Nikos Deligiannis

Ada Gorgun, Fawaz Sammani, Nikos Deligiannis et al.

Diffusion models are usually evaluated by their final outputs, gradually denoising random noise into meaningful images. Yet, generation unfolds along a trajectory, and analyzing this dynamic process is crucial for understanding how controllable, reliable, and predictable these models are in terms of their success/failure modes. In this work, we ask the question: when does noise turn into a specific concept (e.g., age) and lock in the denoising trajectory? We propose PCI (Prompt-Conditioned Intervention) to study this question. PCI is a training-free and model-agnostic framework for analyzing concept dynamics through diffusion time. The central idea is the analysis of Concept Insertion Success (CIS), defined as the probability that a concept inserted at a given timestep is preserved and reflected in the final image, offering a way to characterize the temporal dynamics of concept formation. Applied to several state-of-the-art text-to-image diffusion models and a broad taxonomy of concepts, PCI reveals diverse temporal behaviors across diffusion models, in which certain phases of the trajectory are more favorable to specific concepts even within the same concept type. These findings also provide actionable insights for text-driven image editing, highlighting when interventions are most effective without requiring access to model internals or training, and yielding quantitatively stronger edits that achieve a balance of semantic accuracy and content preservation than strong baselines. Code is available at: https://github.com/adagorgun/PCI-Prompt-Controlled-Interventions

Fawaz Sammani, Tanmoy Mukherjee, Nikos Deligiannis

Natural language explanation (NLE) models aim at explaining the decision-making process of a black box system via generating natural language sentences which are human-friendly, high-level and fine-grained. Current NLE models explain the decision-making process of a vision or vision-language model (a.k.a., task model), e.g., a VQA model, via a language model (a.k.a., explanation model), e.g., GPT. Other than the additional memory resources and inference time required by the task model, the task and explanation models are completely independent, which disassociates the explanation from the reasoning process made to predict the answer. We introduce NLX-GPT, a general, compact and faithful language model that can simultaneously predict an answer and explain it. We first conduct pre-training on large scale data of image-caption pairs for general understanding of images, and then formulate the answer as a text prediction task along with the explanation. Without region proposals nor a task model, our resulting overall framework attains better evaluation scores, contains much less parameters and is 15$\times$ faster than the current SoA model. We then address the problem of evaluating the explanations which can be in many times generic, data-biased and can come in several forms. We therefore design 2 new evaluation measures: (1) explain-predict and (2) retrieval-based attack, a self-evaluation framework that requires no labels. Code is at: https://github.com/fawazsammani/nlxgpt.

Fawaz Sammani, Nikos Deligiannis

Natural Language Explanations (NLE) aim at supplementing the prediction of a model with human-friendly natural text. Existing NLE approaches involve training separate models for each downstream task. In this work, we propose Uni-NLX, a unified framework that consolidates all NLE tasks into a single and compact multi-task model using a unified training objective of text generation. Additionally, we introduce two new NLE datasets: 1) ImageNetX, a dataset of 144K samples for explaining ImageNet categories, and 2) VQA-ParaX, a dataset of 123K samples for explaining the task of Visual Question Answering (VQA). Both datasets are derived leveraging large language models (LLMs). By training on the 1M combined NLE samples, our single unified framework is capable of simultaneously performing seven NLE tasks including VQA, visual recognition and visual reasoning tasks with 7X fewer parameters, demonstrating comparable performance to the independent task-specific models in previous approaches, and in certain tasks even outperforming them. Code is at https://github.com/fawazsammani/uni-nlx

Luuk H. Boulogne, Julian Lorenz, Daniel Kienzle et al.

Challenges drive the state-of-the-art of automated medical image analysis. The quantity of public training data that they provide can limit the performance of their solutions. Public access to the training methodology for these solutions remains absent. This study implements the Type Three (T3) challenge format, which allows for training solutions on private data and guarantees reusable training methodologies. With T3, challenge organizers train a codebase provided by the participants on sequestered training data. T3 was implemented in the STOIC2021 challenge, with the goal of predicting from a computed tomography (CT) scan whether subjects had a severe COVID-19 infection, defined as intubation or death within one month. STOIC2021 consisted of a Qualification phase, where participants developed challenge solutions using 2000 publicly available CT scans, and a Final phase, where participants submitted their training methodologies with which solutions were trained on CT scans of 9724 subjects. The organizers successfully trained six of the eight Final phase submissions. The submitted codebases for training and running inference were released publicly. The winning solution obtained an area under the receiver operating characteristic curve for discerning between severe and non-severe COVID-19 of 0.815. The Final phase solutions of all finalists improved upon their Qualification phase solutions.HSUXJM-TNZF9CHSUXJM-TNZF9C

Abel Díaz Berenguer, Tanmoy Mukherjee, Matias Bossa et al.

Successful data representation is a fundamental factor in machine learning based medical imaging analysis. Deep Learning (DL) has taken an essential role in robust representation learning. However, the inability of deep models to generalize to unseen data can quickly overfit intricate patterns. Thereby, we can conveniently implement strategies to aid deep models in discovering useful priors from data to learn their intrinsic properties. Our model, which we call a dual role network (DRN), uses a dependency maximization approach based on Least Squared Mutual Information (LSMI). The LSMI leverages dependency measures to ensure representation invariance and local smoothness. While prior works have used information theory measures like mutual information, known to be computationally expensive due to a density estimation step, our LSMI formulation alleviates the issues of intractable mutual information estimation and can be used to approximate it. Experiments on CT based COVID-19 Detection and COVID-19 Severity Detection benchmarks demonstrate the effectiveness of our method.

Alexandros Stergiou, Brent De Weerdt, Nikos Deligiannis

Video anomaly detection deals with the recognition of abnormal events in videos. Apart from the visual signal, video anomaly detection has also been addressed with the use of skeleton sequences. We propose a holistic representation of skeleton trajectories to learn expected motions across segments at different times. Our approach uses multitask learning to reconstruct any continuous unobserved temporal segment of the trajectory allowing the extrapolation of past or future segments and the interpolation of in-between segments. We use an end-to-end attention-based encoder-decoder. We encode temporally occluded trajectories, jointly learn latent representations of the occluded segments, and reconstruct trajectories based on expected motions across different temporal segments. Extensive experiments on three trajectory-based video anomaly detection datasets show the advantages and effectiveness of our approach with state-of-the-art results on anomaly detection in skeleton trajectories.

