ABSTRACT
Distributed machine learning (DML) can be an important capability for modern military to take advantage of data and devices distributed at multiple vantage points to adapt and learn. The existing distributed machine learning frameworks, however, cannot realize the full benefits of DML, because they are all based on the simple linear aggregation framework, but linear aggregation cannot handle the divergence challenges arising in military settings: the learning data at different devices can be heterogeneous, i.e Non-IID data, leading to model divergence, but the ability for devices to communicate is substantially limited (i.e., weak connectivity due to sparse and dynamic communications), reducing the ability for devices to reconcile model divergence. In this update, we (1) provide a systematic survey of DML in military settings, and (2) introduce a novel DML framework called aggregation in the mirror space (AIMS) that allows a DML system to introduce a general mirror function to map a model into a mirror space to conduct aggregation and gradient descent. Adapting the convexity of the mirror function according to the divergence force, we show that AIMS allows automatic optimization of DML. In this update, we will provide systematic evaluations including experimental evaluations using EMANE (Extendable Mobile Ad-hoc Network Emulator) for military communications settings.

BIOGRAPHY
Mr. Ryan Yang’s main research area is multi-agent optimization in both cooperative and game-theoretic settings. He has submitted a paper to MILCOM, introducing aggregation in the mirror space (AIMS), another dial that can be used alongside learning rate to accelerate optimization, especially useful in low-communication settings. Additionally, working within the ALTO Working Group, Ryan has had a paper published in SIGCOMM NAI’22, the workshop accompanying SIGCOMM, the flagship conference of computer communications and networking. Although very young, Ryan has received many awards recognizing his accomplishments. Ryan was invited to the USA International Math Olympiad Team Selection Test group and has received a bronze medal in the USA Physics Olympiad.

BIOGRAPHY
Professor Y. Richard Yang is a member of the Computer Systems Lab at Yale, where he founded and leads the Laboratory of Networked Systems (LANS). His research spans areas including computer networks, mobile computing, wireless networking, and network security. He led the P4P project (2008), which is the foundation for the establishment of the IETF Application-Layer Traffic Optimization (ALTO) Working Group and related Internet standards. Systems based on ALTO have been deployed in large-scale, production networks such as Deutsche Telekom since 2016. He led one of the most comprehensive research studies of Internet traffic engineering (2004-2010), with adoption by ATT (domain backup 2007), Cisco (ISP multihoming, 2004), and Google (COPE, 2006). He led one of the first systematic analysis and design of network localization (2004-2006), establishing the systematic network localization theory. He is a core member of the team that designed and implemented the first massive MIMO system called Argos (2012). He is among the first to design systematic, high-level network programming languages (Maple in 2013 and Trident 2018). His research has received extensive citations and featured in mainstream media including Economist, Forbes, Guardian, Chronicle of Higher Education, Information Week, MIT Technology Review, Science Daily, USA Today, Washington Post, and Wired, among others. He has received many awards, including the ACM SIGMobile Test of Time Award, US NSF CAREER Award, the Google Faculty Research Award, the Facebook Network Systems Award.