Intermediate Data Research Articles

A neural network approach for unstructured mesh quality evaluation

PurposeThe quality of the unstructured mesh has a considerable impact on the stability and accuracy of aerodynamic simulation in computational fluid dynamics (CFD). Typically, engineers spend a significant portion of their time on mesh quality evaluation to ensure a valid, high-quality mesh. The extensive manual interaction and a priori knowledge required to undertake an accurate and timely evaluation process have become a bottleneck in the idealized efficient CFD workflow. This paper aims to introduce a neural network-based quality evaluation approach for unstructured meshes to enable higher efficiency and the level of automation.Design/methodology/approachThe paper investigates the capability of deep neural networks for the quality evaluation of unstructured meshes. For training the network, we build a training dataset for mesh quality learning algorithms. The dataset contains a rich variety of unstructured aircraft meshes with different mesh sizes, densities, cell distribution, growth ratios and cell numbers to ensure its diversity and availability. We also design a neural network, AircraftNet, to learn the effect of mesh quality on the convergent properties of the numerical solutions. The proposed network directly manipulates raw point data in mesh source files rather than passing it to an intermediate data representation. During training, AircraftNet extracts non-linear quality features from high-dimensional data spaces and then automatically predicts the overall quality of the input unstructured mesh.FindingsThe paper provides a series of experimental results on GPUs. It shows that AircraftNet is able to effectively analyze the quality-related features like mesh density and distribution from the extracted features and achieve high prediction accuracy on the proposed dataset with even a small number of training runs.Research limitations/implicationsBecause of the limited training dataset, the research results may lack generalizability. Therefore, researchers are encouraged to test the proposed propositions further.Originality/valueThe paper publishes a benchmarking dataset for mesh quality learning algorithms and designs a novel neural network approach for unstructured mesh quality evaluation.

Private compression for intermediate feature in IoT-supported mobile cloud inference

In the emerging Internet of Things (IoT) paradigm, mobile cloud inference serves as an efficient application framework that relieves the computation and storage burden on resource-constrained mobile devices by offloading the workload to cloud servers. However, mobile cloud inference encounters computation, communication, and privacy challenges to ensure efficient system inference and protect the privacy of mobile users’ collected information. To address the deployment of deep neural networks (DNN) with large capacity, we propose splitting computing (SC) where the entire model is divided into two parts, to be executed on mobile and cloud ends respectively. However, the transmission of intermediate data poses a bottleneck to system performance. This paper initially demonstrates the privacy issue arising from the machine analysis-oriented intermediate feature. We conduct a preliminary experiment to intuitively reveal the latent potential for enhancing the privacy-preserving ability of the initial feature. Motivated by this, we propose a framework for privacy-preserving intermediate feature compression, which addresses the limitations in both compression and privacy that arise in the original extracted feature data. Specifically, we propose a method that jointly enhances privacy and encoding efficiency, achieved through the collaboration of the encoding feature privacy enhancement module and the privacy feature ordering enhancement module. Additionally, we develop a gradient-reversal optimization strategy based on information theory to ensure the utmost concealment of core privacy information throughout the entire codec process. We evaluate the proposed method on two DNN models using two datasets, demonstrating its ability to achieve superior analysis accuracy and higher privacy preservation than HEVC. Furthermore, we provide an application case of a wireless sensor network to validate the effectiveness of the proposed method in a real-world scenario.

RepDilPCR: a tool for automated analysis of qPCR assays by the dilution-replicate method

BackgroundThe dilution-replicate experimental design for qPCR assays is especially efficient. It is based on multiple linear regression of multiple 3-point standard curves that are derived from the experimental samples themselves and thus obviates the need for a separate standard curve produced by serial dilution of a standard. The method minimizes the total number of reactions and guarantees that Cq values are within the linear dynamic range of the dilution-replicate standard curves. However, the lack of specialized software has so far precluded the widespread use of the dilution-replicate approach.ResultsHere we present repDilPCR, the first tool that utilizes the dilution-replicate method and extends it by adding the possibility to use multiple reference genes. repDilPCR offers extensive statistical and graphical functions that can also be used with preprocessed data (relative expression values) obtained by usual assay designs and evaluation methods. repDilPCR has been designed with the philosophy to automate and speed up data analysis (typically less than a minute from Cq values to publication-ready plots), and features automatic selection and performance of appropriate statistical tests, at least in the case of one-factor experimental designs. Nevertheless, the program also allows users to export intermediate data and perform more sophisticated analyses with external statistical software, e.g. if two-way ANOVA is necessary.ConclusionsrepDilPCR is a user-friendly tool that can contribute to more efficient planning of qPCR experiments and their robust analysis. A public web server is freely accessible at https://repdilpcr.eu without registration. The program can also be used as an R script or as a locally installed Shiny app, which can be downloaded from https://github.com/deyanyosifov/repDilPCR where also the source code is available.