Fawaz Sammani, Boris Joukovsky, Nikos Deligiannis

Contrastive learning has revolutionized the field of computer vision, learning rich representations from unlabeled data, which generalize well to diverse vision tasks. Consequently, it has become increasingly important to explain these approaches and understand their inner workings mechanisms. Given that contrastive models are trained with interdependent and interacting inputs and aim to learn invariance through data augmentation, the existing methods for explaining single-image systems (e.g., image classification models) are inadequate as they fail to account for these factors and typically assume independent inputs. Additionally, there is a lack of evaluation metrics designed to assess pairs of explanations, and no analytical studies have been conducted to investigate the effectiveness of different techniques used to explaining contrastive learning. In this work, we design visual explanation methods that contribute towards understanding similarity learning tasks from pairs of images. We further adapt existing metrics, used to evaluate visual explanations of image classification systems, to suit pairs of explanations and evaluate our proposed methods with these metrics. Finally, we present a thorough analysis of visual explainability methods for contrastive learning, establish their correlation with downstream tasks and demonstrate the potential of our approaches to investigate their merits and drawbacks.

Arttu Malkamäki, Daniel Balinhas, Letizia Iannucci et al.

This article revisits the widely studied problem of disinformation and related phenomena in online social networks (OSNs) by reframing it as a broader problem of misrepresentation. While disinformation is commonly understood as the intentional spread of false content, its meaning is applied inconsistently and often remains narrowly content-focused. This obscures other forms of manipulation, such as coordinated behavior that distorts the visibility, popularity or perceived legitimacy of actors and discourses without altering content itself. We argue that such limitations hinder a coherent and operational understanding of information campaigning in OSNs. To address this, we introduce strategic misrepresentation as a unifying concept capturing the interplay between content, actors and processes in shaping collective sensemaking. We formalize this concept through a four-dimensional framework encompassing content distortion, actor distortion, process distortion and covertness, reflecting how information campaigns unfold in practice and emphasizing observable behavioral signals. Building on this conceptualization, we conduct an integrative survey of state-of-the-art detection techniques across machine learning, network science and visual analytics. By synthesizing these approaches, we demonstrate how they jointly operationalize strategic misrepresentation in a data-driven manner. Our work provides a novel pragmatic foundation for detecting, classifying, and evaluating legitimate and illegitimate information campaigns within and across OSNs.

Lusine Abrahamyan, Nikos Deligiannis

This paper introduces an efficient patch-based computational module, coined Entropy-based Patch Encoder (EPE) module, for resource-constrained semantic segmentation. The EPE module consists of three lightweight fully-convolutional encoders, each extracting features from image patches with a different amount of entropy. Patches with high entropy are being processed by the encoder with the largest number of parameters, patches with moderate entropy are processed by the encoder with a moderate number of parameters, and patches with low entropy are processed by the smallest encoder. The intuition behind the module is the following: as patches with high entropy contain more information, they need an encoder with more parameters, unlike low entropy patches, which can be processed using a small encoder. Consequently, processing part of the patches via the smaller encoder can significantly reduce the computational cost of the module. Experiments show that EPE can boost the performance of existing real-time semantic segmentation models with a slight increase in the computational cost. Specifically, EPE increases the mIOU performance of DFANet A by 0.9% with only 1.2% increase in the number of parameters and the mIOU performance of EDANet by 1% with 10% increase of the model parameters.

Boris Joukovsky, Brent De Weerdt, Nikos Deligiannis

We propose a data-driven approach to explicitly learn the progressive encoding of a continuous source, which is successively decoded with increasing levels of quality and with the aid of correlated side information. This setup refers to the successive refinement of the Wyner-Ziv coding problem. Assuming ideal Slepian-Wolf coding, our approach employs recurrent neural networks (RNNs) to learn layered encoders and decoders for the quadratic Gaussian case. The models are trained by minimizing a variational bound on the rate-distortion function of the successively refined Wyner-Ziv coding problem. We demonstrate that RNNs can explicitly retrieve layered binning solutions akin to scalable nested quantization. Moreover, the rate-distortion performance of the scheme is on par with the corresponding monolithic Wyner-Ziv coding approach and is close to the rate-distortion bound.

Alexandros Stergiou, Nikos Deligiannis

The success of deep learning models has led to their adaptation and adoption by prominent video understanding methods. The majority of these approaches encode features in a joint space-time modality for which the inner workings and learned representations are difficult to visually interpret. We propose LEArned Preconscious Synthesis (LEAPS), an architecture-independent method for synthesizing videos from the internal spatiotemporal representations of models. Using a stimulus video and a target class, we prime a fixed space-time model and iteratively optimize a video initialized with random noise. Additional regularizers are used to improve the feature diversity of the synthesized videos alongside the cross-frame temporal coherence of motions. We quantitatively and qualitatively evaluate the applicability of LEAPS by inverting a range of spatiotemporal convolutional and attention-based architectures trained on Kinetics-400, which to the best of our knowledge has not been previously accomplished.

Tzoulio Chamiti, Leandro Di Bella, Adrian Munteanu et al.

Tracking multiple objects based on textual queries is a challenging task that requires linking language understanding with object association across frames. Previous works typically train the whole process end-to-end or integrate an additional referring text module into a multi-object tracker, but they both require supervised training and potentially struggle with generalization to open-set queries. In this work, we introduce ReferGPT, a novel zero-shot referring multi-object tracking framework. We provide a multi-modal large language model (MLLM) with spatial knowledge enabling it to generate 3D-aware captions. This enhances its descriptive capabilities and supports a more flexible referring vocabulary without training. We also propose a robust query-matching strategy, leveraging CLIP-based semantic encoding and fuzzy matching to associate MLLM generated captions with user queries. Extensive experiments on Refer-KITTI, Refer-KITTIv2 and Refer-KITTI+ demonstrate that ReferGPT achieves competitive performance against trained methods, showcasing its robustness and zero-shot capabilities in autonomous driving. The codes are available on https://github.com/Tzoulio/ReferGPT

Fawaz Sammani, Tzoulio Chamiti, Paul Gavrikov et al.

