arXiv subjects

Marco Aldinucci

Publications and source records attributed to Marco Aldinucci.

At least 19 recordsLinked to original sources

arXiv2025

Building an Accelerated OpenFOAM Proof-of-Concept Application using Modern C++

The modern trend in High-Performance Computing (HPC) involves the use of accelerators such as Graphics Processing Units (GPUs) alongside Central Processing Units (CPUs) to speed up numerical operations in various applications. Leading manufacturers such as NVIDIA, Intel, and AMD are constantly advancing these architectures, augmenting them with features such as mixed precision, enhanced memory hierarchies, and specialised accelerator silicon blocks (e.g., Tensor Cores on GPU or AMX/SME engines on CPU) to enhance compute performance. At the same time, significant efforts in software development are aimed at optimizing the use of these innovations, seeking to improve usability and accessibility. This work contributes to the state-of-the-art of OpenFOAM development by presenting a working Proof-Of-Concept application built using modern ISO C++ parallel constructs. This approach, combined with an appropriate compiler runtime stack, like the one provided by the NVIDIA HPC SDK, makes it possible to accelerate well-defined kernels, allowing multi-core execution and GPU offloading using a single codebase. The study demonstrates that it is possible to increase the performance of the OpenFOAM laplacianFoam application by offloading the computations on NVIDIA GPUs using the C++ parallel construct.

cs.MS↗

arXiv2025

Decentralized Time Series Classification with ROCKET Features

Time series classification (TSC) is a critical task with applications in various domains, including healthcare, finance, and industrial monitoring. Due to privacy concerns and data regulations, Federated Learning has emerged as a promising approach for learning from distributed time series data without centralizing raw information. However, most FL solutions rely on a client-server architecture, which introduces robustness and confidentiality risks related to the distinguished role of the server, which is a single point of failure and can observe knowledge extracted from clients. To address these challenges, we propose DROCKS, a fully decentralized FL framework for TSC that leverages ROCKET (RandOm Convolutional KErnel Transform) features. In DROCKS, the global model is trained by sequentially traversing a structured path across federation nodes, where each node refines the model and selects the most effective local kernels before passing them to the successor. Extensive experiments on the UCR archive demonstrate that DROCKS outperforms state-of-the-art client-server FL approaches while being more resilient to node failures and malicious attacks. Our code is available at https://anonymous.4open.science/r/DROCKS-7FF3/README.md.

cs.LG↗

arXiv2025

Exploring energy consumption of AI frameworks on a 64-core RV64 Server CPU

In today's era of rapid technological advancement, artificial intelligence (AI) applications require large-scale, high-performance, and data-intensive computations, leading to significant energy demands. Addressing this challenge necessitates a combined approach involving both hardware and software innovations. Hardware manufacturers are developing new, efficient, and specialized solutions, with the RISC-V architecture emerging as a prominent player due to its open, extensible, and energy-efficient instruction set architecture (ISA). Simultaneously, software developers are creating new algorithms and frameworks, yet their energy efficiency often remains unclear. In this study, we conduct a comprehensive benchmark analysis of machine learning (ML) applications on the 64-core SOPHON SG2042 RISC-V architecture. We specifically analyze the energy consumption of deep learning inference models across three leading AI frameworks: PyTorch, ONNX Runtime, and TensorFlow. Our findings show that frameworks using the XNNPACK back-end, such as ONNX Runtime and TensorFlow, consume less energy compared to PyTorch, which is compiled with the native OpenBLAS back-end.

