Federated clustering effectively addresses the Non-IID problem by organizing clients into clusters for the training of personalized models. However, current federated clustering methods often cluster clients based on a single dimension, and fail to simultaneously achieve low computational cost, high accuracy, and strong privacy preservation. To address this problem, this manuscript proposes a novel approach called Gravitational Clustering Federated Learning (GCFL). GCFL treats each client as an object in a latent space, where the position encodes the local model and the mass encodes client importance. By simulating gravitational interactions between clients, GCFL enables adaptive clustering. Extensive experiments on Non-IID datasets validate the effectiveness of GCFL, and comparative analysis with state-of-the-art methods demonstrates that the proposed approach achieves more reasonable clustering and faster convergence.

Knowledge-based Visual Question Answering (VQA) is a challenging task that requires models to access external knowledge for reasoning. Large Language Models (LLMs) have recently been employed for zero-shot knowledge-based VQA due to their inherent knowledge storage and in-context learning capabilities. However, LLMs are commonly perceived as implicit knowledge bases, and their generative and in-context learning potential remains underutilized. Existing works demonstrate that the performance of in-context learning strongly depends on the quality and order of demonstrations in prompts. In light of this, we propose Knowledge Generation with Frozen Language Models (KGFLM), a novel method for generating explicit knowledge statements to improve zero-shot knowledge-based VQA. Our knowledge generation strategy aims to identify effective demonstrations and determine their optimal order, thereby activating the frozen LLM to produce more useful knowledge statements for better predictions. The generated knowledge statements can also serve as interpretable rationales. In our method, the selection and arrangement of demonstrations are based on semantic similarity and quality of demonstrations for each question, without requiring additional annotations. Furthermore, a series of experiments are conducted on A-OKVQA and OKVQA datasets. The results show that our method outperforms some superior zero-shot knowledge-based VQA methods.

The ability to perform short-term traffic flow forecasting is a crucial component of intelligent transportation systems. However, accurate and reliable traffic flow forecasting is still a significant issue due to the complexity and variability of real traffic systems. To improve the accuracy of short-term traffic flow forecasting, this paper presents a novel hybrid prediction framework based on Support Vector Regression (SVR) that uses a Random Forest (RF) to select the most informative feature subset and an enhanced Genetic Algorithm (GA) with chaotic characteristics to identify the optimal forecasting model parameters. The framework is evaluated with real-world traffic data collected from eight sensors located near the I-605 interstate highway in California. Results show that the proposed RF-CGASVR model achieves better performance than other methods.