CNS Core: Small: Wireless Network Control in Uncooperative and Adversarial Environments

最近移动的和媒体丰富应用的增长极大地增加了对无线容量的需求，使无线网络变得紧张。这种需求的急剧增长对当前的无线网络提出了挑战，并要求更好地利用稀缺的无线资源的新网络算法。 此外，现代通信网络经常在不友好的环境中操作，其中一些用户可能是不合作的，甚至是恶意的，并且试图破坏网络服务。 该项目开发了网络控制算法，该算法在对抗性环境中有效地运行，越来越多地表征现实网络设置，从而导致网络性能的显着改善，并实现新兴的无线应用。该项目为网络开发了一个新的优化框架，其中一些节点，以及外部动态（例如，链路速率、外生到达）可能是不合作的，并且表现出对抗甚至恶意的行为。这个新的框架设想了网络控制算法，是“安全的设计”，在面对敌对的动态。此外，开发的控制算法将具有“鲁棒优化”的味道，在这个意义上，它们将从一开始就被设计为在最坏情况下表现良好，同时在正常条件下保持接近最佳的性能。研究议程包括以下任务：（i）不合作环境中的网络优化：使用基于模型的强化学习技术为网络开发控制算法，其中节点的子集是不可控的，并使用一些未知的固定控制策略。(ii)对抗环境中的网络优化：开发在线学习算法，以最大限度地提高网络的吞吐量和网络效用，其中不可控节点可以采取任意和可能的非静态动作。(iii)恶意环境下的网络优化：描述网络在对抗性流注入造成的溢出情况下的性能，为对抗者开发最佳流注入策略，以及减轻这种对抗性流注入影响的网络控制算法。该奖项反映了NSF的法定使命，并通过使用基金会的知识价值和更广泛的影响审查标准进行评估，被认为值得支持。

Recent growth in mobile and media-rich applications has greatly increased the demand for wireless capacity, straining wireless networks. This dramatic increase in demand poses a challenge for current wireless networks, and calls for new network algorithms that make better use of scarce wireless resources. Moreover, modern communication networks frequently operate in unfriendly environments, where some of the users may be uncooperative, or even malicious, and try to disrupt network services. This project develops network control algorithms that operate effectively in adversarial environments that increasingly characterize realistic network settings, thus leading to dramatic improvement in network performance and enabling emerging wireless applications.This project develops a new optimization framework for networks where some of the nodes, as well as the external dynamics (e.g., link rates, exogenous arrivals), may be uncooperative and exhibit adversarial or even malicious behavior. This novel framework envisions network control algorithms that are "secure-by-design", in the face of adversarial dynamics. Moreover, the developed control algorithms will have a "robust optimization" flavor, in the sense that they will be designed from the outset to perform well under worst-case conditions, while maintaining nearly optimal performance under normal conditions. The research agenda includes the following tasks: (i) Network Optimization in Uncooperative Environments: Use techniques from model-based reinforcement learning to develop control algorithms for networks where a subset of nodes are uncontrollable and use some unknown stationary control policy. (ii) Network Optimization in Adversarial Environments: Develop online learning algorithms for maximizing throughput and network utility in networks where uncontrollable nodes can take arbitrary and possibly non-stationary actions. (iii) Network Optimization in Malicious Environments: Characterize the network's performance in overflow due to adversarial flow injections, develop optimal flow injection policies for the adversary, and network control algorithms to mitigate the effect of such adversarial flow injections.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.