Pedro Ribeiro

Ahmad Naser Eddin, Jacopo Bono, David Aparício et al.

Many real-world datasets have an underlying dynamic graph structure, where entities and their interactions evolve over time. Machine learning models should consider these dynamics in order to harness their full potential in downstream tasks. Previous approaches for graph representation learning have focused on either sampling k-hop neighborhoods, akin to breadth-first search, or random walks, akin to depth-first search. However, these methods are computationally expensive and unsuitable for real-time, low-latency inference on dynamic graphs. To overcome these limitations, we propose graph-sprints a general purpose feature extraction framework for continuous-time-dynamic-graphs (CTDGs) that has low latency and is competitive with state-of-the-art, higher latency models. To achieve this, a streaming, low latency approximation to the random-walk based features is proposed. In our framework, time-aware node embeddings summarizing multi-hop information are computed using only single-hop operations on the incoming edges. We evaluate our proposed approach on three open-source datasets and two in-house datasets, and compare with three state-of-the-art algorithms (TGN-attn, TGN-ID, Jodie). We demonstrate that our graph-sprints features, combined with a machine learning classifier, achieve competitive performance (outperforming all baselines for the node classification tasks in five datasets). Simultaneously, graph-sprints significantly reduce inference latencies, achieving close to an order of magnitude speed-up in our experimental setting.

Nicholas Matsumoto, Anil Kumar Saini, Pedro Ribeiro et al.

In many evolutionary computation systems, parent selection methods can affect, among other things, convergence to a solution. In this paper, we present a study comparing the role of two commonly used parent selection methods in evolving machine learning pipelines in an automated machine learning system called Tree-based Pipeline Optimization Tool (TPOT). Specifically, we demonstrate, using experiments on multiple datasets, that lexicase selection leads to significantly faster convergence as compared to NSGA-II in TPOT. We also compare the exploration of parts of the search space by these selection methods using a trie data structure that contains information about the pipelines explored in a particular run.

Ricardo Ribeiro Pereira, Jacopo Bono, João Tiago Ascensão et al.

Machine learning methods to aid defence systems in detecting malicious activity typically rely on labelled data. In some domains, such labelled data is unavailable or incomplete. In practice this can lead to low detection rates and high false positive rates, which characterise for example anti-money laundering systems. In fact, it is estimated that 1.7--4 trillion euros are laundered annually and go undetected. We propose The GANfather, a method to generate samples with properties of malicious activity, without label requirements. We propose to reward the generation of malicious samples by introducing an extra objective to the typical Generative Adversarial Networks (GANs) loss. Ultimately, our goal is to enhance the detection of illicit activity using the discriminator network as a novel and robust defence system. Optionally, we may encourage the generator to bypass pre-existing detection systems. This setup then reveals defensive weaknesses for the discriminator to correct. We evaluate our method in two real-world use cases, money laundering and recommendation systems. In the former, our method moves cumulative amounts close to 350 thousand dollars through a network of accounts without being detected by an existing system. In the latter, we recommend the target item to a broad user base with as few as 30 synthetic attackers. In both cases, we train a new defence system to capture the synthetic attacks.

Ahmad Naser Eddin, Jacopo Bono, David Aparício et al.

Continuous-time dynamic graphs (CTDGs) are essential for modeling interconnected, evolving systems. Traditional methods for extracting knowledge from these graphs often depend on feature engineering or deep learning. Feature engineering is limited by the manual and time-intensive nature of crafting features, while deep learning approaches suffer from high inference latency, making them impractical for real-time applications. This paper introduces Deep-Graph-Sprints (DGS), a novel deep learning architecture designed for efficient representation learning on CTDGs with low-latency inference requirements. We benchmark DGS against state-of-the-art (SOTA) feature engineering and graph neural network methods using five diverse datasets. The results indicate that DGS achieves competitive performance while inference speed improves between 4x and 12x compared to other deep learning approaches on our benchmark datasets. Our method effectively bridges the gap between deep representation learning and low-latency application requirements for CTDGs.