Leakage Model-flexible Deep Learning-based Side-channel Analysis

Profiling side-channel analysis has gained widespread acceptance in both academic and industrial realms due to its robust capacity to unveil protected secrets, even in the presence of countermeasures. To harness this capability, an adversary must access a clone of the target device to acquire profiling measurements, labeling them with leakage models. The challenge of finding an effective leakage model, especially for a protected dataset with a low signal-to-noise ratio or weak correlation between actual leakages and labels, often necessitates an intuitive engineering approach, as otherwise, the attack will not perform well. In this paper, we introduce a deep learning approach with a flexible leakage model, referred to as the multi-bit model. Instead of trying to learn a pre-determined representation of the target intermediate data, we utilize the concept of the stochastic model to decompose the label into bits. Then, the deep learning model is used to classify each bit independently. This versatile multi-bit model can adjust to existing leakage models like the Hamming weight and Most Significant Bit while also possessing the flexibility to adapt to complex leakage scenarios. To further improve the attack efficiency, we extend the multi-bit model to profile all 16 subkey bytes simultaneously, which requires negligible computational effort. The experimental results show that the proposed methods can efficiently break all key bytes across four considered datasets while the conventional leakage models fail. Our work signifies a significant step forward in deep learning-based side-channel attacks, showcasing a high degree of flexibility and efficiency with the proposed leakage model.

Modeling forest canopy structure and developing a stand health index using satellite remote sensing

Biotic and abiotic disturbances modify tree structure and degrade stand health. Accurate geospatial data on stand structure is important for monitoring tree growth, forest health, progression and severity of diseases and pests, and estimating resilience to climate stress. The live crown ratio (LCR) of trees serves as a key health indicator but has been understudied at the landscape level using remote sensing data. This study generated the leaf area index (LAI) and a novel spatial layer of LCR at site and landscape scales using a combination of satellite data and ground observations. We conducted field surveys to collect plot-level (10 m × 10 m) data in four eastern white pine (EWP; Pinus strobus L.)-dominated sites in the state of Maine, USA. The plot-level data were used to develop regression models for LAI and LCR estimation using microwave (Sentinel-1) and optical (Sentinel-2) remote sensing data and applying the Random Forest (RF) and Support Vector Machine (SVM) machine learning algorithms. The RF model showed higher prediction accuracy than the SVM model at the site level. Moreover, the prediction accuracy at the site and landscape levels were comparable for LAI (R2 > 0.76) and LCR (R2 > 0.71) using the RF model. Furthermore, the predicted LAI and LCR were integrated with canopy height and stand density to develop a novel health index map for EWP. The resulting health index map successfully delineated patches representing various health categories. Forestry practitioners and decision-makers can use the derived health index map and intermediate spatial data layers (LAI and LCR) to guide stand management. The developed framework can potentially be applied to other coniferous and broadleaved species for remote sensing-based LCR estimation and forest health assessment upon further studies and verification.

Improved JPEG Lossless Compression for Compression of Intermediate Layers in Neural Networks Based on Compute-In-Memory

With the development of Convolutional Neural Networks (CNNs), there is a growing requirement for their deployment on edge devices. At the same time, Compute-In-Memory (CIM) technology has gained significant attention in edge CNN applications due to its ability to minimize data movement between memory and computing units. However, the deployment of complex deep neural network models on edge devices with restricted hardware resources continues to be challenged by a lack of adequate storage for intermediate layer data. In this article, we propose an optimized JPEG Lossless Compression (JPEG-LS) algorithm that implements serial context parameter updating alongside parallel encoding. This method is designed for the global prediction and efficient compression of intermediate data layers in neural networks employing CIM techniques. The results indicate average compression ratios of 6.44× for VGG16, 3.62× for ResNet34, 1.67× for MobileNetV2, and 2.31× for InceptionV3. Moreover, the implementation achieves a data throughput of 32 bits per cycle at 600 MHz on the TSMC 28 nm, with a hardware cost of 122 K Gate Count.

