Hanane Azzag

Abdellah Madane, Mohamed-djallel Dilmi, Florent Forest et al.

Conditional generation of time-dependent data is a task that has much interest, whether for data augmentation, scenario simulation, completing missing data, or other purposes. Recent works proposed a Transformer-based Time series generative adversarial network (TTS-GAN) to address the limitations of recurrent neural networks. However, this model assumes a unimodal distribution and tries to generate samples around the expectation of the real data distribution. One of its limitations is that it may generate a random multivariate time series; it may fail to generate samples in the presence of multiple sub-components within an overall distribution. One could train models to fit each sub-component separately to overcome this limitation. Our work extends the TTS-GAN by conditioning its generated output on a particular encoded context allowing the use of one model to fit a mixture distribution with multiple sub-components. Technically, it is a conditional generative adversarial network that models realistic multivariate time series under different types of conditions, such as categorical variables or multivariate time series. We evaluate our model on UniMiB Dataset, which contains acceleration data following the XYZ axes of human activities collected using Smartphones. We use qualitative evaluations and quantitative metrics such as Principal Component Analysis (PCA), and we introduce a modified version of the Frechet inception distance (FID) to measure the performance of our model and the statistical similarities between the generated and the real data distributions. We show that this transformer-based CGAN can generate realistic high-dimensional and long data sequences under different kinds of conditions.

Djamel Bouchaffra, Faycal Ykhlef, Mustapha Lebbah et al.

Cooperative multi-agent systems require robust mechanisms for credit assignment under uncertainty. Here we introduce a variational framework, termed the Game-Theoretic Free Energy Principle (GT-FEP), that models coalition formation through a Gibbs distribution over interacting agents. Within this framework, we derive a precision-dependent formulation of cooperative credit assignment and show that an agent's Shapley value exhibits a non-monotonic relationship with sensory precision beta, reflecting a trade-off between noisy inference and overconfident local estimation. Motivated by this observation, we propose Adaptive Precision Control (APC), an online adaptation algorithm that dynamically adjusts observation precision using local estimates of cooperative contribution. We evaluate APC on real-world Swiss roundabout trajectory datasets and on a multi-agent control task derived from the same trajectories. Across both settings, APC adapts to changing noise conditions online and achieves performance comparable to the best fixed precision without prior tuning. Our results connect variational inference, cooperative game theory, and adaptive multi-agent coordination, and suggest that precision adaptation can improve robust cooperation under uncertainty.

Bilal Faye, Abdoulaye Mbaye, Hanane Azzag et al.

Transformers have become the dominant architecture across a wide range of domains, largely due to the effectiveness of multi-head attention in capturing diverse representation subspaces. However, standard multi-head attention activates all heads uniformly for every input, regardless of task requirements or input complexity. In many scenarios, particularly for coarse-grained tasks such as text classification, the relevant information is often global and does not require the full diversity of attention heads. As a consequence, using a fixed number of heads can introduce unnecessary computational cost or lead to suboptimal performance when the allocation does not match the input. To address this limitation, we introduce BudgetFormer, a Transformer architecture equipped with an adaptive multi-head attention mechanism that dynamically allocates computational resources. Our approach learns, for each input, both a head budget corresponding to the number of attention heads required, and a relevance distribution that selects the most informative heads. We also propose a training strategy based on an exploration and exploitation trade-off, allowing the model to discover effective head configurations before converging to efficient usage patterns. Experiments on text classification tasks of varying complexity show that our method reduces inference cost in terms of FLOPs and memory, while also achieving performance that can surpass standard full multi-head attention. These results highlight the potential of adaptive head allocation as a principled approach to improving both efficiency and effectiveness in Transformer models.

Bilal Faye, Hanane Azzag, Mustapha Lebbah

Cross-modal alignment Learning integrates information from different modalities like text, image, audio and video to create unified models. This approach develops shared representations and learns correlations between modalities, enabling applications such as visual question answering and audiovisual content analysis. Current techniques rely on large modality-specific encoders, necessitating fine-tuning or training from scratch on vast aligned datasets (e.g., text-image, text-audio, image-audio). This approach has limitations: (i) it is very expensive due to the need for training large encoders on extensive datasets, (ii) acquiring aligned large paired datasets is challenging, and (iii) adding new modalities requires retraining the entire framework to incorporate these modalities. To address these issues, we propose OneEncoder, a lightweight framework that progressively represents and aligns four modalities (image, text, audio, video). Initially, we train a lightweight Universal Projection module (UP) to align image and text modalities. Then, we freeze the pretrained UP and progressively align future modalities to those already aligned. OneEncoder operates efficiently and cost-effectively, even in scenarios where vast aligned datasets are unavailable, due to its lightweight design. Trained on small paired datasets, it shows strong performance in tasks like classification, querying, and visual question answering, surpassing methods that rely on large datasets and specialized encoders.