Pedro C. Vieira, Pedro Ribeiro, Viacheslav Borovitskiy

While deep ensembles are widely considered to be the default method for uncertainty quantification in deep learning, their effectiveness for graph-structured data is often simply assumed based on successes in domains like computer vision. We investigate standard deep ensembles specifically for message-passing graph neural networks. Benchmarking across seven datasets representing varied tasks and complexities, we reveal that ensembles provide surprisingly little improvement over a single model. Instead, the observed marginal gains stem primarily from stabilizing optimization noise in point predictions rather than yielding meaningfully better uncertainty estimates. Through an aleatoric-epistemic decomposition, we identify epistemic collapse: independently trained networks consistently converge to overly similar predictions. Because disagreement is the fundamental mechanism through which ensembles capture epistemic uncertainty, this lack of diversity neutralizes their key advantage. Analyzing this phenomenon further, we suggest this collapse is driven by functional rather than weight-space convexity, where distinct parameter solutions induce almost identical behavior. Our results suggest that deep ensemble success does not seamlessly transfer to graph machine learning.

Nuno Saavedra, Pedro Ribeiro, André Coelho et al.

Unmanned Aerial Vehicle (UAV)-assisted networks are increasingly foreseen as a promising approach for emergency response, providing rapid, flexible, and resilient communications in environments where terrestrial infrastructure is degraded or unavailable. In such scenarios, voice radio communications remain essential for first responders due to their robustness; however, their unstructured nature prevents direct integration with automated UAV-assisted network management. This paper proposes SIREN, an AI-driven framework that enables voice-driven perception for UAV-assisted networks. By integrating Automatic Speech Recognition (ASR) with Large Language Model (LLM)-based semantic extraction and Natural Language Processing (NLP) validation, SIREN converts emergency voice traffic into structured, machine-readable information, including responding units, location references, emergency severity, and Quality-of-Service (QoS) requirements. SIREN is evaluated using synthetic emergency scenarios with controlled variations in language, speaker count, background noise, and message complexity. The results demonstrate robust transcription and reliable semantic extraction across diverse operating conditions, while highlighting speaker diarization and geographic ambiguity as the main limiting factors. These findings establish the feasibility of voice-driven situational awareness for UAV-assisted networks and show a practical foundation for human-in-the-loop decision support and adaptive network management in emergency response operations.

Carlos Couto, José Mourão, Mário A. T. Figueiredo et al.

Near an optimal learning point of a neural network, the learning performance of gradient descent dynamics is dictated by the Hessian matrix of the loss function with respect to the network parameters. We characterize the Hessian eigenspectrum for some classes of teacher-student problems, when the teacher and student networks have matching weights, showing that the smaller eigenvalues of the Hessian determine long-time learning performance. For linear networks, we analytically establish that for large networks the spectrum asymptotically follows a convolution of a scaled chi-square distribution with a scaled Marchenko-Pastur distribution. We numerically analyse the Hessian spectrum for polynomial and other non-linear networks. Furthermore, we show that the rank of the Hessian matrix can be seen as an effective number of parameters for networks using polynomial activation functions. For a generic non-linear activation function, such as the error function, we empirically observe that the Hessian matrix is always full rank.

André Coelho, Pedro Ribeiro, Helder Fontes et al.

This position paper presents A4FN, an Agentic Artificial Intelligence (AI) architecture for intent-driven automation in Flying Networks (FNs) using Unmanned Aerial Vehicles (UAVs) as access nodes. A4FN leverages Generative AI and Large Language Models (LLMs) to enable real-time, context-aware network control via a distributed agentic system. It comprises two components: the Perception Agent (PA), which semantically interprets multimodal input -- including imagery, audio, and telemetry data -- from UAV-mounted sensors to derive Service Level Specifications (SLSs); and the Decision-and-Action Agent (DAA), which reconfigures the network based on inferred intents. A4FN embodies key properties of Agentic AI, including autonomy, goal-driven reasoning, and continuous perception-action cycles. Designed for mission-critical, infrastructure-limited scenarios such as disaster response, it supports adaptive reconfiguration, dynamic resource management, and interoperability with emerging wireless technologies. The paper details the A4FN architecture, its core innovations, and open research challenges in multi-agent coordination and Agentic AI integration in next-generation FNs.

