ProtoPFL: Enhancing privacy and accuracy in federated time-series forecasting with heterogeneous data

This presentation explores the analysis of heterogeneous geospatial data from various sources through the application of artificial intelligence (AI) tools. Wastewater networks are used as a case study to address challenges such as data completion, multi-source integration, and managing diverse data formats, including Geographic Information Systems (GIS), analog maps, and pipe inspection videos, all derived from real-world data. We will review some solutions developed under the European project Starwars (STormwAteR and WastewAteR networkS heterogeneous data AI-driven management). These solutions are based on innovative models and tools that employ logical and graph-based representations of heterogeneous data. Specifically, we aim to represent different data types — such as GIS, ITV inspection videos, and maps — as annotated graphs, incorporating the uncertainty stemming from incomplete or inconsistent information.

The study of the prediction of stock market volatility is of great significance to rationally control financial market risks and increase excessive investment returns and has received extensive attention from academic and commercial circles. However, as a dynamic and complex system, the stock market is affected by multiple factors and has a comprehensive capability to include complex financial data. Given that the explanatory variables of influencing factors are diverse, heterogeneous and complex, the existing intelligent algorithms have great limitations for the analysis and processing of multi-source heterogeneous data in the stock market. Therefore, this study adopts the edge weight and information transmission mechanism suitable for subgraph data to complete node screening, the gate recurrent unit (GRU) and long short-term memory (LSTM) to aggregate subgraph nodes. The compiled data contain the metapaths of three types of index data, and the introduction of the association relationship attention dimension effectively mines the implicit meanings of multi-source heterogeneous data. The metapath attention mechanism is combined with a graph neural network to complete the classification of multi-source heterogeneous graph data, by which the prediction of stock market volatility is realized. The results show that the above method is feasible for the fusion of heterogeneous stock market data and the mining of implicit semantic information of association relations. The accuracy of the proposed method for the prediction of stock market volatility in this study is 16.64% higher than that of the dimensional reduction index and 14.48% higher than that of other methods for the fusion and prediction of heterogeneous data using the same model.

Cloud storage has recently appeared as a promising solution to handle the immense volume of data produced in nowadays’ rich digital environment. In order to address the departmental information silos, the digital divide, and truly realize the sharing of information, we can outsourced the heterologous and heterogeneous mass data to the cloud so as to reduce the costs of service and improve the fundamental problem of human services. In this paper, we propose an adaptive unified storage scheme for the heterogeneous and heterologous data in cloud center, which manages different storage methods and technology in a unified way. According to the data characteristics and users’ demands, it provides adaptive unified storage services and make the underlying storage technology completely transparent to the upper layer application. Providing consistent storage services through the global cloud storage management can effectively simplify cloud center storage management, reduce storage costs as well as improve the efficiency of data access overall. The simulation results show that the proposed method effectively achieves the strategic thinking of data storage costs is proportional to the value of data.KeywordsUnified storageHeterologous and heterogeneousDistribution on demandCloud storage

As the complexity of intelligent public transportation systems continues to increase and the degree of interconnection of equipment at all levels gradually deepens, building a digital twin system will generate a large amount of multi-source heterogeneous data. Effective processing and in-depth mining of data can provide real-time and intelligent decision-making optimization for the construction of digital twin systems, thereby efficiently handling complex emergencies. This paper conducts a systematic review on the processing methods and technologies of multi-source heterogeneous twin data in traffic dynamic scenarios. First, the content and classification of multi-source heterogeneous twin data in the operation process of intelligent bus systems are clarified; secondly, the multi-source The data processing methods and technologies applied in each stage of data collection, data storage, data fusion and data analysis in heterogeneous twin data processing are analyzed, and the advantages, disadvantages and applications of various methods and technologies are analyzed; finally, corresponding to the multiple This paper summarizes the methods and technologies for processing multi-source heterogeneous twin data, and points out the challenges and development trends faced by the current multi-source heterogeneous twin data processing methods and technologies.

To preserve and restore ecosystems, public environmental policies on the international level now encourage the creation of Green Infrastructure, i.e. networks composed of areas where animal and vegetal species can live (habitat patches) and corridors to circulate between them. In France, each regions, the first subnational administrative level, identified existing habitat patches and corridors in their territories using flexible guidelines. This resulted in heterogeneous data, raising the question of their consistent mapping at a supra-regional level. To answer this question, this study first focuses on the habitat patches of two adjacent regions and explores three ways of homogenizing the map. The first method consists in generalizing the more detailed data using morphologic operators. The second method consists in graphically refining the less detailed data by filling the areas with patterns taken from the more detailed data. The third method consists in drastically changing the level of abstraction of the data from both regions by rasterizing the space. Based on those experiments, we applied the most appropriate method to data collected by all the regions of continental France, a step which itself raises new issues concerning data harmonization and parameters settings.