Influence of short-span scans on trueness in the digital transfer of the maxillomandibular relationship

ObjectivesThis study aimed to assess the influence of the location and distribution of short-span scans, serving as intermediate registration data, on the trueness of digitally transferring the maxillomandibular relationship in full-mouth rehabilitation. MethodsMaxillary and mandibular complete-arch preparation casts mounted on an articulator were scanned, with and without interim restorations, using an intraoral scanner. Four types of short-span scans—right anterior, left anterior, right posterior, and left posterior—were captured from maxillary and mandibular casts. Each scan comprised two prepared teeth and two crowned teeth arranged alternately. These short-span scans served as intermediate data and were cross-registered with complete-arch interim restorations and preparation casts to transfer the maxillomandibular relationship. Based on the number (1 or 2), location (anterior or posterior), and distribution (unilateral or bilateral) of short-span scans, they were categorized into six groups. Trueness was assessed by evaluating the mandibular 3-dimensional (3D), spatial distance, angle, and linear distance deviations between the test and reference scans. ResultsSignificant differences in the trueness of digital cross-mounting were observed among the six groups (P < .05). The group registering the bilateral posterior areas exhibited the smallest 3D, spatial distance, angle, and linear distance deviations, whereas the group registering one anterior area showed the largest deviations. ConclusionThe number, location, and distribution of the registered short-span scans significantly affected the trueness of transferring the maxillomandibular relationship. A posterior short-span scan had higher trueness than an anterior scan. The symmetrical distribution of the two registered larger short-span scans over the cross-arch enhanced the registration trueness. Clinical significanceDigital cross-mounting in full-mouth rehabilitation using short-span scans enhances transfer accuracy, improves patient comfort, and increases chairside efficiency.

Design and Evaluation of Real-Time Data Storage and Signal Processing in a Long-Range Distributed Acoustic Sensing (DAS) Using Cloud-Based Services.

In cloud-based Distributed Acoustic Sensing (DAS) sensor data management, we are confronted with two primary challenges. First, the development of efficient storage mechanisms capable of handling the enormous volume of data generated by these sensors poses a challenge. To solve this issue, we propose a method to address the issue of handling the large amount of data involved in DAS by designing and implementing a pipeline system to efficiently send the big data to DynamoDB in order to fully use the low latency of the DynamoDB data storage system for a benchmark DAS scheme for performing continuous monitoring over a 100 km range at a meter-scale spatial resolution. We employ the DynamoDB functionality of Amazon Web Services (AWS), which allows highly expandable storage capacity with latency of access of a few tens of milliseconds. The different stages of DAS data handling are performed in a pipeline, and the scheme is optimized for high overall throughput with reduced latency suitable for concurrent, real-time event extraction as well as the minimal storage of raw and intermediate data. In addition, the scalability of the DynamoDB-based data storage scheme is evaluated for linear and nonlinear variations of number of batches of access and a wide range of data sample sizes corresponding to sensing ranges of 1-110 km. The results show latencies of 40 ms per batch of access with low standard deviations of a few milliseconds, and latency per sample decreases for increasing the sample size, paving the way toward the development of scalable, cloud-based data storage services integrating additional post-processing for more precise feature extraction. The technique greatly simplifies DAS data handling in key application areas requiring continuous, large-scale measurement schemes. In addition, the processing of raw traces in a long-distance DAS for real-time monitoring requires the careful design of computational resources to guarantee requisite dynamic performance. Now, we will focus on the design of a system for the performance evaluation of cloud computing systems for diverse computations on DAS data. This system is aimed at unveiling valuable insights into performance metrics and operational efficiencies of computations on the data in the cloud, which will provide a deeper understanding of the system's performance, identify potential bottlenecks, and suggest areas for improvement. To achieve this, we employ the CloudSim framework. The analysis reveals that the virtual machine (VM) performance decreases significantly the processing time with more capable VMs, influenced by Processing Elements (PEs) and Million Instructions Per Second (MIPS). The results also reflect that, although a larger number of computations is required as the fiber length increases, with the subsequent increase in processing time, the overall speed of computation is still suitable for continuous real-time monitoring. We also see that VMs with lower performance in terms of processing speed and number of CPUs have more inconsistent processing times compared to those with higher performance, while not incurring significantly higher prices. Additionally, the impact of VM parameters on computation time is explored, highlighting the importance of resource optimization in the DAS system design for efficient performance. The study also observes a notable trend in processing time, showing a significant decrease for every additional 50,000 columns processed as the length of the fiber increases. This finding underscores the efficiency gains achieved with larger computational loads, indicating improved system performance and capacity utilization as the DAS system processes more extensive datasets.