Ricardo Ribeiro Pereira, Jacopo Bono, Hugo Ferreira et al.

When the available data for a target domain is limited, transfer learning (TL) methods can be used to develop models on related data-rich domains, before deploying them on the target domain. However, these TL methods are typically designed with specific, static assumptions on the amount of available labeled and unlabeled target data. This is in contrast with many real world applications, where the availability of data and corresponding labels varies over time. Since the evaluation of the TL methods is typically also performed under the same static data availability assumptions, this would lead to unrealistic expectations concerning their performance in real world settings. To support a more realistic evaluation and comparison of TL algorithms and models, we propose a data manipulation framework that (1) simulates varying data availability scenarios over time, (2) creates multiple domains through resampling of a given dataset and (3) introduces inter-domain variability by applying realistic domain transformations, e.g., creating a variety of potentially time-dependent covariate and concept shifts. These capabilities enable simulation of a large number of realistic variants of the experiments, in turn providing more information about the potential behavior of algorithms when deployed in dynamic settings. We demonstrate the usefulness of the proposed framework by performing a case study on a proprietary real-world suite of card payment datasets. Given the confidential nature of the case study, we also illustrate the use of the framework on the publicly available Bank Account Fraud (BAF) dataset. By providing a methodology for evaluating TL methods over time and in realistic data availability scenarios, our framework facilitates understanding of the behavior of models and algorithms. This leads to better decision making when deploying models for new domains in real-world environments.

Yvonne Murray, Martin Sirevåg, Pedro Ribeiro et al.

As a general trend in industrial robotics, an increasing number of safety functions are being developed or re-engineered to be handled in software rather than by physical hardware such as safety relays or interlock circuits. This trend reinforces the importance of supplementing traditional, input-based testing and quality procedures which are widely used in industry today, with formal verification and model-checking methods. To this end, this paper focuses on a representative safety-critical system in an ABB industrial paint robot, namely the High-Voltage electrostatic Control system (HVC). The practical convergence of the high-voltage produced by the HVC, essential for safe operation, is formally verified using a novel and general co-verification framework where hardware and software models are related via platform mappings. This approach enables the pragmatic combination of highly diverse and specialised tools. The paper's main contribution includes details on how hardware abstraction and verification results can be transferred between tools in order to verify system-level safety properties. It is noteworthy that the HVC application considered in this paper has a rather generic form of a feedback controller. Hence, the co-verification framework and experiences reported here are also highly relevant for any cyber-physical system tracking a setpoint reference.

Ahmad Naser Eddin, Jacopo Bono, David Aparício et al.

Money laundering is a global problem that concerns legitimizing proceeds from serious felonies (1.7-4 trillion euros annually) such as drug dealing, human trafficking, or corruption. The anti-money laundering systems deployed by financial institutions typically comprise rules aligned with regulatory frameworks. Human investigators review the alerts and report suspicious cases. Such systems suffer from high false-positive rates, undermining their effectiveness and resulting in high operational costs. We propose a machine learning triage model, which complements the rule-based system and learns to predict the risk of an alert accurately. Our model uses both entity-centric engineered features and attributes characterizing inter-entity relations in the form of graph-based features. We leverage time windows to construct the dynamic graph, optimizing for time and space efficiency. We validate our model on a real-world banking dataset and show how the triage model can reduce the number of false positives by 80% while detecting over 90% of true positives. In this way, our model can significantly improve anti-money laundering operations.

Vanessa Freitas Silva, Maria Eduarda Silva, Pedro Ribeiro et al.

