Scheduling and Aggregation Design for Asynchronous Federated Learning over Wireless Networks

A recent study has explored the calibration of vision-language models, specifically CLIP, within a federated learning (FL) framework, revealing that fine-tuning under FL can negatively impact calibration metrics. The research highlights the need for improved strategies to enhance reliability in distributed settings, particularly through the proposed FL2oRA method.

A new method called CoLOR has been developed to address the challenges of open-set domain adaptation in machine learning, particularly under conditions where the background distribution of known classes shifts. This method provides theoretical guarantees for effective open-set recognition, even when new classes emerge that were not present during training.

A new study introduces multi-frequency Federated Learning (FL) for Human Activity Recognition (HAR) using head-worn sensors like earbuds and smart glasses. This approach addresses privacy concerns associated with centralized data collection by enabling decentralized model training across devices with varying sampling frequencies.

A recent study proposes the application of machine learning techniques to predict slot usage in Time Slotted Channel Hopping (TSCH) wireless sensor networks (WSNs), aiming to enhance energy efficiency by enabling nodes to enter a deep sleep state during idle periods. This approach is particularly relevant for industrial applications where low power consumption and reliable operation are essential.

A new study on optimal transport (OT) and Gromov-Wasserstein (GW) alignment has been released, focusing on Gaussian measures and their applications in data science and machine learning. This research addresses computational challenges by providing closed-form solutions for Gaussian distributions under quadratic cost, particularly enhancing the understanding of IGW alignment between uncentered Gaussians in separable Hilbert spaces.

A recent study highlights the significant impact of healthcare teamwork on cancer treatment outcomes, emphasizing that collaboration among healthcare professionals (HCPs) can influence patient survival rates. The research utilizes electronic health record (EHR) systems to model HCP interactions as networks and applies machine learning techniques to identify predictive signals related to patient outcomes.

A recent study published on arXiv highlights the challenges of developing accurate soft sensors for real-time batch process monitoring (BPM) in cell culture bioprocessing, particularly due to limited historical data and infrequent feedback. The research benchmarks various machine learning methods aimed at overcoming these obstacles to enhance monitoring of key process variables such as viable cell density and nutrient levels.

A novel decentralized multi-task fair federated learning (DMTFL) algorithm has been introduced to enhance virtual reality (VR) experiences by optimizing content delivery based on individual user needs at base stations (BSs). This approach addresses challenges such as low latency and quality of experience in wireless VR video streaming. The DMTFL algorithm personalizes caching strategies, allowing for more efficient prefetching of users' fields of view (FOV).