Data Representation Research Articles

What do Australians affected by cancer think about oncology researchers sharing research data? A cross-sectional survey.

Previous research has shown patients and the public in Australia generally support medical researchers in making de-identified research data available to other scientists. However, this research has focussed on certain types of data and recipients. We surveyed Australians affected by cancer to characterize their attitudes toward the sharing of research data with multiple third parties, including the public. A short, anonymous online survey of Australians with a previous diagnosis of cancer was advertised between October 27, 2022, and February 27, 2023. Quantitative responses were analyzed with descriptive statistics. Free-text responses were coded deductively and summarised using content analysis. In total, 551 respondents contributed data to the survey. There was strong support for cancer researchers sharing non-human and de-identified human research data with clinicians (90% and 95%, respectively) and non-profit researchers (both 94%). However, fewer participants supported sharing data with for-profit researchers (both 64%) or publicly (both 61%). When asked if they would hypothetically consent to researchers at their treatment location using and sharing their de-identified data publicly, only half agreed. In contrast, after being shown a visual representation of the de-identified survey data, 80% of respondents supported sharing it publicly. Australians affected by cancer support the sharing of research data, particularly with clinicians and non-profit researchers. Our results also imply that visualization of the data to be shared may enhance support for making it publicly available. These results should help alleviate any concerns about research participants' attitudes toward data sharing, as well as boost researchers' motivation for sharing.

Efficient distributed association management method of data, model, and knowledge for digital twin railway

ABSTRACT Digital twin railway is a pivotal foundation for the intelligent construction and maintenance of railway engineering projects within extensive open spaces. Its essence is the integrated representation and association management of multi-granularity spatiotemporal data, executable analysis models, and professional knowledge. These elements are characterized by the prominent characteristics of multi-source, heterogeneity, and massive volume. However, current decentralized and independent management strategies often neglect the dynamic coupling relationships between them, and numerous multi-path joins and conversion aggregation operations exist across various spatial scale applications. Consequently, this results in challenges such as the inability to dynamically couple data-model-knowledge and conduct global association retrieval, thereby limiting the potential for real-time analysis and intelligent application capabilities. To address these problems, we first constructed a tripartite graph model ( DMK − Graph model ) that explicitly associates temporal, spatial, and interactive relationships. Subsequently, an association management architecture was proposed, accompanied by a global association graph index ( DMK − Graph index ) and a global-local indexing mechanism. Finally, a prototype system for railway data-model-knowledge association management was developed. The effectiveness of the distributed association management method was demonstrated by employing a case study of high-temperature safety risk analysis in railway tunnel engineering with multi-physics field coupling.

Latent Representation Learning for Geospatial Entities

Representation learning has been instrumental in the success of machine learning, offering compact and performant data representations for diverse downstream tasks. In the spatial domain, it has been pivotal in extracting latent patterns from various data types, including points, polylines, polygons, and networked structures. However, existing approaches often fall short of explicitly capturing both semantic and spatial information, relying on proxies and synthetic features. This paper presents GeoNN, a novel graph neural network-based model designed to learn spatially-aware embeddings for geospatial entities. GeoNN leverages edge features generated from geodesic functions, dynamically selecting relevant features based on relative locations. It introduces both transductive (GeoNN-T) and inductive (GeoNN-I) models, ensuring effective encoding of geospatial features and scalability with entity changes. Extensive experiments demonstrate GeoNN’s effectiveness in location-sensitive superpixel-based graphs and real-world points of interest, outperforming baselines across various evaluation measures.

Multi-task Information Enhancement Recommendation model for educational Self-Directed Learning System

In Self-Directed Learning Systems (SDLS), suggesting specific exercises to students after viewing educational videos is crucial, which promotes the integration of learning resources and improves learning efficiency. Given that exercises and educational videos are distinct learning resources, suggesting exercises centered on educational videos containing knowledge concepts is referred to as cross-type recommendations. Traditional cross-type recommendation methods usually separate data representation and recommendation tasks, which hampers effective knowledge transfer and information fusion. Furthermore, educational resources frequently feature an irregular distribution of information, posing a significant challenge to existing models in extracting crucial information. Besides, due to the inadequate data, the cross-type recommendation system needs help in developing a robust model. To address these challenges, we propose a Multi-task Information Enhancement Recommendation (MIER) Model in this paper. This model provides a comprehensive recommendation capability by integrating resource representation and merging recommendations. Multi-task learning improves the capacity to integrate information and transfer knowledge. Then, we propose an Information Enhancement Encoder (IEE) module that utilizes the attention mechanism to enhance the model’s ability to extract essential information from educational videos and exercises. Finally, to address the problem of inadequate data, we design a fusion structure that incorporates the concept knowledge graph as expert information into the model. To evaluate the effectiveness of the MIER, we compare it with other models on both predicting concepts and recommending exercises task. The corresponding results demonstrate that the MIER outperforms all the other models. We comprehensively evaluated the MIER model’s performance across a range of datasets varying in labeled data quantity. The findings demonstrate that the model consistently achieves outstanding performance and exhibits robustness, even under conditions of limited label availability.