Being able to capture the characteristics of a time series with a feature vector is a very important task with a multitude of applications, such as classification, clustering or forecasting. Usually, the features are obtained from linear and nonlinear time series measures, that may present several data related drawbacks. In this work we introduce NetF as an alternative set of features, incorporating several representative topological measures of different complex networks mappings of the time series. Our approach does not require data preprocessing and is applicable regardless of any data characteristics. Exploring our novel feature vector, we are able to connect mapped network features to properties inherent in diversified time series models, showing that NetF can be useful to characterize time data. Furthermore, we also demonstrate the applicability of our methodology in clustering synthetic and benchmark time series sets, comparing its performance with more conventional features, showcasing how NetF can achieve high-accuracy clusters. Our results are very promising, with network features from different mapping methods capturing different properties of the time series, adding a different and rich feature set to the literature.

Vanessa Freitas Silva, Maria Eduarda Silva, Pedro Ribeiro et al.

There is nowadays a constant flux of data being generated and collected in all types of real world systems. These data sets are often indexed by time, space or both requiring appropriate approaches to analyze the data. In univariate settings, time series analysis is a mature and solid field. However, in multivariate contexts, time series analysis still presents many limitations. In order to address these issues, the last decade has brought approaches based on network science. These methods involve transforming an initial time series data set into one or more networks, which can be analyzed in depth to provide insight into the original time series. This review provides a comprehensive overview of existing mapping methods for transforming time series into networks for a wide audience of researchers and practitioners in machine learning, data mining and time series. Our main contribution is a structured review of existing methodologies, identifying their main characteristics and their differences. We describe the main conceptual approaches, provide authoritative references and give insight into their advantages and limitations in a unified notation and language. We first describe the case of univariate time series, which can be mapped to single layer networks, and we divide the current mappings based on the underlying concept: visibility, transition and proximity. We then proceed with multivariate time series discussing both single layer and multiple layer approaches. Although still very recent, this research area has much potential and with this survey we intend to pave the way for future research on the topic.

David Aparício, Pedro Ribeiro, Tijana Milenković et al.

Global pairwise network alignment (GPNA) aims to find a one-to-one node mapping between two networks that identifies conserved network regions. GPNA algorithms optimize node conservation (NC) and edge conservation (EC). NC quantifies topological similarity between nodes. Graphlet-based degree vectors (GDVs) are a state-of-the-art topological NC measure. Dynamic GDVs (DGDVs) were used as a dynamic NC measure within the first-ever algorithms for GPNA of temporal networks: DynaMAGNA++ and DynaWAVE. The latter is superior for larger networks. We recently developed a different graphlet-based measure of temporal node similarity, graphlet-orbit transitions (GoTs). Here, we use GoTs instead of DGDVs as a new dynamic NC measure within DynaWAVE, resulting in a new approach, GoT-WAVE. On synthetic networks, GoT-WAVE improves DynaWAVE's accuracy by 25% and speed by 64%. On real networks, when optimizing only dynamic NC, each method is superior ~50% of the time. While DynaWAVE benefits more from also optimizing dynamic EC, only GoT-WAVE can support directed edges. Hence, GoT-WAVE is a promising new temporal GPNA algorithm, which efficiently optimizes dynamic NC. Future work on better incorporating dynamic EC may yield further improvements.

Wei Li, Alvaro Miyazawa, Pedro Ribeiro et al.

Controllers for autonomous robotic systems can be specified using state machines. However, these are typically developed in an ad hoc manner without formal semantics, which makes it difficult to analyse the controller. Simulations are often used during the development, but a rigorous connection between the designed controller and the implementation is often overlooked. This paper presents a state-machine based notation, RoboChart, together with a tool to automatically create code from the state machines, establishing a rigorous connection between specification and implementation. In RoboChart, a robot's controller is specified either graphically or using a textual description language. The controller code for simulation is automatically generated through a direct mapping from the specification. We demonstrate our approach using two case studies (self-organized aggregation and swarm taxis) in swarm robotics. The simulations are presented using two different simulators showing the general applicability of our approach.