ReSG: A Data Structure for Verification of Majority-based In-memory Computing on ReRAM Crossbars

Recent advancements in the fabrication of Resistive Random Access Memory (ReRAM) devices have led to the development of large-scale crossbar structures. In-memory computing architectures relying on ReRAM crossbars aim to mitigate the processor-memory bottleneck that exists with current complementary metal-oxide semiconductor technology. With this motivation, several synthesis and mapping approaches focusing on the realizations of Boolean functions in the ReRAM crossbars have been proposed earlier. Thus far, the verification of the designs realized on ReRAM crossbars is done either through manual inspection or using simulation-based approaches. Since manual inspections and simulation-based approaches are limited to smaller designs, they cannot be applied to the verification of complex designs on large-scale ReRAM crossbars. Motivated by this, we propose, for the first time, an automatic equivalence checking flow that determines the equivalence between the original function specification (e.g., Majority-inverter Graph ) and the crossbar micro-operations file formats. We consider two crossbar structures, zero-transistor, one-memristor (0T1R) and one-transistor, one-memristor (1T1R) to implement the micro-operations. While the micro-operations file format exists for 0T1R crossbar structures, no representations for micro-operations to be executed in 1T1R crossbars exist yet. In this work, we introduce the micro-operation file format for 1T1R crossbar structures to efficiently represent the micro-operations as ReRAM crossbar netlists. Afterwards, we introduce two intermediate data structures, ReRAM Sequence Graph for 0T1R crossbars (ReSG-0T1R) and for 1T1R crossbars (ReSG-1T1R) , that are derived from the 0T1R and 1T1R crossbar micro-operations file formats, respectively. These ReSGs are then translated into Boolean Satisfiability (SAT) formula, and then the verification is done by checking the generated SAT formulae against the golden functional specification (represented in Verilog) using Z3 Satisfiability solver. Experimental evaluations confirm the effectiveness of the proposed verification methodology on MCNC and ISCAS benchmarks.

Data-driven water quality prediction for wastewater treatment plants

Monitoring and managing wastewater treatment plants (WWTPs) is crucial for environmental protection. The presection of the quality of treated water is essential for energy efficient operation. The current research presents a comprehensive comparison of machine learning models for water quality parameter prediction in WWTPs. Four machine learning models presented in MLP, GFFR, MLP-PCA, and RBF were employed in this study. The primary notion of this study is to apply the proposed models using two distinct modeling scenarios. The first scenario represents a straightforward approach by utilizing the inputs and outputs of WWTPs; meanwhile, the second scenario involves using multi-step modeling techniques, which incorporate intermediate outputs induced by primary and secondary settlers. The study also investigates the potential of the adopted models to handle high dimensional data as a result of the multi-step modeling since more data points and outputs are progressively integrated at each step. The results show that the GFFR model outperforms the other models across both scenarios, specifically in the second scenario in predicting conductivity (COND) by providing higher correlation accuracy (R = 0.893) and lower prediction deviations (NRMSE = 0.091 and NMAE = 0.071). However, all models across both scenarios struggle to predict the other water quality parameters, generating significantly lower prediction correlations and higher prediction deviations. Nonetheless, the innovative multi-step technique in scenario two has significantly boosted the prediction capacity of all models, with improvement ranging from 0.2 % to 157 % and an average of 60 %. The implementation of AI models has proven its ability to accomplish high accuracy for WQ parameter prediction, highlighting the impact of leveraging intermediate process data.

A deep learning-based crystal plasticity finite element model

This study presents an innovative deep learning-based surrogate model for the Crystal Plasticity Finite Element (CPFE) method, fundamentally transforming the generation of mechanical properties such as stress-strain curves in the study of crystal plasticity. Stress-strain curves are pivotal in understanding material deformation, elucidating the intricate relationship between a material's structure and its properties. Traditional CPFE methods, though thorough in their analysis, face significant computational challenges, largely due to the complexity of the crystal plasticity framework. The proposed model circumvents this bottleneck by utilizing an autoencoder architecture to learn intermediate data representations, which are then used to predict the plastic component of deformation. This predicted plastic component serves as a foundation for computing stress-strain curves, effectively bypassing the most time-intensive aspect of traditional CPFE methods, the plasticity self-consistency procedure (achieving a 29.3x speed increase without compromising accuracy).

