Search

Systems characterized by evolving interactions, prevalent in social, financial, and biological domains, are effectively modeled as continuous-time dynamic graphs (CTDGs). To manage the scale and complexity of these graph datasets, machine learning (ML) approaches have become essential. However, CTDGs pose challenges for ML because traditional static graph methods do not naturally account for event timings. Newer approaches, such as graph recurrent neural networks (GRNNs), are inherently time-aware and offer advantages over static methods for CTDGs. However, GRNNs face another issue: the short truncation of backpropagation-through-time (BPTT), whose impact has not been properly examined until now. In this work, we demonstrate that this truncation can limit the learning of dependencies beyond a single hop, resulting in reduced performance. Through experiments on a novel synthetic task and real-world datasets, we reveal a performance gap between full backpropagation-through-time (F-BPTT) and the truncated backpropagation-through-time (T-BPTT) commonly used to train GRNN models. We term this gap the "truncation gap" and argue that understanding and addressing it is essential as the importance of CTDGs grows, discussing potential future directions for research in this area., Comment: Published in Transactions on Machine Learning Research (TMLR)

Data used by automated decision-making systems, such as Machine Learning models, often reflects discriminatory behavior that occurred in the past. These biases in the training data are sometimes related to label noise, such as in COMPAS, where more African-American offenders are wrongly labeled as having a higher risk of recidivism when compared to their White counterparts. Models trained on such biased data may perpetuate or even aggravate the biases with respect to sensitive information, such as gender, race, or age. However, while multiple label noise correction approaches are available in the literature, these focus on model performance exclusively. In this work, we propose Fair-OBNC, a label noise correction method with fairness considerations, to produce training datasets with measurable demographic parity. The presented method adapts Ordering-Based Noise Correction, with an adjusted criterion of ordering, based both on the margin of error of an ensemble, and the potential increase in the observed demographic parity of the dataset. We evaluate Fair-OBNC against other different pre-processing techniques, under different scenarios of controlled label noise. Our results show that the proposed method is the overall better alternative within the pool of label correction methods, being capable of attaining better reconstructions of the original labels. Models trained in the corrected data have an increase, on average, of 150% in demographic parity, when compared to models trained in data with noisy labels, across the considered levels of label noise.

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.

Adversarial attacks are a major concern in security-centered applications, where malicious actors continuously try to mislead Machine Learning (ML) models into wrongly classifying fraudulent activity as legitimate, whereas system maintainers try to stop them. Adversarially training ML models that are robust against such attacks can prevent business losses and reduce the work load of system maintainers. In such applications data is often tabular and the space available for attackers to manipulate undergoes complex feature engineering transformations, to provide useful signals for model training, to a space attackers cannot access. Thus, we propose a new form of adversarial training where attacks are propagated between the two spaces in the training loop. We then test this method empirically on a real world dataset in the domain of credit card fraud detection. We show that our method can prevent about 30% performance drops under moderate attacks and is essential under very aggressive attacks, with a trade-off loss in performance under no attacks smaller than 7%.

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., Comment: 9 pages, 5 figures, 7 tables

Monitoring the behavior of automated real-time stream processing systems has become one of the most relevant problems in real world applications. Such systems have grown in complexity relying heavily on high dimensional input data, and data hungry Machine Learning (ML) algorithms. We propose a flexible system, Feature Monitoring (FM), that detects data drifts in such data sets, with a small and constant memory footprint and a small computational cost in streaming applications. The method is based on a multi-variate statistical test and is data driven by design (full reference distributions are estimated from the data). It monitors all features that are used by the system, while providing an interpretable features ranking whenever an alarm occurs (to aid in root cause analysis). The computational and memory lightness of the system results from the use of Exponential Moving Histograms. In our experimental study, we analyze the system's behavior with its parameters and, more importantly, show examples where it detects problems that are not directly related to a single feature. This illustrates how FM eliminates the need to add custom signals to detect specific types of problems and that monitoring the available space of features is often enough., Comment: 10 pages, 5 figures. AutoML, KDD22, August 14-17, 2022, Washington, DC, US

The widespread use of Internet-of-Things (IoT) across different application domains leads to an increased concern regarding their dependability, especially as the number of potentially mission-critical systems becomes considerable. Fault-tolerance has been used to reduce the impact of faults in systems, and their adoption in IoT is becoming a necessity. This work focuses on how to exercise fault-tolerance mechanisms by deliberately provoking its malfunction. We start by describing a proof-of-concept fault-injection add-on to a commonly used publish/subscribe broker. We then present several experiments mimicking real-world IoT scenarios, focusing on injecting faults in systems with (and without) active self-healing mechanisms and comparing their behavior to the baseline without faults. We observe evidence that fault-injection can be used to (a) exercise in-place fault-tolerance apparatus, and (b) detect when these mechanisms are not performing nominally, providing insights into enhancing in-place fault-tolerance techniques., Comment: 8 pages