Bilal Faye, Hanane Azzag, Mustapha Lebbah et al.

Deep Neural network learning for image processing faces major challenges related to changes in distribution across layers, which disrupt model convergence and performance. Activation normalization methods, such as Batch Normalization (BN), have revolutionized this field, but they rely on the simplified assumption that data distribution can be modelled by a single Gaussian distribution. To overcome these limitations, Mixture Normalization (MN) introduced an approach based on a Gaussian Mixture Model (GMM), assuming multiple components to model the data. However, this method entails substantial computational requirements associated with the use of Expectation-Maximization algorithm to estimate parameters of each Gaussian components. To address this issue, we introduce Adaptative Context Normalization (ACN), a novel supervised approach that introduces the concept of "context", which groups together a set of data with similar characteristics. Data belonging to the same context are normalized using the same parameters, enabling local representation based on contexts. For each context, the normalized parameters, as the model weights are learned during the backpropagation phase. ACN not only ensures speed, convergence, and superior performance compared to BN and MN but also presents a fresh perspective that underscores its particular efficacy in the field of image processing.

Bilal Faye, Hanane Azzag, Mustapha Lebbah et al.

Deep neural networks have become a staple in solving intricate problems, proving their mettle in a wide array of applications. However, their training process is often hampered by shifting activation distributions during backpropagation, resulting in unstable gradients. Batch Normalization (BN) addresses this issue by normalizing activations, which allows for the use of higher learning rates. Despite its benefits, BN is not without drawbacks, including its dependence on mini-batch size and the presumption of a uniform distribution of samples. To overcome this, several alternatives have been proposed, such as Layer Normalization, Group Normalization, and Mixture Normalization. These methods may still struggle to adapt to the dynamic distributions of neuron activations during the learning process. To bridge this gap, we introduce Unsupervised Adaptive Normalization (UAN), an innovative algorithm that seamlessly integrates clustering for normalization with deep neural network learning in a singular process. UAN executes clustering using the Gaussian mixture model, determining parameters for each identified cluster, by normalizing neuron activations. These parameters are concurrently updated as weights in the deep neural network, aligning with the specific requirements of the target task during backpropagation. This unified approach of clustering and normalization, underpinned by neuron activation normalization, fosters an adaptive data representation that is specifically tailored to the target task. This adaptive feature of UAN enhances gradient stability, resulting in faster learning and augmented neural network performance. UAN outperforms the classical methods by adapting to the target task and is effective in classification, and domain adaptation.

Bilal Faye, Hanane Azzag, Mustapha Lebbah

Open-vocabulary object detection (OVD) extends recognition beyond fixed taxonomies by aligning visual and textual features, as in MDETR, GLIP, or RegionCLIP. While effective, these models require updating all parameters of large vision--language backbones, leading to prohibitive training cost. Recent efficient OVD approaches, inspired by parameter-efficient fine-tuning methods such as LoRA or adapters, reduce trainable parameters but often face challenges in selecting which layers to adapt and in balancing efficiency with accuracy. We propose UniProj-Det, a lightweight modular framework for parameter-efficient OVD. UniProj-Det freezes pretrained backbones and introduces a Universal Projection module with a learnable modality token, enabling unified vision--language adaptation at minimal cost. Applied to MDETR, our framework trains only about ~2-5% of parameters while achieving competitive or superior performance on phrase grounding, referring expression comprehension, and segmentation. Comprehensive analysis of FLOPs, memory, latency, and ablations demonstrates UniProj-Det as a principled step toward scalable and efficient open-vocabulary detection.

Bilal Faye, Mohamed-Djallel Dilmi, Hanane Azzag et al.

Normalization is a pre-processing step that converts the data into a more usable representation. As part of the deep neural networks (DNNs), the batch normalization (BN) technique uses normalization to address the problem of internal covariate shift. It can be packaged as general modules, which have been extensively integrated into various DNNs, to stabilize and accelerate training, presumably leading to improved generalization. However, the effect of BN is dependent on the mini-batch size and it does not take into account any groups or clusters that may exist in the dataset when estimating population statistics. This study proposes a new normalization technique, called context normalization, for image data. This approach adjusts the scaling of features based on the characteristics of each sample, which improves the model's convergence speed and performance by adapting the data values to the context of the target task. The effectiveness of context normalization is demonstrated on various datasets, and its performance is compared to other standard normalization techniques.

