In federated learning for multimodal models, balancing global generalization and local personalization remains challenging due to heterogeneous client data distributions. To address this, we propose a novel federated learning framework employing dual Low-Rank Adaptation (LoRA) adapters within a frozen Contrastive Language-Image Pre-training (CLIP) backbone. Each client maintains both global and local adapters, dynamically orchestrated via a lightweight gating network that adaptively fuses the adapters based on input-specific features. Unlike traditional parameter averaging, our server aggregates client knowledge through federated distillation on a compact reference dataset, effectively mitigating parameter conflicts across clients. Experiments demonstrate that our adaptive fusion strategy significantly improves personalized representation quality, outperforming standard LoRA-based federated approaches, especially in scenarios with diverse local data distributions.

Motivated by high resource costs of centralized machine learning schemes as well as data privacy concerns, federated learning (FL) emerged as an efficient alternative that relies on aggregating locally trained models rather than collecting clients' potentially private data. In practice, available resources and data distributions vary from one client to another, creating an inherent system heterogeneity that leads to deterioration of the performance of conventional FL algorithms. In this work, we present a federated quantization-based self-supervised learning scheme (Fed-QSSL) designed to address heterogeneity in FL systems. At clients' side, to tackle data heterogeneity we leverage distributed self-supervised learning while utilizing low-bit quantization to satisfy constraints imposed by local infrastructure and limited communication resources. At server's side, Fed-QSSL deploys de-quantization, weighted aggregation and re-quantization, ultimately creating models personalized to both data distribution as well as specific infrastructure of each client's device. We validated the proposed algorithm on real world datasets, demonstrating its efficacy, and theoretically analyzed impact of low-bit training on the convergence and robustness of the learned models.

Federated learning (FL) is widely recognized as a valuable approach for Connected and Automated Vehicles (CAVs) because it facilitates collaborative model development across a multitude of vehicles in a decentralized manner. However, numerous studies on FL algorithms only assessed their performance through experiments conducted in simulated client-server configurations (e.g., where both server and clients run on the same machine) or simplified scenarios that do not account for client downtime. In this paper, we aim to conduct more realistic evaluations for CAV applications leveraging FL. We present a preliminary experimental study as well as offer insights into potential future directions.

Big data disrupts everything it touches, but automotive is probably one of the top industries that enjoy and leverage the benefits. The Automotive Big Data Pipeline (ABDP) is a Big Data pipeline base on the automotive use case and is required to scale up agile and high performance in real-time or in batch. Nonetheless, there're many alternative infrastructure designs but lack of knowledge, which fits the best for the automotive domain. It leads this paper into a question: What kinds of infrastructure design could provide better performance for the ABDP?In this paper, we introduce two well-known infrastructure designs called Hyper-Converged infrastructure (HCI) and Disaggregated Hyper-Converged infrastructure (DHCI). HCI combines standard data center hardware using locally attached storage resources to create fast, common building blocks. However, does single standard hardware fit all the requirements? DHCI scale independently from compute and storage provides an option. It provides a more cost-efficient and flexible solution; however, there is no comparison from the performance point of view. Therefore, to address it, our objective is to conduct an empirical performance comparison to see which one performs better.The experiment result shows that DHCI performs almost the same as HCI on CPU utilization, memory, and network consumption. However, regarding storage and running time metrics, DHCI performs slightly higher storage throughput, IOPs, and less running time than HCI.