Internet-of-Things (IoT) systems have spread among different application domains, from home automation to industrial manufacturing processes. The rushed development by competing vendors to meet the market demand of IoT solutions, the lack of interoperability standards, and the overall lack of a defined set of best practices have resulted in a highly complex, heterogeneous, and frangible ecosystem. Several works have been pushing towards visual programming solutions to abstract the underlying complexity and help humans reason about it. As these solutions begin to meet widespread adoption, their building blocks usually do not consider reliability issues. Node-RED, being one of the most popular tools, also lacks such mechanisms, either built-in or via extensions. In this work we present SHEN (Self-Healing Extensions for Node-RED) which provides 17 nodes that collectively enable the implementation of self-healing strategies within this visual framework. We proceed to demonstrate the feasibility and effectiveness of the approach using real devices and fault injection techniques., Comment: Paper accepted for publication as part of the 3rd International Workshop on Software Engineering Research & Practices for the Internet of Things (SERP4IoT 2021), part of 43rd International Conference on Software Engineering (ICSE 2021)

Industrial Internet-of-Things (IIoT) results from the addition of sensing and actuating capabilities to industrial environments to improve the overall manufacturing processes. Some of these systems have highly-complex tasks of monitorization and control and need to be programmed accordingly. The use of visual programming, such as workflows, is common in these systems due to the abstraction they provide to the systems programmer. However, such programming environments have several deficiencies on what regards debugging capabilities, mostly due to the constraints that difficult the use of traditional mechanisms. The work presented in this paper approaches these issues, delving into the design and implementation of a multi-strategy debugging mechanism into a commercial-grade Manufacturing Execution System. To validate the approach, a proof-of-concept was then developed and validated against different debugging scenarios.

A number of approaches have been proposed to identify service boundaries when decomposing a monolith to microservices. However, only a few use systematic methods and have been demonstrated with replicable empirical studies. We describe a systematic approach for refactoring systems to microservice architectures that uses static analysis to determine the system's structure and dynamic analysis to understand its actual behavior. A prototype of a tool was built using this approach (MonoBreaker) and was used to conduct a case study on a real-world software project. The goal was to assess the feasibility and benefits of a systematic approach to decomposition that combines static and dynamic analysis. The three study participants regarded as positive the decomposition proposed by our tool, and considered that it showed improvements over approaches that rely only on static analysis.

Software has a longstanding association with a state of crisis considering its success rate. The explosion of Internet-connected devices, Internet-of-Things, adds to the complexity of software systems. The particular characteristics of these systems, such as being large-scale and its heterogeneity, pose increasingly new challenges. In this paper, we first briefly introduce the IoT paradigm and the current state of art of software development. Then, we delve into the particularities of developing software for IoT systems and systems of systems, given an overview of what are the current methodologies and tools for design, develop and test such systems. The findings are discussed, revealing open issues and research directions, and reveal that the nowadays IoT software development practices are still lagging behind of what are the current best practices., Comment: 38 pages

Knee injuries are frequent, varied and often require the patient to undergo intensive rehabilitation for several months. Treatment protocols usually contemplate some recurrent measurements in order to assess progress, such as goniometry. The need for specific equipment or the complexity and duration of these tasks cause them to often be neglected. A novel deep learning based solution is presented, supported by the generation of a synthetic image dataset. A 3D human-body model was used for this purpose, simulating a recovering patient. For each image, the coordinates of three key points were registered: the centers of the thigh, the knee and the lower leg. These values are sufficient to estimate the flexion angle. Convolutional neural networks were then trained for predicting these six coordinates. Transfer learning was used with the VGG16 and InceptionV3 models pre-trained on the ImageNet dataset, being an additional custom model trained from scratch. All models were tested with different combinations of data augmentation techniques applied on the training sets. InceptionV3 achieved the best overall results, producing considerably good predictions even on real unedited pictures., Comment: 8 pages, 9 figures

Smart space management can be done in many ways. On one hand, there are conversational assistants such as the Google Assistant or Amazon Alexa that enable users to comfortably interact with smart spaces with only their voice, but these have limited functionality and are usually limited to simple commands. On the other hand, there are visual interfaces such as IBM's Node-RED that enable complex features and dependencies between different devices. However, these are limited since they require users to have a technical knowledge of how the smart devices work and the system's interface is more complicated and harder to use since they require a computer. This project proposes a new conversational assistant - Jarvis - that combines the ease of use of current assistants with the operational complexity of the visual platforms. The goal of Jarvis is to make it easier to manage smart spaces by providing intuitive commands and useful features. Jarvis integrates with already existing user interfaces such as the Google Assistant, Slack or Facebook Messenger, making it very easy to integrate with existing systems. Jarvis also provides an innovative feature - causality queries - that enable users to ask it why something happened. For example, a user can ask "why did the light turn on?" to understand how the system works., Comment: 20 pages, 4 figures, 2 tables

The IoT area has grown significantly in the last few years and is expected to reach a gigantic amount of 50 billion devices by 2020. The appearance of serverless architectures, specifically highlighting FaaS, raises the question of the of using such in IoT environments. Combining IoT with a serverless architectural design can be effective when trying to make use of the local processing power that exists in a local network of IoT devices and creating a fog layer that leverages computational capabilities that are closer to the end-user. In this approach, which is placed between the device and the serverless function, when a device requests for the execution of a serverless function will decide based on previous metrics of execution if the serverless function should be executed locally, in the fog layer of a local network of IoT devices, or if it should be executed remotely, in one of the available cloud servers. Therefore, this approach allows to dynamically allocating functions to the most suitable layer.