Karim Radouane, Hanane Azzag, Mustapha lebbah

We propose a unified framework that integrates object detection (OD) and visual grounding (VG) for remote sensing (RS) imagery. To support conventional OD and establish an intuitive prior for VG task, we fine-tune an open-set object detector using referring expression data, framing it as a partially supervised OD task. In the first stage, we construct a graph representation of each image, comprising object queries, class embeddings, and proposal locations. Then, our task-aware architecture processes this graph to perform the VG task. The model consists of: (i) a multi-branch network that integrates spatial, visual, and categorical features to generate task-aware proposals, and (ii) an object reasoning network that assigns probabilities across proposals, followed by a soft selection mechanism for final referring object localization. Our model demonstrates superior performance on the OPT-RSVG and DIOR-RSVG datasets, achieving significant improvements over state-of-the-art methods while retaining classical OD capabilities. The code will be available in our repository: \url{https://github.com/rd20karim/MB-ORES}.

Florent Forest, Mustapha Lebbah, Hanane Azzag et al.

Self-Organizing Map algorithms have been used for almost 40 years across various application domains such as biology, geology, healthcare, industry and humanities as an interpretable tool to explore, cluster and visualize high-dimensional data sets. In every application, practitioners need to know whether they can \textit{trust} the resulting mapping, and perform model selection to tune algorithm parameters (e.g. the map size). Quantitative evaluation of self-organizing maps (SOM) is a subset of clustering validation, which is a challenging problem as such. Clustering model selection is typically achieved by using clustering validity indices. While they also apply to self-organized clustering models, they ignore the topology of the map, only answering the question: do the SOM code vectors approximate well the data distribution? Evaluating SOM models brings in the additional challenge of assessing their topology: does the mapping preserve neighborhood relationships between the map and the original data? The problem of assessing the performance of SOM models has already been tackled quite thoroughly in literature, giving birth to a family of quality indices incorporating neighborhood constraints, called \textit{topographic} indices. Commonly used examples of such metrics are the topographic error, neighborhood preservation or the topographic product. However, open-source implementations are almost impossible to find. This is the issue we try to solve in this work: after a survey of existing SOM performance metrics, we implemented them in Python and widely used numerical libraries, and provide them as an open-source library, SOMperf. This paper introduces each metric available in our module along with usage examples.

Alex Mourer, Florent Forest, Mustapha Lebbah et al.

Model selection is a major challenge in non-parametric clustering. There is no universally admitted way to evaluate clustering results for the obvious reason that no ground truth is available. The difficulty to find a universal evaluation criterion is a consequence of the ill-defined objective of clustering. In this perspective, clustering stability has emerged as a natural and model-agnostic principle: an algorithm should find stable structures in the data. If data sets are repeatedly sampled from the same underlying distribution, an algorithm should find similar partitions. However, stability alone is not well-suited to determine the number of clusters. For instance, it is unable to detect if the number of clusters is too small. We propose a new principle: a good clustering should be stable, and within each cluster, there should exist no stable partition. This principle leads to a novel clustering validation criterion based on between-cluster and within-cluster stability, overcoming limitations of previous stability-based methods. We empirically demonstrate the effectiveness of our criterion to select the number of clusters and compare it with existing methods. Code is available at https://github.com/FlorentF9/skstab.

Djamel Bouchaffra, Faycal Ykhlef, Mustapha Lebbah et al.

Collective intelligence emerges across biological, physical, and artificial systems without central coordination, yet a unifying principle governing such behaviour remains elusive. The Free Energy Principle explains how individual agents adapt through variational inference, while game theory formalises strategic interactions. Here we introduce the Game-Theoretic Free Energy Principle, a unified framework showing that multi-agent systems performing local free-energy minimisation implicitly implement a stochastic game. We prove that, under bounded rationality and local information constraints, stationary points of collective free energy correspond to approximate Nash equilibria of an induced game. Conversely, a broad class of cooperative games admits a variational representation in which equilibria arise as Gibbs distributions over coalitions, establishing a bridge between Bayesian inference and strategic interaction. To characterise higher-order effects, we introduce a free-energy formulation of the Harsanyi dividend, isolating irreducible multi-agent synergy. This yields a predictive theory of cooperation, including a falsifiable non-monotonic relationship between sensory precision and agent influence. We validate this prediction across neural, biological, and artificial multi-agent systems. These results identify a common variational principle underlying inference, thermodynamics, and game-theoretic equilibrium.