Joint Vision-Language Embedding models such as CLIP typically fail at understanding negation in text queries - for example, failing to distinguish "no" in the query: "a plain blue shirt with no logos". Prior work has largely addressed this limitation through data-centric approaches, fine-tuning CLIP on large-scale synthetic negation datasets. However, these efforts are commonly evaluated using retrieval-based metrics that cannot reliably reflect whether negation is actually understood. In this paper, we identify two key limitations of such evaluation metrics and investigate an alternative evaluation framework based on Multimodal LLMs-as-a-judge, which typically excel at understanding simple yes/no questions about image content, providing a fair evaluation of negation understanding in CLIP models. We then ask whether there already exists a direction in the CLIP embedding space associated with negation. We find evidence that such a direction exists, and show that it can be manipulated through test-time intervention via representation engineering to steer CLIP toward negation-aware behavior without any fine-tuning. Finally, we test negation understanding on non-common image-text samples to evaluate generalization under distribution shifts.

Xianlu Li, Nicolas Nadisic, Shaoguang Huang et al.

Subspace clustering has become widely adopted for the unsupervised analysis of hyperspectral images (HSIs). Recent model-aware deep subspace clustering methods often use a two-stage framework, involving the calculation of a self-representation matrix with complexity of O(n^2), followed by spectral clustering. However, these methods are computationally intensive, generally incorporating solely either local or non-local spatial structure constraints, and their structural constraints fall short of effectively supervising the entire clustering process. We propose a scalable, context-preserving deep clustering method based on basis representation, which jointly captures local and non-local structures for efficient HSI clustering. To preserve local structure (i.e., spatial continuity within subspaces), we introduce a spatial smoothness constraint that aligns clustering predictions with their spatially filtered versions. For non-local structure (i.e., spectral continuity), we employ a mini-cluster-based scheme that refines predictions at the group level, encouraging spectrally similar pixels to belong to the same subspace. Notably, these two constraints are jointly optimized to reinforce each other. Specifically, our model is designed as an one-stage approach in which the structural constraints are applied to the entire clustering process. The time and space complexity of our method is O(n), making it applicable to large-scale HSI data. Experiments on real-world datasets show that our method outperforms state-of-the-art techniques. Our code is available at: https://github.com/lxlscut/SCDSC

Ngoc Tan Le, Tzoulio Chamiti, Eirini Papagiannopoulou et al.

Automated analysis of building facades from street-level imagery has great potential for urban analytics, energy assessment, and emergency planning. However, it requires reasoning over spatially arranged elements rather than solely isolated detections. In this work, we model each facade as a graph over window/door detections with a vertical prior on edges. Additionally, we introduce GATA2Floor, a multi-head Graph Attention v2 (GATv2) based model that predicts the global floor count of a building and, via learnable cross-attention queries, softly assigns elements to latent floor slots, yielding interpretable outputs and robustness to irregular designs. To mitigate the lack of labeled datasets, we demonstrate that the proposed graph-based reasoning can be applied without annotations by leveraging a lightweight label-free proposal mechanism based on self-supervised features and vision-language scoring. Our approach demonstrates the value of graph-attention-based relational reasoning for facade understanding.

Pooya Ashtari, Pourya Behmandpoor, Nikos Deligiannis et al.

Implicit neural representations (INRs) have emerged as powerful tools for encoding signals, yet dominant MLP-based designs often suffer from slow convergence, overfitting to noise, and poor extrapolation. We introduce FUTON (Fourier Tensor Network), which models signals as generalized Fourier series whose coefficients are parameterized by a low-rank tensor decomposition. FUTON implicitly expresses signals as weighted combinations of orthonormal, separable basis functions, combining complementary inductive biases: Fourier bases capture smoothness and periodicity, while the low-rank parameterization enforces low-dimensional spectral structure. We provide theoretical guarantees through a universal approximation theorem and derive an inference algorithm with complexity linear in the spectral resolution and the input dimension. On image and volume representation, FUTON consistently outperforms state-of-the-art MLP-based INRs while training 2--5$\times$ faster. On inverse problems such as image denoising and super-resolution, FUTON generalizes better and converges faster.

Abel Díaz Berenguer, Hichem Sahli, Boris Joukovsky et al.

Our motivating application is a real-world problem: COVID-19 classification from CT imaging, for which we present an explainable Deep Learning approach based on a semi-supervised classification pipeline that employs variational autoencoders to extract efficient feature embedding. We have optimized the architecture of two different networks for CT images: (i) a novel conditional variational autoencoder (CVAE) with a specific architecture that integrates the class labels inside the encoder layers and uses side information with shared attention layers for the encoder, which make the most of the contextual clues for representation learning, and (ii) a downstream convolutional neural network for supervised classification using the encoder structure of the CVAE. With the explainable classification results, the proposed diagnosis system is very effective for COVID-19 classification. Based on the promising results obtained qualitatively and quantitatively, we envisage a wide deployment of our developed technique in large-scale clinical studies.Code is available at https://git.etrovub.be/AVSP/ct-based-covid-19-diagnostic-tool.git.

Fawaz Sammani, Nikos Deligiannis

Contrastive Language-Image Pretraining (CLIP) performs zero-shot image classification by mapping images and textual class representation into a shared embedding space, then retrieving the class closest to the image. This work provides a new approach for interpreting CLIP models for image classification from the lens of mutual knowledge between the two modalities. Specifically, we ask: what concepts do both vision and language CLIP encoders learn in common that influence the joint embedding space, causing points to be closer or further apart? We answer this question via an approach of textual concept-based explanations, showing their effectiveness, and perform an analysis encompassing a pool of 13 CLIP models varying in architecture, size and pretraining datasets. We explore those different aspects in relation to mutual knowledge, and analyze zero-shot predictions. Our approach demonstrates an effective and human-friendly way of understanding zero-shot classification decisions with CLIP.

Kasper Cools, Clara Maathuis, Alexander M. van Oers et al.