Modeling heterogeneous data sources remains a fundamental challenge of acoustic modeling in speech recognition. We call this the multi-condition problem because the speech data come from many different conditions. In this paper, we introduce the fundamental confusability problem in multi-condition learning, then discuss the problem formalization, the taxonomy, and the architectures for multi-condition learning. While the ideas presented are applicable to all classifiers, we focus our attention in this work on acoustic models based on deep neural networks (DNN). We propose four different strategies for multi-condition learning of a DNN that we refer to as a mixed-condition model, a condition-dependent model, a condition-normalizing model, and a condition-aware model. Based on the experimental results on the voice search and short message dictation task and the Aurora 4 task, we show that the confusability introduced when modeling heterogeneous data depends on the source of acoustic distortion itself, the front-end feature extractor, and the classifier. We also demonstrate the best approach for dealing with heterogeneous data may not be to let the model sort it out blindly, even with a classifier as sophisticated as a DNN.

The design of large and complex equipment involves a good number of design software, which require and produce heterogeneous data. Viewing and managing data from various type and format poses a great difficulty for engineering designers. In order to solve these difficulties, we propose an intelligent interface system for handling heterogeneous design data. The intelligent interface system can understand and integrate data from heterogeneous data files. Data of different type and format can be specified, presented and managed in a unified way. From our system implementation we can conclude that our approach can greatly improve production efficiency of engineering designers.

This article is part of the Focus Theme of METHODS of Information in Medicine on "Managing Interoperability and Complexity in Health Systems". The increasing availability of electronic clinical data provides great potential for finding eligible patients for clinical research. However, data heterogeneity makes it difficult for clinical researchers to interrogate sources consistently. Existing standard query languages are often not sufficient to query across diverse representations. Thus, a higher-level domain language is needed so that queries become data-representation agnostic. To this end, we define a clinician-readable computational language for querying whether patients meet eligibility criteria (ECs) from clinical trials. This language is capable of implementing the temporal semantics required by many ECs, and can be automatically evaluated on heterogeneous data sources. By reference to standards and examples of existing ECs, a clinician-readable query language was developed. Using a model-based approach, it was implemented to transform captured ECs into queries that interrogate heterogeneous data warehouses. The query language was evaluated on two types of data sources, each different in structure and content. The query language abstracts the level of expressivity so that researchers construct their ECs with no prior knowledge of the data sources. It was evaluated on two types of semantically and structurally diverse data warehouses. This query language is now used to express ECs in the EHR4CR project. A survey shows that it was perceived by the majority of users to be useful, easy to understand and unambiguous. An EC-specific language enables clinical researchers to express their ECs as a query such that the user is isolated from complexities of different heterogeneous clinical data sets. More generally, the approach demonstrates that a domain query language has potential for overcoming the problems of semantic interoperability and is applicable where the nature of the queries is well understood and the data is conceptually similar but in different representations. Our language provides a strong basis for use across different clinical domains for expressing ECs by overcoming the heterogeneous nature of electronic clinical data whilst maintaining semantic consistency. It is readily comprehensible by target users. This demonstrates that a domain query language can be both usable and interoperable.

The accumulated gene expression and DNA methylation data provide a great opportunity to exploit the mechanisms of biological systems. Current algorithms for the integration of gene expression and methylation data are characterized for undesirable performance because they fail to address the latent relations in the heterogeneous data. To solve this problem, we propose a novel multi-view clustering with self-representation learning and low-rank tensor constraint (MCSL-LTC), where the gene expression and DNA methylation data are treated as complementary views, and MCSL-LTC obtains a consensus partitioning reflecting the structure and features of various views. Specifically, self-representation learning is employed to explore the low-dimensional subspace structures embedded in different views, where the tensor norm is adopted to smooth different views, therefore improving the quality of features. Experimental results demonstrate that the proposed approach outperforms state-of-the-art baselines in terms of accuracy on both the social and cancer data, provides an effective and efficient method for the integration of heterogeneous genomic data.

This paper proposes the use of the bootstrap in penalized model selection for possibly dependent heterogeneous data. The results show that we can establish (at least asymptotically) a direct relationship between estimation error and a data based complexity penalization. This requires redefinition of the target function as the sum of the individual expected predicted risks. In this framework, the wild bootstrap and related approaches can be used to estimate the penalty with no need to account for heterogeneous dependent data. The methodology is highlighted by a simulation study whose results are particularly encouraging.