Factor annealing decoupling compositional training method for imbalanced hyperspectral image classification

Factor annealing decoupling compositional training method for imbalanced hyperspectral image classification

G2-MonoDepth: A General Framework of Generalized Depth Inference From Monocular RGB+X Data.

Monocular depth inference is a fundamental problem for scene perception of robots. Specific robots may be equipped with a camera plus an optional depth sensor of any type and located in various scenes of different scales, whereas recent advances derived multiple individual sub-tasks. It leads to additional burdens to fine-tune models for specific robots and thereby high-cost customization in large-scale industrialization. This article investigates a unified task of monocular depth inference, which infers high-quality depth maps from all kinds of input raw data from various robots in unseen scenes. A basic benchmark G2-MonoDepth is developed for this task, which comprises four components: (a) a unified data representation RGB+X to accommodate RGB plus raw depth with diverse scene scale/semantics, depth sparsity ([0%, 100%]) and errors (holes/noises/blurs), (b) a novel unified loss to adapt to diverse depth sparsity/errors of input raw data and diverse scales of output scenes, (c) an improved network to well propagate diverse scene scales from input to output, and (d) a data augmentation pipeline to simulate all types of real artifacts in raw depth maps for training. G2-MonoDepth is applied in three sub-tasks including depth estimation, depth completion with different sparsity, and depth enhancement in unseen scenes, and it always outperforms SOTA baselines on both real-world data and synthetic data.

Emotion recognition with reduced channels using CWT based EEG feature representation and a CNN classifier

Objective. Although emotion recognition has been studied for decades, a more accurate classification method that requires less computing is still needed. At present, in many studies, EEG features are extracted from all channels to recognize emotional states, however, there is a lack of an efficient feature domain that improves classification performance and reduces the number of EEG channels. Approach. In this study, a continuous wavelet transform (CWT)-based feature representation of multi-channel EEG data is proposed for automatic emotion recognition. In the proposed feature, the time-frequency domain information is preserved by using CWT coefficients. For a particular EEG channel, each CWT coefficient is mapped into a strength-to-entropy component ratio to obtain a 2D representation. Finally, a 2D feature matrix, namely CEF2D, is created by concatenating these representations from different channels and fed into a deep convolutional neural network architecture. Based on the CWT domain energy-to-entropy ratio, effective channel and CWT scale selection schemes are also proposed to reduce computational complexity. Main results. Compared with previous studies, the results of this study show that valence and arousal classification accuracy has improved in both 3-class and 2-class cases. For the 2-class problem, the average accuracies obtained for valence and arousal dimensions are 98.83% and 98.95%, respectively, and for the 3-class, the accuracies are 98.25% and 98.68%, respectively. Significance. Our findings show that the entropy-based feature of EEG data in the CWT domain is effective for emotion recognition. Utilizing the proposed feature domain, an effective channel selection method can reduce computational complexity.

BIMO: Bootstrap Inter–Intra Modality at Once Unsupervised Learning for Multivariate Time Series

It is difficult to learn meaningful representations of time-series data since they are sparsely labeled and unpredictable. Hence, we propose bootstrap inter–intra modality at once (BIMO), an unsupervised representation learning method based on time series. Unlike previous works, the proposed BIMO method learns both inter-sample and intra-temporal modality representations simultaneously without negative pairs. BIMO comprises a main network and two auxiliary networks, namely inter-auxiliary and intra-auxiliary networks. The main network is trained to learn inter–intra modality representations sequentially by regulating the use of each auxiliary network dynamically. Thus, BIMO thoroughly learns inter–intra modality representations simultaneously. The experimental results demonstrate that the proposed BIMO method outperforms the state-of-the-art unsupervised methods and achieves comparable performance to existing supervised methods.