The increasing reliance on machine learning systems has made their security a critical concern. Evasion attacks enable adversaries to manipulate the decision-making processes of AI systems, potentially causing security breaches or misclassification of targets. Vision Transformers (ViTs) have gained significant traction in modern machine learning due to increased 1) performance compared to Convolutional Neural Networks (CNNs) and 2) robustness against adversarial perturbations. However, ViTs remain vulnerable to evasion attacks, particularly to adversarial patches, unique patterns designed to manipulate AI classification systems. These vulnerabilities are investigated by designing realistic adversarial patches to cause misclassification in person vs. non-person classification tasks using the Creases Transformation (CT) technique, which adds subtle geometric distortions similar to those occurring naturally when wearing clothing. This study investigates the transferability of adversarial attack techniques used in CNNs when applied to ViT classification models. Experimental evaluation across four fine-tuned ViT models on a binary person classification task reveals significant vulnerability variations: attack success rates ranged from 40.04% (google/vit-base-patch16-224-in21k) to 99.97% (facebook/dino-vitb16), with google/vit-base-patch16-224 achieving 66.40% and facebook/dinov3-vitb16 reaching 65.17%. These results confirm the cross-architectural transferability of adversarial patches from CNNs to ViTs, with pre-training dataset scale and methodology strongly influencing model resilience to adversarial attacks.

Fawaz Sammani, Jonas Fischer, Nikos Deligiannis

Visual classifiers provide high-dimensional feature representations that are challenging to interpret and analyze. Text, in contrast, provides a more expressive and human-friendly interpretable medium for understanding and analyzing model behavior. We propose a simple, yet powerful method for reformulating any pretrained visual classifier so that it can be queried with free-form text without compromising its original performance. Our approach is label-free, data and compute-efficient, and is trained to preserve the underlying classifiers distribution and decision-making processes. Our method unlocks several zero-shot text interpretability applications for any visual classifier. We apply our method on 40 visual classifiers and demonstrate two primary applications: 1) building both label-free and zero-shot concept bottleneck models and therefore converting any visual classifier to be inherently-interpretable and 2) zero-shot decoding of visual features into natural language sentences. In both tasks we establish new state-of-the-art results, outperforming existing works and surpassing CLIP-based baselines with ImageNet-only trained classifiers, while using up to 400x fewer images and 400,000x less text during training.

Lusine Abrahamyan, Anh Minh Truong, Wilfried Philips et al.

Recent success in the field of single image super-resolution (SISR) is achieved by optimizing deep convolutional neural networks (CNNs) in the image space with the L1 or L2 loss. However, when trained with these loss functions, models usually fail to recover sharp edges present in the high-resolution (HR) images for the reason that the model tends to give a statistical average of potential HR solutions. During our research, we observe that gradient maps of images generated by the models trained with the L1 or L2 loss have significantly lower variance than the gradient maps of the original high-resolution images. In this work, we propose to alleviate the above issue by introducing a structure-enhancing loss function, coined Gradient Variance (GV) loss, and generate textures with perceptual-pleasant details. Specifically, during the training of the model, we extract patches from the gradient maps of the target and generated output, calculate the variance of each patch and form variance maps for these two images. Further, we minimize the distance between the computed variance maps to enforce the model to produce high variance gradient maps that will lead to the generation of high-resolution images with sharper edges. Experimental results show that the GV loss can significantly improve both Structure Similarity (SSIM) and peak signal-to-noise ratio (PSNR) performance of existing image super-resolution (SR) deep learning models.

Xiangyu Yang, Giannis Bekoulis, Nikos Deligiannis

In this paper, we introduce the new problem of extracting fine-grained traffic information from Twitter streams by also making publicly available the two (constructed) traffic-related datasets from Belgium and the Brussels capital region. In particular, we experiment with several models to identify (i) whether a tweet is traffic-related or not, and (ii) in the case that the tweet is traffic-related to identify more fine-grained information regarding the event (e.g., the type of the event, where the event happened). To do so, we frame (i) the problem of identifying whether a tweet is a traffic-related event or not as a text classification subtask, and (ii) the problem of identifying more fine-grained traffic-related information as a slot filling subtask, where fine-grained information (e.g., where an event has happened) is represented as a slot/entity of a particular type. We propose the use of several methods that process the two subtasks either separately or in a joint setting, and we evaluate the effectiveness of the proposed methods for solving the traffic event detection problem. Experimental results indicate that the proposed architectures achieve high performance scores (i.e., more than 95% in terms of F$_{1}$ score) on the constructed datasets for both of the subtasks (i.e., text classification and slot filling) even in a transfer learning scenario. In addition, by incorporating tweet-level information in each of the tokens comprising the tweet (for the BERT-based model) can lead to a performance improvement for the joint setting.

Lusine Abrahamyan, Valentin Ziatchin, Yiming Chen et al.

Compact convolutional neural networks (CNNs) have witnessed exceptional improvements in performance in recent years. However, they still fail to provide the same predictive power as CNNs with a large number of parameters. The diverse and even abundant features captured by the layers is an important characteristic of these successful CNNs. However, differences in this characteristic between large CNNs and their compact counterparts have rarely been investigated. In compact CNNs, due to the limited number of parameters, abundant features are unlikely to be obtained, and feature diversity becomes an essential characteristic. Diverse features present in the activation maps derived from a data point during model inference may indicate the presence of a set of unique descriptors necessary to distinguish between objects of different classes. In contrast, data points with low feature diversity may not provide a sufficient amount of unique descriptors to make a valid prediction; we refer to them as random predictions. Random predictions can negatively impact the optimization process and harm the final performance. This paper proposes addressing the problem raised by random predictions by reshaping the standard cross-entropy to make it biased toward data points with a limited number of unique descriptive features. Our novel Bias Loss focuses the training on a set of valuable data points and prevents the vast number of samples with poor learning features from misleading the optimization process. Furthermore, to show the importance of diversity, we present a family of SkipNet models whose architectures are brought to boost the number of unique descriptors in the last layers. Our Skipnet-M can achieve 1% higher classification accuracy than MobileNetV3 Large.

Lusine Abrahamyan, Yiming Chen, Giannis Bekoulis et al.