Biological signaling networks process extracellular cues to control important cell decisions such as death–survival, growth–quiescence, and proliferation–differentiation. After receptor activation, intracellular signaling proteins change in abundance, modification state, and enzymatic activity. Many of the proteins in signaling networks have been identified, but it is not known how signaling molecules work together to control cell decisions. To begin to address this issue, we report the use of partial least squares regression as an analytical method to glean signal–response relationships from heterogeneous multivariate signaling data collected from HT-29 human colon carcinoma cells stimulated to undergo programmed cell death. By partial least squares modeling, we relate dynamic and quantitative measurements of 20–30 intracellular signals to cell survival after treatment with tumor necrosis factor alpha (a death factor) and insulin (a survival factor). We find that partial least squares models can distinguish highly informative signals from redundant uninformative signals to generate a reduced model that retains key signaling features and signal–response relationships. In these models, measurements of biochemical characteristics, based on very different techniques (Western blots, kinase assays, etc.), are grouped together as covariates, showing that heterogenous data have been effectively fused. Importantly, informative protein predictors of cell responses are always multivariate, demonstrating the multicomponent nature of the decision process.

During an epidemic, the spatial, temporal and demographical patterns of disease transmission are determined by multiple factors. Besides the physiological properties of pathogenes and hosts, the social contacts of host population, which characterize individuals' reciprocal exposures of infection in view of demographical structures and various social activities, are also pivotal to understand and further predict the prevalence of infectious diseases. The means of measuring social contacts will dominate the extent how precisely we can forecast the dynamics of infections in the real world. Most current works focus their efforts on modeling the spatial patterns of static social contacts. In this work, we address the problem on how to characterize and measure dynamical social contacts during an epidemic from a novel perspective. We propose an epidemic-model-based tensor deconvolution framework to address this issue, in which the spatiotemporal patterns of social contacts are represented by the factors of tensors, which can be discovered by a tensor deconvolution procedure with an integration of epidemic models from rich types of data, mainly including heterogeneous outbreak surveillance, social-demographic census and physiological data from medical reports. Taking SIR model as a case study, the efficacy of the proposed method is theoretically analyzed and empirically validated through a set of rigorous experiments on both synthetic and real-world data.

Process task duration data often exhibit multiple peaks, indicating differences in, for example, customer ages and preferences, resource capabilities or the day/hour of a week. This heterogeneous data, which captures diverse customer patterns, should be represented using different models, resulting in an overall mixture model. This paper introduces gamma mixture models to represent various customer patterns in task duration data, with a focus on automating the fitting process. The approach involves a two-stage procedure: first, divide-and-conquer using peak-, equidistance- and cluster-based techniques to partition data, and automatically fit gamma distributions to each subset. The second stage then improves the fitted mixture model by directly searching the log-likelihood surface. The method is compared with the expectation–maximization (EM) algorithm and an open tool (HyperStar), using both artificially generated datasets and a publicly available hospital billing dataset, demonstrating its effectiveness and time efficiency in modelling heterogeneous process duration data. Furthermore, a case study on process conformance checking is conducted using the hospital billing dataset, highlighting a potential application area for the method in process mining.

Mobile crowdsensing is a new paradigm that can utilize pervasive smartphones to collect and analyze data to benefit users. However, sensory data gathered by smartphone usually involves different data types because of different granularity and multiple sensor sources. Besides, the data are also time labelled. The heterogeneous and time sequential data raise new challenges for data analyzing. Some existing solutions try to learn each type of data one by one and analyze them separately without considering time information. In addition, the traditional methods also have to determine phone orientation because some sensors equipped in smartphone are orientation related. In this paper, we think that a combination of multiple sensors can represent an invariant feature for a crowdsensing context. Therefore, we propose a new representation learning method of heterogeneous data with time labels to extract typical features using deep learning. We evaluate that our proposed method can adapt data generated by different orientations effectively. Furthermore, we test the performance of the proposed method by recognizing two group mobile activities, walking/cycling and driving/bus with smartphone sensors. It achieves precisions of98.6%and93.7%in distinguishing cycling from walking and bus from driving, respectively.

Rock grade is a key indicator guiding tunnel construction. In order to ensure the efficiency and safety of construction, it is necessary to accurately predict the rock grade of the unexcavated part of a tunnel. Currently, geological sketches and geophysical exploration methods can be employed to obtain multi-source and heterogeneous detection data. However, the key challenge lies in how to integrate various types of exploration data to predict the rock grade, which is the focus of the current research. In this paper, we propose a multi-source information fusion-based rock-grade hybrid model for the tunnel construction process. The proposed approach consists of several steps. In the first step, homogenization processing of the acquired multi-source and heterogeneous data, such as geological and TSP (Tunnel Seismic Prediction) detection data, is performed. This primarily includes feature extraction, spatial registration, and the filtering of anomalous data, aimed at enhancing the quality of the data. In the second step, considering the variations in the geological conditions of the construction face, this paper first stratifies the rock grades at the construction face. Subsequently, utilizing TSP detection data, a rock-grade prediction model is established by combining knowledge-driven and data-driven approaches. In the third step, based on the rock grade predictions obtained from the rock grade forecasting model established in the second step, an intelligent decision-making process is conducted by comparing these predictions with the rock grades anticipated during the design stage. This results in the determination of the final rock grade. Finally, the effectiveness of the proposed method is validated through comparison with experimental results.