Handling Unstructured Image using Generative AI and Dev-Ops

Abstract: This study explores the application of generative artificial intelligence (AI) to convert unstructured medical image data into structured formats. Medical imaging is vital for diagnosis and treatment, yet handling the large volume of unstructured data presents challenges. Utilizing generative AI, particularly convolutional neural networks (CNNs), enables the transformation of raw medical images into structured data representations. By integrating AI techniques, this approach enhances the accessibility and interoperability of medical imaging data, providing valuable insights for healthcare professionals. The proposed methodology aims to streamline the process, optimizing scalability and reproducibility without extensive reliance on DevOps practices. This research signifies a significant step towards leveraging AI for efficient utilization of medical imaging data in healthcare applications

A Review on the Impact of Data Representation on Model Explainability

In recent years, advanced machine learning and artificial intelligence techniques have gained popularity due to their ability to solve problems across various domains with high performance and quality. However, these techniques are often so complex that they fail to provide simple and understandable explanations for the outputs they generate. To address this issue, the field of explainable artificial intelligence has recently emerged. On the other hand, most data generated in different domains are inherently structural; that is, they consist of parts and relationships among them. Such data can be represented using either a simple data-structure or form , such as a vector , or a complex data-structure, such as a graph . The effect of this representation form on the explainability and interpretability of machine learning models is not extensively discussed in the literature. In this survey paper, we review efficient algorithms proposed for learning from inherently structured data, emphasizing how their representation form affects the explainability of learning models. A conclusion of our literature review is that using complex forms or data-structures for data representation improves not only the learning performance, but also the explainability and transparency of the model.

Neighborhood convolutional graph neural network

Graph convolutional networks (GCNs) have emerged as powerful tools for learning representations of graph structured data. A recent advancement is a decoupled GCN, which separates neighborhood aggregation from feature transformation within each graph convolutional layer and yields promising results. However, our investigation on node-level operations reveals its limitations in terms of modeling capacity and scalability, which hinder effective learning of rich node representations of diverse graphs. To address these challenges, we introduce a new paradigm of decoupled GCNs, called a neighborhood convolutional graph neural network (NCGNN). The NCGNN uses a novel neighborhood aggregator to generate a feature matrix of multi-hop neighborhoods of each node before model training. This avoids training the model via the adjacency matrix, effectively controls the training cost, and enhances scalability. Additionally, the NCGNN integrates advanced feature learning components, including a dedicated convolutional neural network for rich neighborhood information extraction and a feature fusion module to seamlessly integrate the raw attributes with the extracted features. Extensive experiments validate the superiority of the NCGNN over representative graph neural networks in node classification tasks, demonstrating its effectiveness and scalability for diverse graph structures.

Uniqueness of Iris Pattern Based on the Auto-Regressive Model

In this paper, we evaluate the uniqueness of a hypothetical iris recognition system that relies upon a nonlinear mapping of iris data into a space of Gaussian codewords with independent components. Given the new data representation, we develop and apply a sphere packing bound for Gaussian codewords and a bound similar to Daugman’s to characterize the maximum iris population as a function of the relative entropy between Gaussian codewords of distinct iris classes. As a potential theoretical approach leading toward the realization of the hypothetical mapping, we work with the auto-regressive model fitted into iris data, after some data manipulation and preprocessing. The distance between a pair of codewords is measured in terms of the relative entropy (log-likelihood ratio statistic is an alternative) between distributions of codewords, which is also interpreted as a measure of iris quality. The new approach to iris uniqueness is illustrated using two toy examples involving two small datasets of iris images. For both datasets, the maximum sustainable population is presented as a function of image quality expressed in terms of relative entropy. Although the auto-regressive model may not be the best model for iris data, it lays the theoretical framework for the development of a high-performance iris recognition system utilizing a nonlinear mapping from the space of iris data to the space of Gaussian codewords with independent components.

Introspection of UBNIN and Modified-UBNIN Algorithms for Structural MRI. Reply to Kelly et al. A Comparison of Brain-State Representations of Binary Neuroimaging Connectivity Data. Comment on “Samantaray et al. Unique Brain Network Identification Number for Parkinson’s and Healthy Individuals Using Structural MRI. Brain Sci. 2023, 13, 1297”

The purpose of this reply is to address the comments given by Kelly et al. on our original paper “Unique Brain Network Identification Number for Parkinson’s and Healthy Individuals using Structural MRI”. We agree to the inadvertent rounding pitfall in our original paper due to the non-inclusion of symbolic math toolbox (MATLAB). We now provide the actual ranges (with decimal values) of the UBNIN values of healthy individuals and those with Parkinson’s disease and further observations. Upon further introspection, we propose another variant, called Modified-UBNIN (UBNIN-MT,MN) which is highly weighted on the node with the highest network degree (i.e., connections). The italicized sentences within inverted commas are statements from Kelly et al.’s comment paper.

A Comparison of Brain-State Representations of Binary Neuroimaging Connectivity Data. Comment on Samantaray et al. Unique Brain Network Identification Number for Parkinson’s and Healthy Individuals Using Structural MRI. Brain Sci. 2023, 13, 1297

Samantaray et al [...]