Training deep neural networks on large datasets containing high-dimensional data requires a large amount of computation. A solution to this problem is data-parallel distributed training, where a model is replicated into several computational nodes that have access to different chunks of the data. This approach, however, entails high communication rates and latency because of the computed gradients that need to be shared among nodes at every iteration. The problem becomes more pronounced in the case that there is wireless communication between the nodes (i.e. due to the limited network bandwidth). To address this problem, various compression methods have been proposed including sparsification, quantization, and entropy encoding of the gradients. Existing methods leverage the intra-node information redundancy, that is, they compress gradients at each node independently. In contrast, we advocate that the gradients across the nodes are correlated and propose methods to leverage this inter-node redundancy to improve compression efficiency. Depending on the node communication protocol (parameter server or ring-allreduce), we propose two instances of the LGC approach that we coin Learned Gradient Compression (LGC). Our methods exploit an autoencoder (i.e. trained during the first stages of the distributed training) to capture the common information that exists in the gradients of the distributed nodes. We have tested our LGC methods on the image classification and semantic segmentation tasks using different convolutional neural networks (ResNet50, ResNet101, PSPNet) and multiple datasets (ImageNet, Cifar10, CamVid). The ResNet101 model trained for image classification on Cifar10 achieved an accuracy of 93.57%, which is lower than the baseline distributed training with uncompressed gradients only by 0.18%.

Esther Rodrigo Bonet, Duc Minh Nguyen, Nikos Deligiannis

Temporal collaborative filtering (TCF) methods aim at modelling non-static aspects behind recommender systems, such as the dynamics in users' preferences and social trends around items. State-of-the-art TCF methods employ recurrent neural networks (RNNs) to model such aspects. These methods deploy matrix-factorization-based (MF-based) approaches to learn the user and item representations. Recently, graph-neural-network-based (GNN-based) approaches have shown improved performance in providing accurate recommendations over traditional MF-based approaches in non-temporal CF settings. Motivated by this, we propose a novel TCF method that leverages GNNs to learn user and item representations, and RNNs to model their temporal dynamics. A challenge with this method lies in the increased data sparsity, which negatively impacts obtaining meaningful quality representations with GNNs. To overcome this challenge, we train a GNN model at each time step using a set of observed interactions accumulated time-wise. Comprehensive experiments on real-world data show the improved performance obtained by our method over several state-of-the-art temporal and non-temporal CF models.

Giannis Bekoulis, Christina Papagiannopoulou, Nikos Deligiannis

We study the fact checking problem, which aims to identify the veracity of a given claim. Specifically, we focus on the task of Fact Extraction and VERification (FEVER) and its accompanied dataset. The task consists of the subtasks of retrieving the relevant documents (and sentences) from Wikipedia and validating whether the information in the documents supports or refutes a given claim. This task is essential and can be the building block of applications such as fake news detection and medical claim verification. In this paper, we aim at a better understanding of the challenges of the task by presenting the literature in a structured and comprehensive way. We describe the proposed methods by analyzing the technical perspectives of the different approaches and discussing the performance results on the FEVER dataset, which is the most well-studied and formally structured dataset on the fact extraction and verification task. We also conduct the largest experimental study to date on identifying beneficial loss functions for the sentence retrieval component. Our analysis indicates that sampling negative sentences is important for improving the performance and decreasing the computational complexity. Finally, we describe open issues and future challenges, and we motivate future research in the task.

Huynh Van Luong, Boris Joukovsky, Yonina C. Eldar et al.

Deep unfolded neural networks are designed by unrolling the iterations of optimization algorithms. They can be shown to achieve faster convergence and higher accuracy than their optimization counterparts. This paper proposes a new deep-unfolding-based network design for the problem of Robust Principal Component Analysis (RPCA) with application to video foreground-background separation. Unlike existing designs, our approach focuses on modeling the temporal correlation between the sparse representations of consecutive video frames. To this end, we perform the unfolding of an iterative algorithm for solving reweighted $\ell_1$-$\ell_1$ minimization; this unfolding leads to a different proximal operator (a.k.a. different activation function) adaptively learned per neuron. Experimentation using the moving MNIST dataset shows that the proposed network outperforms a recently proposed state-of-the-art RPCA network in the task of video foreground-background separation.

Iman Marivani, Evaggelia Tsiligianni, Bruno Cornelis et al.

Multimodal image super-resolution (SR) is the reconstruction of a high resolution image given a low-resolution observation with the aid of another image modality. While existing deep multimodal models do not incorporate domain knowledge about image SR, we present a multimodal deep network design that integrates coupled sparse priors and allows the effective fusion of information from another modality into the reconstruction process. Our method is inspired by a novel iterative algorithm for coupled convolutional sparse coding, resulting in an interpretable network by design. We apply our model to the super-resolution of near-infrared image guided by RGB images. Experimental results show that our model outperforms state-of-the-art methods.

Tien Huu Do, Duc Minh Nguyen, Giannis Bekoulis et al.

Graph convolutional neural networks (GCNNs) have received much attention recently, owing to their capability in handling graph-structured data. Among the existing GCNNs, many methods can be viewed as instances of a neural message passing motif; features of nodes are passed around their neighbors, aggregated and transformed to produce better nodes' representations. Nevertheless, these methods seldom use node transition probabilities, a measure that has been found useful in exploring graphs. Furthermore, when the transition probabilities are used, their transition direction is often improperly considered in the feature aggregation step, resulting in an inefficient weighting scheme. In addition, although a great number of GCNN models with increasing level of complexity have been introduced, the GCNNs often suffer from over-fitting when being trained on small graphs. Another issue of the GCNNs is over-smoothing, which tends to make nodes' representations indistinguishable. This work presents a new method to improve the message passing process based on node transition probabilities by properly considering the transition direction, leading to a better weighting scheme in nodes' features aggregation compared to the existing counterpart. Moreover, we propose a novel regularization method termed DropNode to address the over-fitting and over-smoothing issues simultaneously. DropNode randomly discards part of a graph, thus it creates multiple deformed versions of the graph, leading to data augmentation regularization effect. Additionally, DropNode lessens the connectivity of the graph, mitigating the effect of over-smoothing in deep GCNNs. Extensive experiments on eight benchmark datasets for node and graph classification tasks demonstrate the effectiveness of the proposed methods in comparison with the state of the art.

