With the development of the Internet of things (IoT), federated learning (FL) has received increasing attention as a distributed machine learning (ML) framework that does not require data exchange. However, current FL frameworks follow an idealized setup in which the task size is fixed and the storage space is unlimited, which is impossible in the real world. In fact, new classes of these participating clients always emerge over time, and some samples are overwritten or discarded due to storage limitations. We urgently need a new framework to adapt to the dynamic task sequences and strict storage constraints in the real world. Continuous learning or incremental learning is the ultimate goal of deep learning, and we introduce incremental learning into FL to describe a new federated learning framework. New generation federated learning (NGFL) is probably the most desirable framework for FL, in which, in addition to the basic task of training the server, each client needs to learn its private tasks, which arrive continuously independent of communication with the server. We give a rigorous mathematical representation of this framework, detail several major challenges faced under this framework, and address the main challenges of combining incremental learning with federated learning (aggregation of heterogeneous output layers and the task transformation mutual knowledge problem), and show the lower and upper baselines of the framework.

With the growing adoption of low-altitude economy (LAE), such as urban air mobility and smart logistics, federated learning (FL) has become a key technology for enabling distributed intelligence while preserving data privacy. However, edge clients in such systems often suffer from limited resources and heterogeneous data, making them especially vulnerable to covert backdoor attacks. These threats can compromise critical functions such as navigation control or data confidentiality. Existing backdoor methods typically target specific clients, but their performance significantly degrades on edge clients due to insufficient data and computation power. To address this challenge, we propose LBKD, a novel backdoor attack strategy that combines large language models (LLMs) with bidirectional knowledge distillation. In our method, the attacker first obtains the target client's model and data, then uses LLMs guided by poisoning and data generation instructions to produce both poisoned and synthetic samples. During training, a more expressive distillation model engages in two-way knowledge exchange with the client model, enhancing its ability to learn and retain the backdoor pattern. The final poisoned model is submitted to the server, successfully embedding the backdoor into the global model. Experimental results show that LBKD significantly improves attack success rates (ASR), resulting in up to a 75.4% improvement over existing methods. Even under defense mechanisms, the ASR remains above 90% in some cases, demonstrating the method's effectiveness and robustness on edge clients.

The data security issue in federated learning is critical. While federated learning allows clients to collaboratively participate in the global model training without sharing private data, external eavesdroppers may intercept the model uploaded by the client to the server, revealing some sensitive information. Noise interference, i.e., adding noise to the client model before transmission, is an effective and efficient privacy-preserving method, but it degrades the learning performance of the system at the same time. In this paper, to address the challenge of system performance degradation caused by noise interference, we propose the FedNoise algorithm, which adopts two separate learning rates at the client and server respectively. By carefully tuning these learning rates, the global model can converge to the optimum. We provide theoretical proofs of the convergence of FedNoise for both strongly convex and non-convex loss functions and conduct simulations on real tasks. Numerical experimental results demonstrate that, under the same privacy protection level, FedNoise significantly outperforms the state-of-art scheme on datasets MNIST, Fashion-MNIST, and CIFAR10.

A key feature of federated learning (FL) is that not all clients participate in every communication epoch of each global model update. The rationality for such partial client selection is largely to reduce the communication overhead. However, in many cases, the unselected clients are still able to compute their local model updates, but are not “authorized” to upload the updates in this round, which is a waste of computation capacity. In this work, we propose an algorithm FedUmf—Federated Learning with Unauthorized Model Fusion that utilizes the model updates from the unselected clients. More specifically, a client computes the stochastic gradient descent (SGD) even if it is not selected to upload in the current communication epoch. Then, if this client is selected in the next round, it non-trivially merges the outdated SGD stored in the previous round with the current global model before it starts to compute the new local model. A rigorous convergence analysis is established for FedUmf, which shows a faster convergence rate than the vanilla FedAvg. Comprehensive numerical experiments on several standard classification tasks demonstrate its advantages, which corroborate the theoretical results.

Federated Learning (FL) is widely used in mobile computing as a communication-efficient distributed machine learning (ML) paradigm; however, it faces challenges such as model convergence to local optima or slow convergence due to the heterogeneity of client data. To mitigate data heterogeneity, the Nesterov Accelerated Gradient (NAG) method demonstrates its effectiveness by predictively updating the gradient to improve system performance. However, the performance of NAG depends heavily on the choice of decay coefficients; larger coefficients have greater acceleration but may lead to an unstable convergence process due to their unreasonable prediction of the descent gradient. To solve the above problems, this paper proposes the first radical heavy ball (RHB) method that combines momentum and NAG. In Stochastic Gradient Descent (SGD), momentum stabilizes the gradient descent process by integrating the historical gradients to update the parameters, and the RHB strategy decouples a single decay coefficient into an NAG component and a momentum component. The RHB introduces a gradient recall after each gradient acceleration by the NAG to strengthen the NAG’s perception of the historical gradients, thus stabilizing the gradient descent process. By weighing the historical gradients and the predicted gradient, RHB effectively mitigates the instability of NAG convergence and demonstrates better performance. As a result, the algorithm further mitigates the impact of customer data heterogeneity in FL and can effectively deliver global update information to participants without additional communication costs. We conduct comprehensive experiments in a binary function, single node, and federated model environment to analyze the convergence properties in non-convex loss functions. RHB exhibits better performance and less computational overhead than many existing algorithms.