cs.DC↗

arXiv2025

The Gaia AVU-GSR solver: a CPU + GPU parallel code toward Exascale systems

The solver module of the Astrometric Verification Unit - Global Sphere Reconstruction (AVU-GSR) pipeline aims to find the astrometric parameters of $\sim$$10^8$ stars in the Milky Way, besides the attitude and instrumental settings of the Gaia satellite and the parametrized post Newtonian parameter $γ$ with a resolution of 10-100 micro-arcseconds. To perform this task, the code solves a system of linear equations with the iterative Least Squares (LSQR) algorithm, where the coefficient matrix is large (10-50 TB) and sparse and the iterations stop when convergence is reached in the least squares sense. The two matrix-by-vector products performed at each LSQR step were GPU-ported, firstly with OpenACC and then with CUDA, resulting in a $\sim$$1.5$x and $\sim$$14$x speedup, respectively, over an original code version entirely parallelized on the CPU with MPI + OpenMP. The CUDA code was further optimized and then ported with programming frameworks portable across different GPU architectures, obtaining a further $\sim$$2$x acceleration factor. One critical section of the code consists in the computation of covariances, whose total number is $N_{\rm unk} \times (N_{\rm unk} - 1)/2$ and occupy $\sim$1 EB, being $N_{\rm unk}$$\sim$$5 \times 10^8$ the total number of unknowns. This "Big Data" issue cannot be faced with standard approaches: we defined an I/O-based pipeline made of two concurrently launched jobs, where one job, i.e., the LSQR, writes the files and the second job reads them, iteratively computes the covariances and deletes them. The pipeline does not present significant bottlenecks until a number of covariances elements equal to $\sim$$8 \times 10^6$. The code currently runs in production on Leonardo CINECA infrastructure.

astro-ph.IM↗

arXiv2024

Stream parallel skeleton optimization

We discuss the properties of the composition of stream parallel skeletons such as pipelines and farms. By looking at the ideal performance figures assumed to hold for these skeletons, we show that any stream parallel skeleton composition can always be rewritten into an equivalent "normal form" skeleton composition, delivering a service time which is equal or even better to the service time of the original skeleton composition, and achieving a better utilization of the processors used. The normal form is defined as a single farm built around a sequential worker code. Experimental results are discussed that validate this normal form.

cs.DC↗

arXiv2024

A Fault Tolerance Mechanism for Hybrid Scientific Workflows

In large distributed systems, failures are a daily event occurring frequently, especially with growing numbers of computation tasks and locations on which they are deployed. The advantage of representing an application with a workflow is the possibility of exploiting Workflow Management System (WMS) features such as portability. A relevant feature that some WMSs supply is reliability. Over recent years, the emergence of hybrid workflows has posed new and intriguing challenges by increasing the possibility of distributing computations involving heterogeneous and independent environments. Consequently, the number of possible points of failure in the execution increased, creating different important challenges that are interesting to study. This paper presents the implementation of a fault tolerance mechanism for hybrid workflows based on the recovery and rollback approach. A representation of the hybrid workflows with the formal framework is provided, together with the experiments demonstrating the functionality of implementing approach.

cs.DC↗

arXiv2024

Introducing SWIRL: An Intermediate Representation Language for Scientific Workflows

In the ever-evolving landscape of scientific computing, properly supporting the modularity and complexity of modern scientific applications requires new approaches to workflow execution, like seamless interoperability between different workflow systems, distributed-by-design workflow models, and automatic optimisation of data movements. In order to address this need, this article introduces SWIRL, an intermediate representation language for scientific workflows. In contrast with other product-agnostic workflow languages, SWIRL is not designed for human interaction but to serve as a low-level compilation target for distributed workflow execution plans. The main advantages of SWIRL semantics are low-level primitives based on the send/receive programming model and a formal framework ensuring the consistency of the semantics and the specification of translating workflow models represented by Directed Acyclic Graphs (DAGs) into SWIRL workflow descriptions. Additionally, SWIRL offers rewriting rules designed to optimise execution traces, accompanied by corresponding equivalence. An open-source SWIRL compiler toolchain has been developed using the ANTLR Python3 bindings.

cs.DC↗

arXiv2024

DALLMi: Domain Adaption for LLM-based Multi-label Classifier

Large language models (LLMs) increasingly serve as the backbone for classifying text associated with distinct domains and simultaneously several labels (classes). When encountering domain shifts, e.g., classifier of movie reviews from IMDb to Rotten Tomatoes, adapting such an LLM-based multi-label classifier is challenging due to incomplete label sets at the target domain and daunting training overhead. The existing domain adaptation methods address either image multi-label classifiers or text binary classifiers. In this paper, we design DALLMi, Domain Adaptation Large Language Model interpolator, a first-of-its-kind semi-supervised domain adaptation method for text data models based on LLMs, specifically BERT. The core of DALLMi is the novel variation loss and MixUp regularization, which jointly leverage the limited positively labeled and large quantity of unlabeled text and, importantly, their interpolation from the BERT word embeddings. DALLMi also introduces a label-balanced sampling strategy to overcome the imbalance between labeled and unlabeled data. We evaluate DALLMi against the partial-supervised and unsupervised approach on three datasets under different scenarios of label availability for the target domain. Our results show that DALLMi achieves higher mAP than unsupervised and partially-supervised approaches by 19.9% and 52.2%, respectively.