Huynh Van Luong, Boris Joukovsky, Nikos Deligiannis

Deep unfolding methods---for example, the learned iterative shrinkage thresholding algorithm (LISTA)---design deep neural networks as learned variations of optimization methods. These networks have been shown to achieve faster convergence and higher accuracy than the original optimization methods. In this line of research, this paper develops a novel deep recurrent neural network (coined reweighted-RNN) by the unfolding of a reweighted $\ell_1$-$\ell_1$ minimization algorithm and applies it to the task of sequential signal reconstruction. To the best of our knowledge, this is the first deep unfolding method that explores reweighted minimization. Due to the underlying reweighted minimization model, our RNN has a different soft-thresholding function (alias, different activation functions) for each hidden unit in each layer. Furthermore, it has higher network expressivity than existing deep unfolding RNN models due to the over-parameterizing weights. Importantly, we establish theoretical generalization error bounds for the proposed reweighted-RNN model by means of Rademacher complexity. The bounds reveal that the parameterization of the proposed reweighted-RNN ensures good generalization. We apply the proposed reweighted-RNN to the problem of video frame reconstruction from low-dimensional measurements, that is, sequential frame reconstruction. The experimental results on the moving MNIST dataset demonstrate that the proposed deep reweighted-RNN significantly outperforms existing RNN models.

Iman Marivani, Evaggelia Tsiligianni, Bruno Cornelis et al.

The reconstruction of a high resolution image given a low resolution observation is an ill-posed inverse problem in imaging. Deep learning methods rely on training data to learn an end-to-end mapping from a low-resolution input to a high-resolution output. Unlike existing deep multimodal models that do not incorporate domain knowledge about the problem, we propose a multimodal deep learning design that incorporates sparse priors and allows the effective integration of information from another image modality into the network architecture. Our solution relies on a novel deep unfolding operator, performing steps similar to an iterative algorithm for convolutional sparse coding with side information; therefore, the proposed neural network is interpretable by design. The deep unfolding architecture is used as a core component of a multimodal framework for guided image super-resolution. An alternative multimodal design is investigated by employing residual learning to improve the training efficiency. The presented multimodal approach is applied to super-resolution of near-infrared and multi-spectral images as well as depth upsampling using RGB images as side information. Experimental results show that our model outperforms state-of-the-art methods.

Peng Xiao, Bin Liao, Nikos Deligiannis

We present DeepFPC, a novel deep neural network designed by unfolding the iterations of the fixed-point continuation algorithm with one-sided l1-norm (FPC-l1), which has been proposed for solving the 1-bit compressed sensing problem. The network architecture resembles that of deep residual learning and incorporates prior knowledge about the signal structure (i.e., sparsity), thereby offering interpretability by design. Once DeepFPC is properly trained, a sparse signal can be recovered fast and accurately from quantized measurements. The proposed model is evaluated in the task of direction-of-arrival (DOA) estimation and is shown to outperform state-of-the-art algorithms, namely, the iterative FPC-l1 algorithm and the 1-bit MUSIC method.

Iman Marivani, Evaggelia Tsiligianni, Bruno Cornelis et al.

Deep learning methods have been successfully applied to various computer vision tasks. However, existing neural network architectures do not per se incorporate domain knowledge about the addressed problem, thus, understanding what the model has learned is an open research topic. In this paper, we rely on the unfolding of an iterative algorithm for sparse approximation with side information, and design a deep learning architecture for multimodal image super-resolution that incorporates sparse priors and effectively utilizes information from another image modality. We develop two deep models performing reconstruction of a high-resolution image of a target image modality from its low-resolution variant with the aid of a high-resolution image from a second modality. We apply the proposed models to super-resolve near-infrared images using as side information high-resolution RGB\ images. Experimental results demonstrate the superior performance of the proposed models against state-of-the-art methods including unimodal and multimodal approaches.

Evaggelia Tsiligianni, Nikos Deligiannis

In linear inverse problems, the goal is to recover a target signal from undersampled, incomplete or noisy linear measurements. Typically, the recovery relies on complex numerical optimization methods; recent approaches perform an unfolding of a numerical algorithm into a neural network form, resulting in a substantial reduction of the computational complexity. In this paper, we consider the recovery of a target signal with the aid of a correlated signal, the so-called side information (SI), and propose a deep unfolding model that incorporates SI. The proposed model is used to learn coupled representations of correlated signals from different modalities, enabling the recovery of multimodal data at a low computational cost. As such, our work introduces the first deep unfolding method with SI, which actually comes from a different modality. We apply our model to reconstruct near-infrared images from undersampled measurements given RGB images as SI. Experimental results demonstrate the superior performance of the proposed framework against single-modal deep learning methods that do not use SI, multimodal deep learning designs, and optimization algorithms.

Tien Huu Do, Xiao Luo, Duc Minh Nguyen et al.

Rumours have existed for a long time and have been known for serious consequences. The rapid growth of social media platforms has multiplied the negative impact of rumours; it thus becomes important to early detect them. Many methods have been introduced to detect rumours using the content or the social context of news. However, most existing methods ignore or do not explore effectively the propagation pattern of news in social media, including the sequence of interactions of social media users with news across time. In this work, we propose a novel method for rumour detection based on deep learning. Our method leverages the propagation process of the news by learning the users' representation and the temporal interrelation of users' responses. Experiments conducted on Twitter and Weibo datasets demonstrate the state-of-the-art performance of the proposed method.

Hung Duy Le, Huynh Van Luong, Nikos Deligiannis

We propose a new deep recurrent neural network (RNN) architecture for sequential signal reconstruction. Our network is designed by unfolding the iterations of the proximal gradient method that solves the l1-l1 minimization problem. As such, our network leverages by design that signals have a sparse representation and that the difference between consecutive signal representations is also sparse. We evaluate the proposed model in the task of reconstructing video frames from compressive measurements and show that it outperforms several state-of-the-art RNN models.