Bilal Faye, Hanane Azzag, Mustapha Lebbah et al.

Deep learning models face persistent challenges in training, particularly due to internal covariate shift and label shift. While single-mode normalization methods like Batch Normalization partially address these issues, they are constrained by batch size dependencies and limiting distributional assumptions. Multi-mode normalization techniques mitigate these limitations but struggle with computational demands when handling diverse Gaussian distributions. In this paper, we introduce a new approach to multi-mode normalization that leverages prior knowledge to improve neural network representations. Our method organizes data into predefined structures, or "contexts", prior to training and normalizes based on these contexts, with two variants: Context Normalization (CN) and Context Normalization - Extended (CN-X). When contexts are unavailable, we introduce Adaptive Context Normalization (ACN), which dynamically builds contexts in the latent space during training. Across tasks in image classification, domain adaptation, and image generation, our methods demonstrate superior convergence and performance.

Bilal Faye, Hanane Azzag, Mustapha Lebbah et al.

Low-cost cross-modal representation learning is crucial for deriving semantic representations across diverse modalities such as text, audio, images, and video. Traditional approaches typically depend on large specialized models trained from scratch, requiring extensive datasets and resulting in high resource and time costs. To overcome these challenges, we introduce a novel approach named Lightweight Cross-Modal Representation Learning (LightCRL). This method uses a single neural network titled Deep Fusion Encoder (DFE), which projects data from multiple modalities into a shared latent representation space. This reduces the overall parameter count while still delivering robust performance comparable to more complex systems.

Bilal Faye, Hanane Azzag, Mustapha Lebbah

Diffusion models have emerged as a leading framework for high-quality image generation, offering stable training and strong performance across diverse domains. However, they remain computationally intensive, particularly during the iterative denoising process. Latent-space models like Stable Diffusion alleviate some of this cost by operating in compressed representations, though at the expense of fine-grained detail. More recent approaches such as Retrieval-Augmented Diffusion Models (RDM) address efficiency by conditioning denoising on similar examples retrieved from large external memory banks. While effective, these methods introduce drawbacks: they require costly storage and retrieval infrastructure, depend on static vision-language models like CLIP for similarity, and lack adaptability during training. We propose the Prototype Diffusion Model (PDM), a method that integrates prototype learning directly into the diffusion process for efficient and adaptive visual conditioning - without external memory. Instead of retrieving reference samples, PDM constructs a dynamic set of compact visual prototypes from clean image features using contrastive learning. These prototypes guide the denoising steps by aligning noisy representations with semantically relevant visual patterns, enabling efficient generation with strong semantic grounding. Experiments show that PDM maintains high generation quality while reducing computational and storage overhead, offering a scalable alternative to retrieval-based conditioning in diffusion models.

Bilal Faye, Hanane Azzag, Mustapha Lebbah

Single-trajectory reinforcement learning (RL) methods aim to optimize policies from datasets consisting of (prompt, response, reward) triplets, where scalar rewards are directly available. This supervision format is highly practical, as it mirrors real-world human feedback, such as thumbs-up/down signals, and avoids the need for structured preference annotations. In contrast, pairwise preference-based methods like Direct Preference Optimization (DPO) rely on datasets with both preferred and dispreferred responses, which are harder to construct and less natural to collect. Among single-trajectory approaches, Direct Reward Optimization (DRO) has shown strong empirical performance due to its simplicity and stability. However, DRO requires approximating a value function, which introduces several limitations: high off-policy variance, coupling between policy and value learning, and a lack of absolute supervision on the policy itself. We introduce Reward Partitioning Optimization (RPO), a new method that resolves these limitations by removing the need to model the value function. Instead, RPO normalizes observed rewards using a partitioning approach estimated directly from data. This leads to a straightforward supervised learning objective on the policy, with no auxiliary models and no joint optimization. RPO provides direct and stable supervision on the policy, making it robust and easy to implement in practice. We validate RPO on scalar-feedback language modeling tasks using Flan-T5 encoder-decoder models. Our results demonstrate that RPO outperforms existing single-trajectory baselines such as DRO and Kahneman-Tversky Optimization (KTO). These findings confirm that RPO is a simple, effective, and theoretically grounded method for single-trajectory policy optimization.

Djamel Bouchaffra, Fayçal Ykhlef, Bilal Faye et al.