The Long Way to Deforestation: A Type Inference and Elaboration Technique for Removing Intermediate Data Structures

The Long Way to Deforestation: A Type Inference and Elaboration Technique for Removing Intermediate Data Structures

Efficient and Secure Management of Medical Data Sharing Based on Blockchain Technology

In the current landscape of medical data management, processing data across diverse institutions and maximizing their value are paramount. However, traditional methods lack a secure and efficient mechanism for end-to-end traceability and supervision, posing challenges in distributed scenarios lacking mutual trust. Leveraging blockchain’s decentralized, tamper-proof, and traceable features, this paper introduces a blockchain-based medical data management platform. This platform enables full-process management of raw data, operational behaviors, intermediate data, and final data, meeting the needs of trusted storage and supervision of data. We propose two methods, namely, naive method and DAG-based method, to realize forward tracking and backward tracing of medical data stored on the blockchain, respectively. We validated and analyzed the storage and query performance of the medical data management platform on real medical data, and we also conducted experimental analyses on the efficiency of the proposed traceability algorithm under different data scales and processing path lengths. The results demonstrate that our platform and traceability methods effectively meet the management needs of medical data distributed across institutions.

An interpretable ensemble trees method with joint analysis of static and dynamic features for myocardial infarction detection

Objective. In recent years, artificial intelligence-based electrocardiogram (ECG) methods have been massively applied to myocardial infarction (MI). However, the joint analysis of static and dynamic features to achieve accurate and interpretable MI detection has not been comprehensively addressed. Approach. This paper proposes a simplified ensemble tree method with a joint analysis of static and dynamic features to solve this issue for MI detection. Initially, the dynamic features are extracted by modeling the intrinsic dynamics of ECG via dynamic learning in addition to extracting classical static features. Secondly, a two-stage feature selection strategy is designed to identify a few significant features, which substitute the original variables that are employed in constructing the ensemble tree. This approach enhances the discriminative ability by selecting significant static and dynamic features. Subsequently, this paper presents an interpretable classification method named StackTree by introducing a stacked ensemble scheme to modify the ensemble tree simplification algorithm. The representative rules of the raw ensemble trees are selected as the intermediate training data that is used to retrain a decision tree with performance close to that of the source ensemble model. Using this scheme, the significant precision and interpretability of MI detection are thus comprehensively addressed. Main results. The effectiveness of our method in detecting MI is evaluated using the Physikalisch-Technische Bundesanstalt (PTB) and clinical database. The findings suggest that our algorithm outperforms the traditional methods based on a single type of feature. Additionally, it is comparable to the conventional random forest, achieving 97.1% accuracy under the inter-patient framework on the PTB database. Furthermore, feature subsets trained on PTB are validated using the clinical database, resulting in an accuracy of 84.5%. The chosen important features demonstrate that both static and dynamic information have crucial roles in MI detection. Crucially, the proposed method provides clear internal workings in an easy-to-understand visual manner.

CONSTRUCTION OF PROBABILISTIC CAUSAL RELATIONSHIPS BETWEEN EQUIVALENCE CLASSES OF DATA IN AN INTELLIGENT INFORMATION SYSTEM

The subject of this research is the processes involved in generating explanations for decision-making in artificial intelligence systems. Explanations in such systems enable the decision-making process to be transparent and comprehensible for the user, thereby increasing user trust in the obtained results. The aim of this work is to develop an approach for constructing a probabilistic causal explanation model that takes into account the equivalence classes of input, intermediate, and resulting data. Solving this problem creates conditions for building explanations in the form of causal relationships based on the available information about the properties of input data as well as the properties of the results obtained in the artificial intelligence system. To achieve this aim, the following tasks are addressed: developing a causal dependency model between the equivalence classes of input and output data; developing methods for constructing equivalence classes of data in the decision-making process and a method for constructing causal explanations. A probabilistic model of causal dependency is proposed, which includes a causal relationship between the equivalence classes of input or intermediate and resulting data obtained during the decision-making process in the artificial intelligence system. This relationship considers the estimates of the possibility and necessity of such a dependency. The model creates conditions for explaining the possible causes of the obtained decision. A set of methods for constructing equivalence classes of data in the decision-making process and for constructing causal explanations is proposed, establishing a causal relationship between the equivalence classes. When constructing equivalence classes, relations of mandatory and optional data refinement, requirements or exclusions of data, as well as data conjunctions, are established. When constructing causal explanations, the possibility and limitations of the necessity of such a dependency are calculated, allowing explanations to be built based on the available information about the obtained decisions and the input and intermediate data used to form these decisions.

Cclib 2.0: An updated architecture for interoperable computational chemistry.