Duc Minh Nguyen, Robert Calderbank, Nikos Deligiannis

The problem of completing high-dimensional matrices from a limited set of observations arises in many big data applications, especially, recommender systems. Existing matrix completion models generally follow either a memory- or a model-based approach, whereas, geometric matrix completion models combine the best from both approaches. Existing deep-learning-based geometric models yield good performance, but, in order to operate, they require a fixed structure graph capturing the relationships among the users and items. This graph is typically constructed by evaluating a pre-defined similarity metric on the available observations or by using side information, e.g., user profiles. In contrast, Markov-random-fields-based models do not require a fixed structure graph but rely on handcrafted features to make predictions. When no side information is available and the number of available observations becomes very low, existing solutions are pushed to their limits. In this paper, we propose a geometric matrix completion approach that addresses these challenges. We consider matrix completion as a structured prediction problem in a conditional random field (CRF), which is characterized by a maximum a posterior (MAP) inference, and we propose a deep model that predicts the missing entries by solving the MAP inference problem. The proposed model simultaneously learns the similarities among matrix entries, computes the CRF potentials, and solves the inference problem. Its training is performed in an end-to-end manner, with a method to supervise the learning of entry similarities. Comprehensive experiments demonstrate the superior performance of the proposed model compared to various state-of-the-art models on popular benchmark datasets and underline its superior capacity to deal with highly incomplete matrices.

Duc Minh Nguyen, Evaggelia Tsiligianni, Nikos Deligiannis

Matrix completion is one of the key problems in signal processing and machine learning. In recent years, deep-learning-based models have achieved state-of-the-art results in matrix completion. Nevertheless, they suffer from two drawbacks: (i) they can not be extended easily to rows or columns unseen during training; and (ii) their results are often degraded in case discrete predictions are required. This paper addresses these two drawbacks by presenting a deep matrix factorization model and a generic method to allow joint training of the factorization model and the discretization operator. Experiments on a real movie rating dataset show the efficacy of the proposed models.

Tien Huu Do, Duc Minh Nguyen, Evaggelia Tsiligianni et al.

Inferring air quality from a limited number of observations is an essential task for monitoring and controlling air pollution. Existing inference methods typically use low spatial resolution data collected by fixed monitoring stations and infer the concentration of air pollutants using additional types of data, e.g., meteorological and traffic information. In this work, we focus on street-level air quality inference by utilizing data collected by mobile stations. We formulate air quality inference in this setting as a graph-based matrix completion problem and propose a novel variational model based on graph convolutional autoencoders. Our model captures effectively the spatio-temporal correlation of the measurements and does not depend on the availability of additional information apart from the street-network topology. Experiments on a real air quality dataset, collected with mobile stations, shows that the proposed model outperforms state-of-the-art approaches.

Duc Minh Nguyen, Evaggelia Tsiligianni, Robert Calderbank et al.

Autoencoders are popular among neural-network-based matrix completion models due to their ability to retrieve potential latent factors from the partially observed matrices. Nevertheless, when training data is scarce their performance is significantly degraded due to overfitting. In this paper, we mit- igate overfitting with a data-dependent regularization technique that relies on the principles of multi-task learning. Specifically, we propose an autoencoder-based matrix completion model that performs prediction of the unknown matrix values as a main task, and manifold learning as an auxiliary task. The latter acts as an inductive bias, leading to solutions that generalize better. The proposed model outperforms the existing autoencoder-based models designed for matrix completion, achieving high reconstruction accuracy in well-known datasets.

Duc Minh Nguyen, Evaggelia Tsiligianni, Nikos Deligiannis

Matrix completion is one of the key problems in signal processing and machine learning, with applications ranging from image pro- cessing and data gathering to classification and recommender sys- tems. Recently, deep neural networks have been proposed as la- tent factor models for matrix completion and have achieved state- of-the-art performance. Nevertheless, a major problem with existing neural-network-based models is their limited capabilities to extend to samples unavailable at the training stage. In this paper, we propose a deep two-branch neural network model for matrix completion. The proposed model not only inherits the predictive power of neural net- works, but is also capable of extending to partially observed samples outside the training set, without the need of retraining or fine-tuning. Experimental studies on popular movie rating datasets prove the ef- fectiveness of our model compared to the state of the art, in terms of both accuracy and extendability.

Tien Huu Do, Duc Minh Nguyen, Evaggelia Tsiligianni et al.

Predicting the geographical location of users on social networks like Twitter is an active research topic with plenty of methods proposed so far. Most of the existing work follows either a content-based or a network-based approach. The former is based on user-generated content while the latter exploits the structure of the network of users. In this paper, we propose a more generic approach, which incorporates not only both content-based and network-based features, but also other available information into a unified model. Our approach, named Multi-Entry Neural Network (MENET), leverages the latest advances in deep learning and multiview learning. A realization of MENET with textual, network and metadata features results in an effective method for Twitter user geolocation, achieving the state of the art on two well-known datasets.

Huynh Van Luong, Nikos Deligiannis, Søren Forchhammer et al.

We consider a decomposition method for compressive streaming data in the context of online compressive Robust Principle Component Analysis (RPCA). The proposed decomposition solves an $n$-$\ell_1$ cluster-weighted minimization to decompose a sequence of frames (or vectors), into sparse and low-rank components, from compressive measurements. Our method processes a data vector of the stream per time instance from a small number of measurements in contrast to conventional batch RPCA, which needs to access full data. The $n$-$\ell_1$ cluster-weighted minimization leverages the sparse components along with their correlations with multiple previously-recovered sparse vectors. Moreover, the proposed minimization can exploit the structures of sparse components via clustering and re-weighting iteratively. The method outperforms the existing methods for both numerical data and actual video data.

Tien Huu Do, Duc Minh Nguyen, Evaggelia Tsiligianni et al.

The problem of predicting the location of users on large social networks like Twitter has emerged from real-life applications such as social unrest detection and online marketing. Twitter user geolocation is a difficult and active research topic with a vast literature. Most of the proposed methods follow either a content-based or a network-based approach. The former exploits user-generated content while the latter utilizes the connection or interaction between Twitter users. In this paper, we introduce a novel method combining the strength of both approaches. Concretely, we propose a multi-entry neural network architecture named MENET leveraging the advances in deep learning and multiview learning. The generalizability of MENET enables the integration of multiple data representations. In the context of Twitter user geolocation, we realize MENET with textual, network, and metadata features. Considering the natural distribution of Twitter users across the concerned geographical area, we subdivide the surface of the earth into multi-scale cells and train MENET with the labels of the cells. We show that our method outperforms the state of the art by a large margin on three benchmark datasets.