Interpretable representations in explainable AI: from theory to practice

AbstractInterpretable representations are the backbone of many explainers that target black-box predictive systems based on artificial intelligence and machine learning algorithms. They translate the low-level data representation necessary for good predictive performance into high-level human-intelligible concepts used to convey the explanatory insights. Notably, the explanation type and its cognitive complexity are directly controlled by the interpretable representation, tweaking which allows to target a particular audience and use case. However, many explainers built upon interpretable representations overlook their merit and fall back on default solutions that often carry implicit assumptions, thereby degrading the explanatory power and reliability of such techniques. To address this problem, we study properties of interpretable representations that encode presence and absence of human-comprehensible concepts. We demonstrate how they are operationalised for tabular, image and text data; discuss their assumptions, strengths and weaknesses; identify their core building blocks; and scrutinise their configuration and parameterisation. In particular, this in-depth analysis allows us to pinpoint their explanatory properties, desiderata and scope for (malicious) manipulation in the context of tabular data where a linear model is used to quantify the influence of interpretable concepts on a black-box prediction. Our findings lead to a range of recommendations for designing trustworthy interpretable representations; specifically, the benefits of class-aware (supervised) discretisation of tabular data, e.g., with decision trees, and sensitivity of image interpretable representations to segmentation granularity and occlusion colour.

Multi-task self-supervised time-series representation learning

Time-series representation learning is crucial for extracting meaningful representations from time-series data with temporal dynamics and sparse labels. Contrastive learning, a powerful technique for exploiting the inherent data patterns, has been applied to explore the diverse consistencies in time-series data, achieved through careful selection of contrastive pairs and design of appropriate loss. Encouraging such consistency is essential for acquiring comprehensive representations of time-series data. In this paper, we propose a new framework for time-series representation learning that combines the advantages of contextual, temporal, and transformation consistencies. This framework enables the network to learn general representations suitable for different tasks and domains. First, positive and negative pairs are generated to establish a multi-task learning setup. Then, contrastive losses are formulated to explore contextual, temporal, and transformation consistencies, which are jointly optimized to learn general time-series representations. In addition, we investigate an uncertainty weighting approach to enhance the effectiveness of multi-task learning. To evaluate the performance of our framework, we conduct experiments on three downstream tasks: time-series classification, forecasting, and anomaly detection. The experimental results demonstrate the superior performance of our framework compared to benchmark models across different tasks. Furthermore, our framework shows efficiency in cross-domain transfer learning scenarios.

Interpretable machine learning models for failure cause prediction in imbalanced oil pipeline data

Pipelines are critical arteries in the oil and gas industry and require massive capital investment to safely construct networks that transport hydrocarbons across diverse environments. However, these pipeline systems are prone to integrity failure, which results in significant economic losses and environmental damage. Accurate prediction of pipeline failure events using historical oil pipeline accident data enables asset managers to plan sufficient maintenance, rehabilitation, and repair activities to prevent catastrophic failures. However, learning the complex interdependencies between pipeline attributes and rare failure events presents several analytical challenges. This study proposes a novel machine learning (ML) framework to accurately predict pipeline failure causes on highly class-imbalanced data compiled by the United States Pipeline and Hazardous Materials Safety Administration. Natural language processing techniques were leveraged to extract informative features from unstructured text data. Furthermore, class imbalance in the dataset was addressed via oversampling and intrinsic cost-sensitive learning (CSL) strategies adapted for the multi-class case. Nine machine and deep learning architectures were benchmarked, with LightGBM demonstrating superior performance. The integration of CSL yielded an 86% F1 score and a 0.82 Cohen kappa score, significantly advancing prior research. This study leveraged a comprehensive Shapley Additive explanation analysis to interpret the predictions from the LightGBM algorithm, revealing the key factors driving failure probabilities. Leveraging sentiment analysis allowed the models to capture a richer, more multifaceted representation of the textual data. This study developed a novel CSL approach that integrates domain knowledge regarding the varying cost impacts of misclassifying different failure types into ML models. This research demonstrated an effective fusion of text insights from inspection reports with structured pipeline data that enhances model interpretability. The resulting AI modeling framework generated data-driven predictions of the causes of failure that could enable transportation agencies with actionable insights. These insights enable tailored preventative maintenance decisions to proactively mitigate emerging pipeline failures.

