstract This paper investigates the theory of robustness against adversarial attacks. We focus on randomized classifiers (i.e. classifiers that output random variables) and provide a thorough analysis of their behavior through the lens of statistical learning theory and information theory. To this aim, we introduce a new notion of robustness for randomized classifiers, enforcing local Lipschitzness using probability metrics. Equipped with this definition, we make two new contributions. The first one consists in devising a new upper bound on the adversarial generalization gap of randomized classifiers. More precisely, we devise bounds on the generalization gap and the adversarial gap i.e. the gap between the risk and the worst-case risk under attack) of randomized classifiers. The second contribution presents a yet simple but efficient noise injection method to design robust randomized classifiers. We show that our results are applicable to a wide range of machine learning models under mild hypotheses. We further corroborate our findings with experimental results using deep neural networks on standard image datasets, namely CIFAR-10 and CIFAR-100. On these tasks, we manage to design robust models that simultaneously achieve state-of-the-art accuracy (over 0.82 clean accuracy on CIFAR-10) and enjoy guaranteed robust accuracy bounds (0.45 against $$\ell _{2}$$ ℓ 2 adversaries with magnitude 0.5 on CIFAR-10).

Riemannian BCI based on EEG covariance have won many data competitions and achieved very high classification results on BCI datasets. To increase the accuracy of BCI systems, we propose an approach grounded on Riemannian geometry that extends this framework to functional connectivity measures. This paper describes the approach submitted to the Clinical BCI Challenge-WCCI2020 and that ranked 1ston the task 1 of the competition.

Dimensionality reduction of high-dimensional electroencephalography (EEG) covariance matrices is crucial for effective utilization of Riemannian geometry in Brain-Computer Interfaces (BCI). In this paper, we propose a novel similarity-based classification method that relies on dimensionality reduction of EEG covariance matrices. Conventionally, the dimension of the original high-dimensional space is reduced by projecting into one low-dimensional space, and the similarity is learned only based on the single space. In contrast, our method, MUltiple SUbspace Mdm Estimation (MUSUME), obtains multiple low-dimensional spaces that enhance class separability by solving the proposed optimization problem, then the similarity is learned in each low-dimensional space. This multiple projection approach encourages finding the space that is more useful for similarity learning. Experimental evaluation with high-dimensionality EEG datasets (128 channels) confirmed that MUSUME proved significant improvement for classification (p < 0.001) and also it showed the potential to beat the existing method relying on only one subspace representation.

Ordinary supervised learning is useful when we have paired training data of input $X$ and output $Y$. However, such paired data can be difficult to collect in practice. In this paper, we consider the task of predicting $Y$ from $X$ when we have no paired data of them, but we have two separate, independent datasets of $X$ and $Y$ each observed with some mediating variable $U$, that is, we have two datasets $S_X = \{(X_i, U_i)\}$ and $S_Y = \{(U’_j, Y’_j)\}$. A naive approach is to predict $U$ from $X$ using $S_X$ and then $Y$ from $U$ using $S_Y$, but we show that this is not statistically consistent. Moreover, predicting $U$ can be more difficult than predicting $Y$ in practice, e.g., when $U$ has higher dimensionality. To circumvent the difficulty, we propose a new method that avoids predicting $U$ but directly learns $Y = f(X)$ by training $f(X)$ with $S_{X}$ to predict $h(U)$ which is trained with $S_{Y}$ to approximate $Y$. We prove statistical consistency and error bounds of our method and experimentally confirm its practical usefulness.

Symmetric positive definite (SPD) matrices permeates numerous scientific disciplines, including machine learning, optimization, and signal processing. Equipped with a Riemannian geometry, the space of SPD matrices benefits from compelling properties and its derived Riemannian mean is now the gold standard in some applications, e.g. brain-computer interfaces (BCI). This paper addresses the problem of averaging covariance matrices with missing variables. This situation often occurs with inexpensive or unreliable sensors, or when artifact-suppression techniques remove corrupted sensors leading to rank deficient matrices, hindering the use of the Riemannian geometry in covariance-based approaches. An alternate but questionable method consists in removing the matrices with missing variables, thus reducing the training set size. We address those limitations and propose a new formulation grounded in geodesic convexity. Our approach is evaluated on generated datasets with a controlled number of missing variables and a known baseline, demonstrating the robustness of the proposed estimator. The practical interest of this approach is assessed on real BCI datasets. Our results show that the proposed average is more robust and better suited for classification than classical data imputation methods.

