Valeria Mele

Deep learning (DL) has significantly advanced Industry 4.0 by leveraging data from the Industrial Internet of Things (IIoT) to enable smart manufacturing, predictive maintenance, and data-driven product marketing. However, multimodal industrial data presents challenges for traditional frameworks, including scalability, data privacy, and integration efficiency. This article introduces an efficient product retrieval framework for e-commerce systems, addressing privacy and performance challenges through federated learning (FL). Specifically, we propose FL for augmented industrial retrieval (FLAIR), a novel part retrieval system where distributed warehouses collaboratively train a multimodal foundation model, contrastive language-image pretraining (CLIP), by fine-tuning only the Adapter module via FL, ensuring data privacy and efficiency. To address the limited availability of multimodal industrial data, our framework incorporates effective data augmentation strategies to enhance the diversity and quality of the training dataset. Comprehensive experiments on the industrial language-image dataset (ILID) highlight that FLAIR holds effective privacy safeguards and strong retrieval capabilities. Additionally, an advanced e-commerce recommendation system built on FLAIR showcases its practical effectiveness. FLAIR represents the first application of FL for industrial product retrieval, optimizing part searches, inventory management, and customer experience while maintaining data security. The complete code is available at https://github.com/MODAL-UNINA/FLAIR.

Advancements in computing platform deployment have acted as both push and pull elements for the advancement of engineering design and scientific knowledge. Historically, improvements in computing platforms were mostly dependent on simultaneous developments in hardware, software, architecture, and algorithms (a process known as co-design), which raised the performance of computational models. But, there are many obstacles to using the Exascale Computing Era sophisticated computing platforms effectively. These include but are not limited to massive parallelism, effective exploitation, and high complexity in programming, such as heterogeneous computing facilities. So, now is the time to create new algorithms that are more resilient, energy-aware, and able to address the demands of increasing data locality and achieve much higher concurrency through high levels of scalability and granularity. In this context, some methods, such as those based on hierarchical matrices (HMs), have been declared among the most promising in the use of new computing resources precisely because of their strongly hierarchical nature. This work aims to start to assess the advantages, and limits, of the use of HMs in operations such as the evaluation of matrix polynomials, which are crucial, for example, in a Graph Convolutional Deep Neural Network (GC-DNN) context. A case study from the GCNN context provides some insights into the effectiveness, in terms of accuracy, of the employment of HMs.

Accidental fires present a significant and growing threat to ecosystems, communities, and economies worldwide. This paper presents the development and implementation of Smoketracer, a novel on-demand application empowered by advanced computational workflows for fast response to accidental fire and effects mitigation. The DAGonStar workflow engine orchestrates the application leveraging a brand-new ad-hoc file system, turning the workflow:// schema used to define the tasks' data dependencies into an actual user-space distributed in-RAM storage. Although the evaluation results of the DAGon File System (DAGonFS) are still preliminary, the performance of file copy operations is from 31% up to 63% faster than the baseline (shared scratch directory hosted on mechanical hard drives). The accidental fires application wall clock is reduced by 57% when using the DAGonFS on four data nodes.

We herein briefly describe a multilevel approach to analyze parallel algorithms performances. The main outcome of such an approach is that the algorithm is described using a set of operators related to each other according to the problem decomposition. A set of block matrices (called decomposition and execution matrices) highlights fundamental characteristics of the algorithm, such as inherent parallelism and sources of overheads, and all the involved factors and their relationships, in a general but modular, flexible and adaptive fashion. The work aims to show how we can rewrite the well-known Ware-Amdhal's Law: in a previous work we already gave an expression for the Amdhal's law in our framework, but, even with the same meaning, it wasn't immediately comparable with the original one. Here we focus on that law and show how it comes exactly from our parameters. Moreover, we extend the law with a more general expression, that we call Generalized Amdhal's Law: the classical law will come as a particular case of the generalized one.

Summary: This special issue on the Pervasive Nature of HPC (PN‐HPC) collects an extension of the most valuable works presented at the sixth Workshop on Models, Algorithms and Methodologies for Hybrid Parallelism in New HPC Systems (MAMHYP‐22), held in Gdansk (Poland) in September 2022, jointly with the 14th conference on Parallel Processing and Applied Mathematics (PPAM‐22). New original papers related to the workshop themes are also included. The final aim is to provide a glimpse of the current state of knowledge related to the development of efficient methodologies and algorithms for HPC systems with multiple forms of parallelism.

Summary: This work describes some first efforts to design a new Linpack‐like benchmark useful to evaluate the performance of Heterogeneous Computing Resources. The benchmark is based on the Schur Complement reformulation of the solution of a linear equation system. Details about its implementation and evaluation, mainly in terms of performance scalability, are presented for a computing environment based on multi NVIDIA GP‐GPUs nodes connected by an Infiniband network.