INFOCOM 2023

IEEE International Conference on Computer Communications

举办时间&地点：17–20 May 2023 // New York area // USA

这次会议共收到1312篇论文投稿，最终录用252篇，录用率仅为19.2%。

keynotes:

Technology ecosystems often undergo significant transformations as they mature. For example, telephony, the Internet, and PCs all started with a single provider, but each is now served by a competitive market that uses comprehensive and universal technology standards to provide compatibility. In this talk, I will present our view on how the cloud ecosystem, just over fifteen years old, could evolve as it matures.

随着技术生态系统的成熟，它们往往会经历重大转变。例如，电话、互联网和个人电脑都是从单一的供应商开始的，但现在每个供应商都由一个竞争激烈的市场提供服务，这个市场使用全面和通用的技术标准来提供兼容性。在这次演讲中，我将展示我们对云生态系统的看法，这个刚刚超过15年的生态系统将如何随着它的成熟而发展。

国外论文：

Robust Federated Learning against Backdoor Attackers

Priyesh Ranjan; Ashish Gupta; Federico CorÚ; Sajal K. Das

Federated learning is a privacy-preserving alternative to regular edge learning that involves data transfer to the server. However, anonymous submission of gradients to the server can be used by adversaries to influence the model outcome, particularly through backdoor attacks where a trigger pattern is added to the data shard to manipulate the model outcomes on a specific sub-task. This work aims to identify such adversaries and isolate their weight updates to mitigate the effects of the attack. We propose two graph theoretic algorithms to identify the adversaries while maintaining data privacy. Under a classification task, our experimental results show that the proposed algorithms are superior to existing methods especially when numbers of attackers are more than the normal clients. Our algorithms are effective and robust to the attackers adding backdoor patterns at different location in the targeted images.

联邦学习是常规边缘学习的一种隐私保护替代方案，涉及到将数据传输到服务器。然而，攻击者可以使用匿名向服务器提交梯度来影响模型结果，特别是通过后门攻击，在数据分片中添加触发模式来操纵特定子任务上的模型结果。这项工作旨在识别这样的对手，并隔离他们的权重更新，以减轻攻击的影响。我们提出了两种图论算法来识别攻击者，同时保持数据隐私。在分类任务下，实验结果表明，当攻击者数量大于正常客户端时，本文提出的算法优于现有方法。我们的算法对于攻击者在目标图像的不同位置添加后门模式具有有效和鲁棒性。

 国内论文：

论文：RED: Distributed Program Deployment for Resource-aware Programmable Switches

RED:资源感知可编程交换机的分布式程序部署　　

作者：贾星薪（Northeastern University东北大学），李福亮*，陈松林，高程希，王鹏飞，王兴伟*

        论文创新地提出了资源感知的可编程交换机分布式程序部署方案，实现整个数据平面资源的高效、均衡利用，增强了可编程数据平面的灵活性和可扩展性。团队从基础方案的设计到实验部署测试，历时一年多经过多次修改和调整达到预期效果，并且基于该方案进行一系列后续研究工作，依托此成果已与字节跳动和阿里巴巴展开研究合作。