cs.CL↗

arXiv2023

Model-Agnostic Federated Learning

Since its debut in 2016, Federated Learning (FL) has been tied to the inner workings of Deep Neural Networks (DNNs). On the one hand, this allowed its development and widespread use as DNNs proliferated. On the other hand, it neglected all those scenarios in which using DNNs is not possible or advantageous. The fact that most current FL frameworks only allow training DNNs reinforces this problem. To address the lack of FL solutions for non-DNN-based use cases, we propose MAFL (Model-Agnostic Federated Learning). MAFL marries a model-agnostic FL algorithm, AdaBoost.F, with an open industry-grade FL framework: Intel OpenFL. MAFL is the first FL system not tied to any specific type of machine learning model, allowing exploration of FL scenarios beyond DNNs and trees. We test MAFL from multiple points of view, assessing its correctness, flexibility and scaling properties up to 64 nodes. We optimised the base software achieving a 5.5x speedup on a standard FL scenario. MAFL is compatible with x86-64, ARM-v8, Power and RISC-V.

cs.LG↗

arXiv2023

A Federated Learning Benchmark for Drug-Target Interaction

Aggregating pharmaceutical data in the drug-target interaction (DTI) domain has the potential to deliver life-saving breakthroughs. It is, however, notoriously difficult due to regulatory constraints and commercial interests. This work proposes the application of federated learning, which we argue to be reconcilable with the industry's constraints, as it does not require sharing of any information that would reveal the entities' data or any other high-level summary of it. When used on a representative GraphDTA model and the KIBA dataset it achieves up to 15% improved performance relative to the best available non-privacy preserving alternative. Our extensive battery of experiments shows that, unlike in other domains, the non-IID data distribution in the DTI datasets does not deteriorate FL performance. Additionally, we identify a material trade-off between the benefits of adding new data, and the cost of adding more clients.

cs.LG↗

arXiv2023

Experimenting with Emerging RISC-V Systems for Decentralised Machine Learning

Decentralised Machine Learning (DML) enables collaborative machine learning without centralised input data. Federated Learning (FL) and Edge Inference are examples of DML. While tools for DML (especially FL) are starting to flourish, many are not flexible and portable enough to experiment with novel processors (e.g., RISC-V), non-fully connected network topologies, and asynchronous collaboration schemes. We overcome these limitations via a domain-specific language allowing us to map DML schemes to an underlying middleware, i.e. the FastFlow parallel programming library. We experiment with it by generating different working DML schemes on x86-64 and ARM platforms and an emerging RISC-V one. We characterise the performance and energy efficiency of the presented schemes and systems. As a byproduct, we introduce a RISC-V porting of the PyTorch framework, the first publicly available to our knowledge.

cs.DC↗

arXiv2023

The MPI + CUDA Gaia AVU-GSR Parallel Solver Toward Next-generation Exascale Infrastructures

We ported to the GPU with CUDA the Astrometric Verification Unit-Global Sphere Reconstruction (AVU-GSR) Parallel Solver developed for the ESA Gaia mission, by optimizing a previous OpenACC porting of this application. The code aims to find, with a [10,100]$μ$as precision, the astrometric parameters of $\sim$$10^8$ stars, the attitude and instrumental settings of the Gaia satellite, and the global parameter $γ$ of the parametrized Post-Newtonian formalism, by solving a system of linear equations, $A\times x=b$, with the LSQR iterative algorithm. The coefficient matrix $A$ of the final Gaia dataset is large, with $\sim$$10^{11} \times 10^8$ elements, and sparse, reaching a size of $\sim$10-100 TB, typical for the Big Data analysis, which requires an efficient parallelization to obtain scientific results in reasonable timescales. The speedup of the CUDA code over the original AVU-GSR solver, parallelized on the CPU with MPI+OpenMP, increases with the system size and the number of resources, reaching a maximum of $\sim$14x, >9x over the OpenACC application. This result is obtained by comparing the two codes on the CINECA cluster Marconi100, with 4 V100 GPUs per node. After verifying the agreement between the solutions of a set of systems with different sizes computed with the CUDA and the OpenMP codes and that the solutions showed the required precision, the CUDA code was put in production on Marconi100, essential for an optimal AVU-GSR pipeline and the successive Gaia Data Releases. This analysis represents a first step to understand the (pre-)Exascale behavior of a class of applications that follow the same structure of this code. In the next months, we plan to run this code on the pre-Exascale platform Leonardo of CINECA, with 4 next-generation A200 GPUs per node, toward a porting on this infrastructure, where we expect to obtain even higher performances.