We present a novel deep graphical representation that seamlessly merges principles of game theory with laws of statistical mechanics. It performs feature extraction, dimensionality reduction, and pattern classification within a single learning framework. Our approach draws an analogy between neurons in a network and players in a game theory model. Furthermore, each neuron viewed as a classical particle (subject to statistical physics' laws) is mapped to a set of actions representing specific activation value, and neural network layers are conceptualized as games in a sequential cooperative game theory setting. The feed-forward process in deep learning is interpreted as a sequential game, where each game comprises a set of players. During training, neurons are iteratively evaluated and filtered based on their contributions to a payoff function, which is quantified using the Shapley value driven by an energy function. Each set of neurons that significantly contributes to the payoff function forms a strong coalition. These neurons are the only ones permitted to propagate the information forward to the next layers. We applied this methodology to the task of facial age estimation and gender classification. Experimental results demonstrate that our approach outperforms both multi-layer perceptron and convolutional neural network models in terms of efficiency and accuracy.

Etienne Goffinet, Anthony Coutant, Mustapha Lebbah et al.

Autonomous driving systems validation remains one of the biggest challenges car manufacturers must tackle in order to provide safe driverless cars. The high complexity stems from several factors: the multiplicity of vehicles, embedded systems, use cases, and the very high required level of reliability for the driving system to be at least as safe as a human driver. In order to circumvent these issues, large scale simulations reproducing this huge variety of physical conditions are intensively used to test driverless cars. Therefore, the validation step produces a massive amount of data, including many time-indexed ones, to be processed. In this context, building a structure in the feature space is mandatory to interpret the various scenarios. In this work, we propose a new co-clustering approach adapted to high-dimensional time series analysis, that extends the standard model-based co-clustering. The FunCLBM model extends the recently proposed Functional Latent Block Model and allows to create a dependency structure between row and column clusters. This structured partition acts as a feature selection method, that provides several clustering views of a dataset, while discriminating irrelevant features. In this workflow, times series are projected onto a common interpolated low-dimensional frequency space, which allows to optimize the projection basis. In addition, FunCLBM refines the definition of each latent block by performing block-wise dimension reduction and feature selection. We propose a SEM-Gibbs algorithm to infer this model, as well as a dedicated criterion to select the optimal nested partition. Experiments on both simulated and real-case Renault datasets shows the effectiveness of the proposed tools and the adequacy to our use case.

Andriantsiory Dina Faneva, Mustapha Lebbah, Hanane Azzag et al.

Consider a data set collected by (individuals-features) pairs in different times. It can be represented as a tensor of three dimensions (Individuals, features and times). The tensor biclustering problem computes a subset of individuals and a subset of features whose signal trajectories over time lie in a low-dimensional subspace, modeling similarity among the signal trajectories while allowing different scalings across different individuals or different features. This approach are based on spectral decomposition in order to build the desired biclusters. We evaluate the quality of the results from each algorithms with both synthetic and real data set.

Gaël Beck, Tarn Duong, Mustapha Lebbah et al.

We describe in this paper the theory and practice behind a new modal clustering method for binary data. Our approach (BinNNMS) is based on the nearest neighbor median shift. The median shift is an extension of the well-known mean shift, which was designed for continuous data, to handle binary data. We demonstrate that BinNNMS can discover accurately the location of clusters in binary data with theoretical and experimental analyses.

Gaël Beck, Tarn Duong, Mustapha Lebbah et al.

In this paper we target the class of modal clustering methods where clusters are defined in terms of the local modes of the probability density function which generates the data. The most well-known modal clustering method is the k-means clustering. Mean Shift clustering is a generalization of the k-means clustering which computes arbitrarily shaped clusters as defined as the basins of attraction to the local modes created by the density gradient ascent paths. Despite its potential, the Mean Shift approach is a computationally expensive method for unsupervised learning. Thus, we introduce two contributions aiming to provide clustering algorithms with a linear time complexity, as opposed to the quadratic time complexity for the exact Mean Shift clustering. Firstly we propose a scalable procedure to approximate the density gradient ascent. Second, our proposed scalable cluster labeling technique is presented. Both propositions are based on Locality Sensitive Hashing (LSH) to approximate nearest neighbors. These two techniques may be used for moderate sized datasets. Furthermore, we show that using our proposed approximations of the density gradient ascent as a pre-processing step in other clustering methods can also improve dedicated classification metrics. For the latter, a distributed implementation, written for the Spark/Scala ecosystem is proposed. For all these considered clustering methods, we present experimental results illustrating their labeling accuracy and their potential to solve concrete problems.