Pingfan Song, Xin Deng, João F. C. Mota et al.

Real-world data processing problems often involve various image modalities associated with a certain scene, including RGB images, infrared images or multi-spectral images. The fact that different image modalities often share certain attributes, such as certain edges, textures and other structure primitives, represents an opportunity to enhance various image processing tasks. This paper proposes a new approach to construct a high-resolution (HR) version of a low-resolution (LR) image given another HR image modality as reference, based on joint sparse representations induced by coupled dictionaries. Our approach, which captures the similarities and disparities between different image modalities in a learned sparse feature domain in \emph{lieu} of the original image domain, consists of two phases. The coupled dictionary learning phase is used to learn a set of dictionaries that couple different image modalities in the sparse feature domain given a set of training data. In turn, the coupled super-resolution phase leverages such coupled dictionaries to construct a HR version of the LR target image given another related image modality. One of the merits of our sparsity-driven approach relates to the fact that it overcomes drawbacks such as the texture copying artifacts commonly resulting from inconsistency between the guidance and target images. Experiments on real multimodal images demonstrate that incorporating appropriate guidance information via joint sparse representation induced by coupled dictionary learning brings notable benefits in the super-resolution task with respect to the state-of-the-art. Of particular relevance, the proposed approach also demonstrates better robustness than competing deep-learning-based methods in the presence of noise.

Duc Minh Nguyen, Evaggelia Tsiligianni, Nikos Deligiannis

Compressed sensing (CS) is a sampling theory that allows reconstruction of sparse (or compressible) signals from an incomplete number of measurements, using of a sensing mechanism implemented by an appropriate projection matrix. The CS theory is based on random Gaussian projection matrices, which satisfy recovery guarantees with high probability; however, sparse ternary {0, -1, +1} projections are more suitable for hardware implementation. In this paper, we present a deep learning approach to obtain very sparse ternary projections for compressed sensing. Our deep learning architecture jointly learns a pair of a projection matrix and a reconstruction operator in an end-to-end fashion. The experimental results on real images demonstrate the effectiveness of the proposed approach compared to state-of-the-art methods, with significant advantage in terms of complexity.

Huynh Van Luong, Nikos Deligiannis, Jurgen Seiler et al.

We consider an online version of the robust Principle Component Analysis (PCA), which arises naturally in time-varying source separations such as video foreground-background separation. This paper proposes a compressive online robust PCA with prior information for recursively separating a sequences of frames into sparse and low-rank components from a small set of measurements. In contrast to conventional batch-based PCA, which processes all the frames directly, the proposed method processes measurements taken from each frame. Moreover, this method can efficiently incorporate multiple prior information, namely previous reconstructed frames, to improve the separation and thereafter, update the prior information for the next frame. We utilize multiple prior information by solving $n\text{-}\ell_{1}$ minimization for incorporating the previous sparse components and using incremental singular value decomposition ($\mathrm{SVD}$) for exploiting the previous low-rank components. We also establish theoretical bounds on the number of measurements required to guarantee successful separation under assumptions of static or slowly-changing low-rank components. Using numerical experiments, we evaluate our bounds and the performance of the proposed algorithm. In addition, we apply the proposed algorithm to online video foreground and background separation from compressive measurements. Experimental results show that the proposed method outperforms the existing methods.

Huynh Van Luong, Nikos Deligiannis, Søren Forchhammer et al.

In support of applications involving multiview sources in distributed object recognition using lightweight cameras, we propose a new method for the distributed coding of sparse sources as visual descriptor histograms extracted from multiview images. The problem is challenging due to the computational and energy constraints at each camera as well as the limitations regarding inter-camera communication. Our approach addresses these challenges by exploiting the sparsity of the visual descriptor histograms as well as their intra- and inter-camera correlations. Our method couples distributed source coding of the sparse sources with a new joint recovery algorithm that incorporates multiple side information signals, where prior knowledge (low quality) of all the sparse sources is initially sent to exploit their correlations. Experimental evaluation using the histograms of shift-invariant feature transform (SIFT) descriptors extracted from multiview images shows that our method leads to bit-rate saving of up to 43% compared to the state-of-the-art distributed compressed sensing method with independent encoding of the sources.

Nikos Deligiannis, Joao F. C. Mota, Bruno Cornelis et al.

In support of art investigation, we propose a new source separation method that unmixes a single X-ray scan acquired from double-sided paintings. In this problem, the X-ray signals to be separated have similar morphological characteristics, which brings previous source separation methods to their limits. Our solution is to use photographs taken from the front and back-side of the panel to drive the separation process. The crux of our approach relies on the coupling of the two imaging modalities (photographs and X-rays) using a novel coupled dictionary learning framework able to capture both common and disparate features across the modalities using parsimonious representations; the common component models features shared by the multi-modal images, whereas the innovation component captures modality-specific information. As such, our model enables the formulation of appropriately regularized convex optimization procedures that lead to the accurate separation of the X-rays. Our dictionary learning framework can be tailored both to a single- and a multi-scale framework, with the latter leading to a significant performance improvement. Moreover, to improve further on the visual quality of the separated images, we propose to train coupled dictionaries that ignore certain parts of the painting corresponding to craquelure. Experimentation on synthetic and real data - taken from digital acquisition of the Ghent Altarpiece (1432) - confirms the superiority of our method against the state-of-the-art morphological component analysis technique that uses either fixed or trained dictionaries to perform image separation.

Nikos Deligiannis, João F. C. Mota, Bruno Cornelis et al.

In support of art investigation, we propose a new source sepa- ration method that unmixes a single X-ray scan acquired from double-sided paintings. Unlike prior source separation meth- ods, which are based on statistical or structural incoherence of the sources, we use visual images taken from the front- and back-side of the panel to drive the separation process. The coupling of the two imaging modalities is achieved via a new multi-scale dictionary learning method. Experimental results demonstrate that our method succeeds in the discrimination of the sources, while state-of-the-art methods fail to do so.