astro-ph.IM↗

arXiv2023

Benchmarking FedAvg and FedCurv for Image Classification Tasks

Classic Machine Learning techniques require training on data available in a single data lake. However, aggregating data from different owners is not always convenient for different reasons, including security, privacy and secrecy. Data carry a value that might vanish when shared with others; the ability to avoid sharing the data enables industrial applications where security and privacy are of paramount importance, making it possible to train global models by implementing only local policies which can be run independently and even on air-gapped data centres. Federated Learning (FL) is a distributed machine learning approach which has emerged as an effective way to address privacy concerns by only sharing local AI models while keeping the data decentralized. Two critical challenges of Federated Learning are managing the heterogeneous systems in the same federated network and dealing with real data, which are often not independently and identically distributed (non-IID) among the clients. In this paper, we focus on the second problem, i.e., the problem of statistical heterogeneity of the data in the same federated network. In this setting, local models might be strayed far from the local optimum of the complete dataset, thus possibly hindering the convergence of the federated model. Several Federated Learning algorithms, such as FedAvg, FedProx and Federated Curvature (FedCurv), aiming at tackling the non-IID setting, have already been proposed. This work provides an empirical assessment of the behaviour of FedAvg and FedCurv in common non-IID scenarios. Results show that the number of epochs per round is an important hyper-parameter that, when tuned appropriately, can lead to significant performance gains while reducing the communication cost. As a side product of this work, we release the non-IID version of the datasets we used so to facilitate further comparisons from the FL community.

cs.LG↗

arXiv2023

Experimenting with Normalization Layers in Federated Learning on non-IID scenarios

Training Deep Learning (DL) models require large, high-quality datasets, often assembled with data from different institutions. Federated Learning (FL) has been emerging as a method for privacy-preserving pooling of datasets employing collaborative training from different institutions by iteratively globally aggregating locally trained models. One critical performance challenge of FL is operating on datasets not independently and identically distributed (non-IID) among the federation participants. Even though this fragility cannot be eliminated, it can be debunked by a suitable optimization of two hyper-parameters: layer normalization methods and collaboration frequency selection. In this work, we benchmark five different normalization layers for training Neural Networks (NNs), two families of non-IID data skew, and two datasets. Results show that Batch Normalization, widely employed for centralized DL, is not the best choice for FL, whereas Group and Layer Normalization consistently outperform Batch Normalization. Similarly, frequent model aggregation decreases convergence speed and mode quality.

cs.LG↗

arXiv2023

FedER: Federated Learning through Experience Replay and Privacy-Preserving Data Synthesis

In the medical field, multi-center collaborations are often sought to yield more generalizable findings by leveraging the heterogeneity of patient and clinical data. However, recent privacy regulations hinder the possibility to share data, and consequently, to come up with machine learning-based solutions that support diagnosis and prognosis. Federated learning (FL) aims at sidestepping this limitation by bringing AI-based solutions to data owners and only sharing local AI models, or parts thereof, that need then to be aggregated. However, most of the existing federated learning solutions are still at their infancy and show several shortcomings, from the lack of a reliable and effective aggregation scheme able to retain the knowledge learned locally to weak privacy preservation as real data may be reconstructed from model updates. Furthermore, the majority of these approaches, especially those dealing with medical data, relies on a centralized distributed learning strategy that poses robustness, scalability and trust issues. In this paper we present a federated and decentralized learning strategy, FedER, that, exploiting experience replay and generative adversarial concepts, effectively integrates features from local nodes, providing models able to generalize across multiple datasets while maintaining privacy. FedER is tested on two tasks -- tuberculosis and melanoma classification -- using multiple datasets in order to simulate realistic non-i.i.d. medical data scenarios. Results show that our approach achieves performance comparable to standard (non-federated) learning and significantly outperforms state-of-the-art federated methods in their centralized (thus, more favourable) formulation. Code is available at https://github.com/perceivelab/FedER

