The rapid proliferation of mobile devices within Internet of Things (IoT) has substantially heightened the demand for mobile edge computing (MEC). Fog computing (FC) is a more advanced form of edge computing that allows computing nodes to cooperate with each other. Reconfigurable intelligent surfaces (RIS) have emerged as a critical technology for optimizing wireless communication environments, attracting considerable attention. In this paper, we develop an online optimization problem for RIS-aided mobile FC deployed across wireless networks with computing nodes at the base stations (BS). We propose a Lyapunov-drift-plus-penalty-based, space aggregation-assisted proximal policy optimization (LSAPPO) algorithm to tackle the challenges in online optimization problem in RIS-aided mobile FC system. Our technique integrates a reinforcement learning (RL) algorithm employing the proximal policy optimization (PPO) agent, further enhanced by Lyapunov drift-plus-penalty optimization. The space aggregation technique effectively consolidates excessive decision variables and channel state information (CSI) into a manageable set of parameters to streamline the computing framework. Numerical simulation result shows that our proposed algorithm surpasses the benchmarks, underscoring the effectiveness in complicated wireless networks. Furthermore, we introduce the multi-agent LSAPPO algorithm to address the distributed demands of practical scenarios. The multi-agent LSAPPO algorithm enhances convergence speed and performs better in large-scale problems.

Transfer learning leverages existing models to help train new models, rather than training the new models from scratch. Unfortunately, realizing transfer learning in distributed cloud-edge networks faces critical challenges such as online training, uncertain network environments, time-coupled control decisions, and the balance between resource consumption and model accuracy. In this paper, targeting classification tasks, we study the settings of both homogeneous and heterogeneous transfer learning in cloud-edge networks via orchestrating model placement, data dispatching, and inference aggregation. We formulate non-linear mixed-integer programs of long-term cost optimization over consecutive time slots, and design polynomial-time online algorithms by exploiting the real-time trade-off between preserving previous control decisions and applying new control decisions. Our approaches produce new models by combining the existing pre-trained offline models and the online models that are continuously updated based on the inference results of data samples arriving in streams. We rigorously prove that our approaches only incur the number of inference mistakes no greater than a constant times that of the single best model in hindsight, and achieve constant competitive ratios for the total cost. Evaluations have confirmed the superior performance of our approaches compared to other state-of-the-art methods upon real-world data traces, under text classification transfer learning tasks.

Due to the influence of stochastic gradients, the existing algorithms suffer from slow convergence, noise explosion, and even failure to converge in practice, which motivates us to propose an accelerated algorithm to tackle these issues. Recognizing the potential of gradient, momentum, and conjugate gradient as promising search directions, we propose a 3-D acceleration algorithm, which uses a weighted combination of these three basis. Specifically, in order to analyze the dynamics of the discrete-time algorithm during the update process, we provide a general framework for approximating the discrete-time algorithm in the weak sense by a continuous-time stochastic differential equation. We exploit the continuous-time formulation together with Lyapunov drift optimization to derive novel adaptive step sizes, which effectively improve the performance of the algorithm in stabilizing noise and accelerating convergence. Extensive numerical experiments demonstrate the proposed algorithm’s superiority in convergence rate, computation complexity, and noise robustness compared to state-of-the-art baselines.

Transfer learning does not train from scratch but leverages existing models to help train the new model of better accuracy. Unfortunately, realizing transfer learning in distributed cloud-edge networks faces critical challenges such as online training, uncertain network environments, time-coupled control decisions, and the balance between resource consumption and model accuracy. We formulate distributed transfer learning as a non-linear mixed-integer program of long-term cost optimization. We design polynomial-time online algorithms by exploiting the real-time trade-off between preserving previous decisions and applying new decisions, based on primal-dual one-shot solutions for each single time slot. While orchestrating model placement, data dispatching, and inference aggregation, our approach produces new models via combining the existing offline models and the online models being trained using weights adaptively updated based on inference upon data samples that dynamically arrive. Our approach provably incurs the number of inference mistakes no greater than a constant times that of the single best model in hindsight, and achieves a constant competitive ratio for the total cost. Evaluations have confirmed the superior performance of our approach compared to alternatives on real-world traces.

Due to the enormous energy consumption and the wide geographic distribution, Electrical Vehicle (EV) charging stations are believed to have great potential in Emergency Demand Response (EDR) participation. However, EDR limits the electricity drawn from the power grid by the charging station, and can pose threats to satisfying EVs’ charging demand. In this paper, in order to complement the charging station’s energy supply to meet the dynamic EV charging demand, we formulate an online EV charging scheduling problem under EDR as a non-linear mixed-integer program, and propose a novel polynomial-time online algorithm and auction mechanism to jointly incentivize EVs with energy to sell their energy and utilize the charging station’s local generator to produce energy. Our approach conducts an auction in each single round based on a primal-dual method and ties these auctions over time to optimize the system’s long-term social cost, while accommodating the local generator’ on/off-state control, each EV bidder’s cumulative energy budget constraint, and the power grid’s EDR energy cap. Our approach achieves the economic properties of truthfulness, individual rationality, and computational efficiency simultaneously for each auction, and a parameterized-constant competitive ratio for the long-term social cost. By rigorous theoretical analysis and trace-driven experimental studies, the results exhibit that our approach outperforms multiple alternative algorithms regarding the social cost, attains the economic properties, and also executes efficiently in practice.

Mobile users are often reluctant to participate in federated learning to train models, due to the excessive consumption of the limited resources such as the mobile devices’ energy. We propose an auction-based online incentive mechanism, FLORA, which allows users to submit bids dynamically and repetitively and compensates such bids subject to each user’s long-term battery capacity. We formulate a nonlinear mixed-integer program to capture the social cost minimization in the federated learning system. Then we design multiple polynomial-time online algorithms, including a fractional online algorithm and a randomized rounding algorithm to select winning bids and control training accuracy, as well as a payment allocation algorithm to calculate the remuneration based on the bid-winning probabilities. Maintaining the satisfiable quality of the global model that is trained, our approach works on the fly without relying on the unknown future inputs, and achieves provably a sublinear regret and a sublinear fit over time while attaining the economic properties of truthfulness and individual rationality in expectation. Extensive trace-driven evaluations have confirmed the practical superiority of FLORA over existing alternatives.