Interoperability in computational chemistry is elusive, impeded by the independent development of software packages and idiosyncratic nature of their output files. The cclib library was introduced in 2006 as an attempt to improve this situation by providing a consistent interface to the results of various quantum chemistry programs. The shared API across programs enabled by cclib has allowed users to focus on results as opposed to output and to combine data from multiple programs or develop generic downstream tools. Initial development, however, did not anticipate the rapid progress of computational capabilities, novel methods, and new programs; nor did it foresee the growing need for customizability. Here, we recount this history and present cclib 2, focused on extensibility and modularity. We also introduce recent design pivots-the formalization of cclib's intermediate data representation as a tree-based structure, a new combinator-based parser organization, and parsed chemical properties as extensible objects.

Genome-wide analysis and visualization of copy number with CNVpytor in igv.js.

Copy number variation (CNV) and alteration (CNA) analysis is a crucial component in many genomic studies and its applications span from basic research to clinic diagnostics and personalized medicine. CNVpytor is a tool featuring a read depth-based caller and combined read depth and B-allele frequency (BAF) based 2D caller to find CNVs and CNAs. The tool stores processed intermediate data and CNV/CNA calls in a compact HDF5 file-pytor file. Here, we describe a new track in igv.js that utilizes pytor and whole genome variant files as input for on-the-fly read depth and BAF visualization, CNV/CNA calling and analysis. Embedding into HTML pages and Jupiter Notebooks enables convenient remote data access and visualization simplifying interpretation and analysis of omics data. The CNVpytor track is integrated with igv.js and available at https://github.com/igvteam/igv.js. The documentation is available at https://github.com/igvteam/igv.js/wiki/cnvpytor. Usage can be tested in the IGV-Web app at https://igv.org/app and also on https://github.com/abyzovlab/CNVpytor.

The Early Data Release of the Dark Energy Spectroscopic Instrument

The Dark Energy Spectroscopic Instrument (DESI) completed its 5 month Survey Validation in 2021 May. Spectra of stellar and extragalactic targets from Survey Validation constitute the first major data sample from the DESI survey. This paper describes the public release of those spectra, the catalogs of derived properties, and the intermediate data products. In total, the public release includes good-quality spectral information from 466,447 objects targeted as part of the Milky Way Survey, 428,758 as part of the Bright Galaxy Survey, 227,318 as part of the Luminous Red Galaxy sample, 437,664 as part of the Emission Line Galaxy sample, and 76,079 as part of the Quasar sample. In addition, the release includes spectral information from 137,148 objects that expand the scope beyond the primary samples as part of a series of secondary programs. Here, we describe the spectral data, data quality, data products, Large-Scale Structure science catalogs, access to the data, and references that provide relevant background to using these spectra.

FPGA-based tiling scheme and on-chip memory scheduling scheme for multi-branch semantic segmentation neural network accelerator

Abstract According to the characteristics of a multi-branch semantic segmentation neural network with a large amount of intermediate data during the inference process, we reduce the requirement of off-chip memory access and the redundant calculation due to tiling through better blocking scheme, better on-chip memory partitioning scheme, and on-chip memory scheduling scheme, improving the energy efficiency to 77.4 GOPS/W.

Fusion of multi-temporal PlanetScope data and very high-resolution aerial imagery for urban tree species mapping

Detailed assessment of the ecosystem services provided by urban green spaces requires data on urban tree species. While many approaches for mapping of urban trees from remotely sensed data have been proposed, the fusion of multi-temporal satellite imagery with very high resolution orthophotos remains relatively underexplored. In this research, we assess the potential of a multimodal deep learning approach with intermediate data fusion for classifying common tree species found in the Brussels Capital Region. Our method combines two image sources: (a) multi-temporal PlanetScope data and (b) high-resolution aerial imagery. To evaluate the contribution of each image source, we separately train and assess each branch of the network. Both image sources demonstrate the ability to predict prevalent tree species with high accuracy. However, the fusion of the two image sources yields the best results, achieving an overall accuracy of 0.88 for the five most common tree species in the region. Our approach is compared to two conventional machine learning methods: a random forest (RF) and a support vector machine classifier (SVM) and outperforms both with a 11 percentage point increase in overall accuracy over RF and a 10 percentage point increase over SVM. Increasing the number of species from five to thirteen, including all species with more than 500 tree samples, results in a marginal decrease in accuracy (from 0.88 to 0.84). Overall, our deep learning approach demonstrates its efficacy in classifying common tree species in urban settings and provides a foundation for a comprehensive quantification of ecosystem services offered by urban trees through remote sensing data.