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.

Sharding is a promising solution for improving blockchain scalability by distributing the workload across smaller groups of nodes called shards. However, it presents significant challenges, such as balancing tradeoffs between transactions per second (TPS), cross-shard transactions (CSTx), and confirmation latency in dynamic, online environments. While increasing the number of shards enhances TPS within individual shards, it also raises CSTx frequency, leading to higher failure probabilities, increased communication overhead, and prolonged confirmation times. Moreover, managing the number of shards and their configurations under one-sided feedback scenarios adds further complexity. To address these challenges, we model sharding as an online one-sided feedback optimization problem, focusing on maximizing long-term utilities. We introduce a polynomial-time online algorithm that adapts selection probabilities based on feedback information to address this NP-hard problem. Through rigorous analysis, we demonstrate that our approach achieves dynamic regret that grows sub-linearly over time. Extensive evaluations on real-world datasets confirm that our method outperforms existing baseline algorithms in terms of practical performance.

Distributed edge computing systems that participate in Emergency Demand Response (EDR) programs can adjust workload across heterogenous edges to reduce total energy consumption. Unfortunately, this approach may not always reduce sufficient energy as required by EDR. In this paper, we propose to leverage Electrical Vehicles (EVs) and Vehicle-to-Grid (V2G) techniques to provide energy to the edge system, and design an auction mechanism to incentivize EVs to discharge energy for the edges. Yet, we face critical challenges, such as the uncertainty of EV bid arrivals, the restriction of discharging deadlines, and the desire to achieve required economic efficiency. To overcome such challenges, we design a novel online approach,$E^{3}$DR, of multiple algorithms that decompose our original NP-hard social cost minimization problem into two subproblems, solve the first subproblem via reformulation, the primal-dual optimization theory, and a careful payment design, and solve the second subproblem via standard solvers. We have rigorously proved that our approach finishes in polynomial time, achieves truthfulness and individual rationality economically, and leads to a parameterized competitive ratio for the long-term social cost. Through extensive evaluations using real-world data traces, we have validated the superior practical performance of our approach compared to existing algorithms.