Chenhao Ying

#federated learning Oct 2026

FedNereus: Incentive-Aware Heterogeneous Federated Learning Under Resource Constraint

As public concern about data privacy continues to grow, federated learning (FL), as a machine learning technique that does not require the disclosure of user raw data, has attracted widespread attention. However, it still has many common problems. The first problem stems from the absence of incentive mechanisms utilized to incentivize more participation to improve the performance of learning models. The second problem arises from the limited resources of mobile devices that can be utilized for model training, which will seriously decrease the performance of learning models. To address the aforementioned issues, in this paper, we propose a novel incentive-aware collaborative construction of heterogeneous <underline>Fed</underline>erated lear<underline>N</underline>ing und<underline>e</underline>r <underline>re</underline>so<underline>u</underline>rce con<underline>s</underline>traint, namely FedNereus. In fact, FedNereus incentivizes more workers to participate in model training by applying an auction-based incentive mechanism. It is proved that FedNereus allows participating workers to report their actual cost as their bidding price, which is referred to as <italic>truthfulness</italic>. Furthermore, it also allows workers to obtain non-negative reward, which is also referred to as the <italic>individual rationality</italic>. In order to break the resource limitation, FedNereus meticulously designs the rules of model selection so that workers can only train a portion of the learning model based on their limited hardware capabilities, while maintaining the performance guarantee of learning models. The excess empirical risk of FedNereus is shown to be upper bounded by <inline-formula><tex-math notation="LaTeX">$\mathcal{O}(\frac{1}{T})$</tex-math><alternatives><mml:math><mml:mrow><mml:mi mathvariant="script">O</mml:mi></mml:mrow><mml:mo stretchy="false">(</mml:mo><mml:mfrac><mml:mn>1</mml:mn><mml:mi>T</mml:mi></mml:mfrac><mml:mo stretchy="false">)</mml:mo></mml:math><inline-graphic xlink:href="ying-ieq1-3713425.gif"/></alternatives></inline-formula>, where <inline-formula><tex-math notation="LaTeX">$T$</tex-math><alternatives><mml:math><mml:mi>T</mml:mi></mml:math><inline-graphic xlink:href="ying-ieq2-3713425.gif"/></alternatives></inline-formula> is the number of training rounds. Finally, extensive experiments are conducted, whose results show that FedNereus outperforms state-of-the-art approaches in different learning tasks.

Fu-Yuan Xia, Chen-Hao Ying, Xikun Jiang et al. · IEEE transactions on compute... · 0 citations

2026

Scalable Traffic Allocation in Dynamic Networks via End-to-End Imitation Learning

Networks with highly dynamic data transmission demands and network topologies are common in real world. A fundamental problem in such networks is achieving scalable traffic allocation to maximize long-term total throughput under link capacity constraints. However, state-of-the-art (SOTA) works lack scalability. This is primarily due to two reasons in large-scale networks: first, they require solving constrained optimization problems online, which leads to high decision latency; second, they rely on reinforcement learning algorithms for policy optimization, which are inefficient in exploration and challenging to train effectively. To address these issues, we propose the Fast Networked Control (FNC) policy framework, which firstly utilizes parallelizable neural network modules to process the state and generate raw decisions, followed by basic operations such as normalizations and comparisons, which do not require iteration or optimization, to obtain decisions that satisfy the constraints. Hence, FNC policy avoids solving constrained optimization problems and supports parallel execution, significantly reducing decision latency. Furthermore, this policy preserves gradient flow and supports backpropagation, which enable us to design an imitation learning algorithm to efficiently train the policy in an end-to-end manner. Experiments in large-scale networks show that our FNC policy achieves an average 8% improvement in demands satisfaction and 10 times reduction in decision latency versus SOTA works.

Zhaoxing Yang, Guiyun Fan, An-Jie Cao et al. · IEEE Transactions on Network... · 0 citations