cs.LG↗

arXiv2022

The Gaia AVU-GSR parallel solver: preliminary studies of a LSQR-based application in perspective of exascale systems

The Gaia Astrometric Verification Unit-Global Sphere Reconstruction (AVU-GSR) Parallel Solver aims to find the astrometric parameters for $\sim$10$^8$ stars in the Milky Way, the attitude and the instrumental specifications of the Gaia satellite, and the global parameter $γ$ of the post Newtonian formalism. The code iteratively solves a system of linear equations, $\mathbf{A} \times \vec{x} = \vec{b}$, where the coefficient matrix $\mathbf{A}$ is large ($\sim$$10^{11} \times 10^8$ elements) and sparse. To solve this system of equations, the code exploits a hybrid implementation of the iterative PC-LSQR algorithm, where the computation related to different horizontal portions of the coefficient matrix is assigned to separate MPI processes. In the original code, each matrix portion is further parallelized over the OpenMP threads. To further improve the code performance, we ported the application to the GPU, replacing the OpenMP parallelization language with OpenACC. In this port, $\sim$95% of the data is copied from the host to the device at the beginning of the entire cycle of iterations, making the code $compute$ $bound$ rather than $data$$-$$transfer$ $bound$. The OpenACC code presents a speedup of $\sim$1.5 over the OpenMP version but further optimizations are in progress to obtain higher gains. The code runs on multiple GPUs and it was tested on the CINECA supercomputer Marconi100, in anticipation of a port to the pre-exascale system Leonardo, that will be installed at CINECA in 2022.

astro-ph.IM↗

arXiv2022

Modelling Biological and Ecological Systems with the Calculus of Wrapped Compartments

The modelling and analysis of biological systems has deep roots in Mathematics, specifically in the field of Ordinary Differential Equations. Alternative approaches based on formal calculi, often derived from process algebras or term rewriting systems, provide a quite complementary way to analyse the behaviour of biological systems. These calculi allow to cope in a natural way with notions like compartments and membranes, which are not easy (sometimes impossible) to handle with purely numerical approaches, and are often based on stochastic simulation methods. The Calculus of Wrapped Compartments is a framework based on stochastic multiset rewriting in a compartmentalised setting used for the description of biological and ecological systems. We provide an extended presentation of the Calculus of Wrapped Compartments, sketch a few modelling guidelines to encode biological and ecological interactions, show how spatial properties can be handled within the framework and define a hybrid simulation algorithm. Several applications in Biology and Ecology are proposed as modelling case studies.

q-bio.QM↗

arXiv2022

Neural Transformers for Intraductal Papillary Mucosal Neoplasms (IPMN) Classification in MRI images

Early detection of precancerous cysts or neoplasms, i.e., Intraductal Papillary Mucosal Neoplasms (IPMN), in pancreas is a challenging and complex task, and it may lead to a more favourable outcome. Once detected, grading IPMNs accurately is also necessary, since low-risk IPMNs can be under surveillance program, while high-risk IPMNs have to be surgically resected before they turn into cancer. Current standards (Fukuoka and others) for IPMN classification show significant intra- and inter-operator variability, beside being error-prone, making a proper diagnosis unreliable. The established progress in artificial intelligence, through the deep learning paradigm, may provide a key tool for an effective support to medical decision for pancreatic cancer. In this work, we follow this trend, by proposing a novel AI-based IPMN classifier that leverages the recent success of transformer networks in generalizing across a wide variety of tasks, including vision ones. We specifically show that our transformer-based model exploits pre-training better than standard convolutional neural networks, thus supporting the sought architectural universalism of transformers in vision, including the medical image domain and it allows for a better interpretation of the obtained results.

cs.CV↗