Thermodynamic signatures and phase transitions in high-energy Au-Au collisions

Abstract In this study, we systematically investigate the dynamics of various particles namely π+, π−, K+, K−, p, ̄p, Λ, ̄Λ, Ξ− and ̄Ξ+ produced in central Au-Au collisions. We analyze data of AGS and RHIC, which span a broad range of collision energies, ranging from √sN N = 1.9 to 200 GeV. To analyze the transverse momentum (pT ) and transverse mass (mT ) distributions, we employ a two-component standard distribution function, achieving a very good representation of the experimental data across these energy regimes. We extract key thermodynamic parameters, including the effective temperature T , the mean transverse momentum ⟨pT ⟩, and the initial temperature Ti, and analyze their dependence on the values of collision energy and particle mass. Our findings reveal a distinct transition behaviour around √sN N = 19.6 GeV. Below √sN N = 19.6 GeV, the values of T , ⟨pT ⟩, and Ti increase monotonically for all particles due to higher energy transfer into the system. Above this energy threshold, these extracted parameters plateau, suggesting that the additional energy is utilized as latent heat for phase transition rather than increasing the system’s temperature. These observations delineate two distinct regions: a baryon-dominated region at lower energies and a parton-dominated region at higher energies, each potentially indicative of different phases of matter, with the latter possibly signalling the onset of a Quark-Gluon Plasma (QGP). The study thus provides critical insights into the complex interplay of thermodynamics, phase transitions, and particle interactions in high-energy Au-Au collisions.

Prototype Learning for Medical Time Series Classification via Human-Machine Collaboration.

Deep neural networks must address the dual challenge of delivering high-accuracy predictions and providing user-friendly explanations. While deep models are widely used in the field of time series modeling, deciphering the core principles that govern the models' outputs remains a significant challenge. This is crucial for fostering the development of trusted models and facilitating domain expert validation, thereby empowering users and domain experts to utilize them confidently in high-risk decision-making contexts (e.g., decision-support systems in healthcare). In this work, we put forward a deep prototype learning model that supports interpretable and manipulable modeling and classification of medical time series (i.e., ECG signal). Specifically, we first optimize the representation of single heartbeat data by employing a bidirectional long short-term memory and attention mechanism, and then construct prototypes during the training phase. The final classification outcomes (i.e., normal sinus rhythm, atrial fibrillation, and other rhythm) are determined by comparing the input with the obtained prototypes. Moreover, the proposed model presents a human-machine collaboration mechanism, allowing domain experts to refine the prototypes by integrating their expertise to further enhance the model's performance (contrary to the human-in-the-loop paradigm, where humans primarily act as supervisors or correctors, intervening when required, our approach focuses on a human-machine collaboration, wherein both parties engage as partners, enabling more fluid and integrated interactions). The experimental outcomes presented herein delineate that, within the realm of binary classification tasks-specifically distinguishing between normal sinus rhythm and atrial fibrillation-our proposed model, albeit registering marginally lower performance in comparison to certain established baseline models such as Convolutional Neural Networks (CNNs) and bidirectional long short-term memory with attention mechanisms (Bi-LSTMAttns), evidently surpasses other contemporary state-of-the-art prototype baseline models. Moreover, it demonstrates significantly enhanced performance relative to these prototype baseline models in the context of triple classification tasks, which encompass normal sinus rhythm, atrial fibrillation, and other rhythm classifications. The proposed model manifests a commendable prediction accuracy of 0.8414, coupled with macro precision, recall, and F1-score metrics of 0.8449, 0.8224, and 0.8235, respectively, achieving both high classification accuracy as well as good interpretability.

Spatial–Temporal Fusion Gated Transformer Network (STFGTN) for Traffic Flow Prediction

Traffic flow prediction is essential for smart city management and planning, aiding in optimizing traffic scheduling and improving overall traffic conditions. However, due to the correlation and heterogeneity of traffic data, effectively integrating the captured temporal and spatial features remains a significant challenge. This paper proposes a model spatial–temporal fusion gated transformer network (STFGTN), which is based on an attention mechanism that integrates temporal and spatial features. This paper proposes an attention mechanism-based model to address these issues and model complex spatial–temporal dependencies in road networks. The self-attention mechanism enables the model to achieve long-term dependency modeling and global representation of time series data. Regarding temporal features, we incorporate a time embedding layer and a time transformer to learn temporal dependencies. This capability contributes to a more comprehensive and accurate understanding of spatial–temporal dynamic patterns throughout the entire time series. As for spatial features, we utilize DGCN and spatial transformers to capture both global and local spatial dependencies, respectively. Additionally, we propose two fusion gate mechanisms to effectively accommodate to the complex correlation and heterogeneity of spatial–temporal information, resulting in a more accurate reflection of the actual traffic flow. Our experiments on three real-world datasets illustrate the superior performance of our approach.