This paper investigates the theory of robustness against adversarial attacks. It focuses on the family of randomization techniques that consist in injecting noise in the network at inference time. These techniques have proven effective in many contexts, but lack theoretical arguments. We close this gap by presenting a theo- retical analysis of these approaches, hence explaining why they perform well in practice. More precisely, we make two new contributions. The first one relates the randomization rate to robustness to adversarial attacks. This result applies for the general family of exponential distributions, and thus extends and unifies the previous approaches. The second contribution consists in devising a new upper bound on the adversarial risk gap of randomized neural networks. We support our theoretical claims with a set of experiments.

Uplift modeling is aimed at estimating the incremental impact of an action on an individual's behavior, which is useful in various application domains such as targeted marketing (advertisement campaigns) and personalized medicine (medical treatments). Conventional methods of uplift modeling require every instance to be jointly equipped with two types of labels: the taken action and its outcome. However, obtaining two labels for each instance at the same time is difficult or expensive in many real-world problems. In this paper, we propose a novel method of uplift modeling that is applicable to a more practical setting where only one type of labels is available for each instance. We show a mean squared error bound for the proposed estimator and demonstrate its effectiveness through experiments.

The artistic style (or artistic movement) of a painting is a rich descriptor that captures both visual and historical information about the painting. Correctly identifying the artistic style of a paintings is crucial for indexing large artistic databases. In this paper, we investigate the use of deep residual neural to solve the problem of detecting the artistic style of a painting and outperform existing approaches to reach an accuracy of $62%$ on the Wikipaintings dataset (for 25 different style). To achieve this result, the network is first pre-trained on ImageNet, and deeply retrained for artistic style. We empirically evaluate that to achieve the best performance, one need to retrain about 20 layers. This suggests that the two tasks are as similar as expected, and explain the previous success of hand crafted features. We also demonstrate that the style detected on the Wikipaintings dataset are consistent with styles detected on an independent dataset and describe a number of experiments we conducted to validate this approach both qualitatively and quantitatively.

We deal with online learning of acyclic Conditional Preference networks (CP-nets) from data streams, possibly corrupted with noise. We introduce a new, efficient algorithm relying on (i) information-theoretic measures defined over the induced preference rules, which allow us to deal with corrupted data in a principled way, and on (ii) the Hoeffding bound to define an asymptotically optimal decision criterion for selecting the best conditioned variable to update the learned network. This is the first algorithm dealing with online learning of CP-nets in the presence of noise. We provide a thorough theoretical analysis of the algorithm, and demonstrate its effectiveness through an empirical evaluation on synthetic and on real datasets.

In many real-world applications data exhibits non-stationarity, i.e., its distribution changes over time. One approach to handling non-stationarity is to remove or minimize it before attempting to analyze the data. In the context of brain computer interface (BCI) data analysis this is sometimes achieved using stationary subspace analysis (SSA). The classic SSA method finds a matrix that projects the data onto a stationary subspace by optimizing a cost function based on a matrix divergence. In this work we present an alternative method for SSA based on a symmetrized version of this matrix divergence. We show that this frames the problem in terms of distances between symmetric positive definite (SPD) matrices, suggesting a geometric interpretation of the problem. Stemming from this geometric viewpoint, we introduce and analyze a method which utilizes the geometry of the SPD matrix manifold and the invariance properties of its metrics. Most notably we show that these invariances alleviate the need to whiten the input matrices, a common step in many SSA methods which often introduces error. We demonstrate the usefulness of our technique in experiments on both synthetic and real-world data.

Principal Component Analysis (PCA) is a canonical and well-studied tool for dimensionality reduction. However, when few data are available, the poor quality of the covariance estimator at its core may compromise its performance. We leverage this issue by casting the PCA into a multitask framework, and doing so, we show how to solve simultaneously several related PCA problems. Hence, we propose a novel formulation of the PCA problem relying on a novel regularization. This regularization is based on a distance between subspaces, and the whole problem is solved as an optimization problem over a Riemannian manifold. We experimentally demonstrate the usefulness of our approach as pre-processing for EEG signals.

This paper addresses the problem of principal subspace tracking in presence of a colored noise. We propose to extend the YAST algorithm to handle such a case. We also propose a Riemannian framework that could benefit to other classical trackers. Finally, as a proof of concept, our method is compared to the only oblique tracker of the literature on a toy dataset.

This paper addresses the problem of automatic wavelet feature extraction for signal classication. We propose to jointly learn wavelet-based features (including scale and translation of the wavelet as well as its shape) and a decision function by casting the problem as a Multi-Kernel Learning problem. A novel active constraints algorithm is then proposed. Our method has been tested on a toy dataset and compared to classical methods with competitive results.