ACM SIGMETRICS 2021
June 14-18, 2021
Opening the Black Box of Deep Learning: Some Lessons and Take-aways
Deep learning has rapidly come to dominate AI and machine learning in the past decade. These successes have come despite deep learning largely being a "black box." A small subdiscipline has grown up trying to derive better understanding of the underlying mathematical properties. Via a tour d'horizon of recent theoretical analyses of deep learning in some concrete settings, we illustrate how the black box view can miss out on (or even be wrong about) special phenomena going on during training. These phenomena are also not captured by the training objective. We argue that understanding such phenomena via mathematical understanding will be crucial for enabling the full range of future applications.
Sanjeev Arora is Charles C. Fitzmorris Professor of Computer Science at Princeton University. He has received Packard Fellowship (1997), Simons Investigator Award (2012), Gödel Prize (2001 and 2010), ACM Prize in Computing (2012), and the Fulkerson Prize (2012). He is a Fellow of the AAAS and Member of NAS.
AI for System – Infusing AI into Cloud Computing Systems
In the past fifteen years, the most significant paradigm shift in the computing industry is the migration to cloud computing, which brings unprecedented opportunities of digital transformation to business, society, and human life. The implication of this is profound. It means that cloud computing platforms have become part of the basic infrastructure of the world. Therefore, the non-functional properties of cloud computing platforms, including availability, reliability, performance, efficiency, security, sustainability, etc., become immensely important. The distributed nature, massive scale, and high complexity of cloud computing platforms ranging from storage to networking, computing and beyond present huge challenges to achieve effective and efficient building and operation of such software systems.
There is huge wealth of various types of data available throughout the entire development lifecycle of software systems. This is manifested even stronger with the paradigm shift to cloud computing as much more data are available on system runtime and workloads. Leveraging the amount of data, AI for System is to utilize AI/ML technologies to design and build high-quality cloud systems at scale. In this talk, I will first introduce the concept of AI for System and its research landscape. Then using a few projects at Microsoft as examples, I will talk about the work from Microsoft Research on AI for System and its impact. I will also discuss the research challenges and opportunities in AI for System moving forward.
Dr. Hsiao-Wuen Hon is corporate vice president of Microsoft, chairman of Microsoft’s Asia-Pacific R&D Group, and managing director of Microsoft Research Asia. He drives Microsoft’s strategy for research and development activities in the Asia-Pacific region, as well as collaborations with academia.
Dr. Hon has been with Microsoft since 1995. He joined Microsoft Research Asia in 2004 as deputy managing director, stepping into the role of managing director in 2007. He founded and managed Microsoft Search Technology Center from 2005 to 2007 and led development of Microsoft’s search products (Bing) in Asia-Pacific. In 2014, Dr. Hon was appointed as chairman of Microsoft Asia-Pacific R&D Group.
Prior to joining Microsoft Research Asia, Dr. Hon was the founding member and architect of the Natural Interactive Services Division at Microsoft Corporation. Besides overseeing architectural and technical aspects of the award-winning Microsoft Speech Server product, Natural User Interface Platform and Microsoft Assistance Platform, he was also responsible for managing and delivering statistical learning technologies and advanced search. Dr. Hon joined Microsoft Research as a senior researcher in 1995 and has been a key contributor to Microsoft’s SAPI and speech engine technologies. He previously worked at Apple, where he led research and development for Apple’s Chinese Dictation Kit.
An IEEE Fellow and a distinguished scientist of Microsoft, Dr. Hon is an internationally recognized expert in speech technology. Dr. Hon has published more than 100 technical papers in international journals and at conferences. He co-authored a book, Spoken Language Processing, which is a graduate-level textbook and reference book in the area of speech technology used in universities around the world. Dr. Hon holds three dozen patents in several technical areas.
Dr. Hon received a PhD in Computer Science from Carnegie Mellon University and a B.S. in Electrical Engineering from National Taiwan University.
Enabling Intelligent Services at the Network Edge
The proliferation of novel mobile applications and the associated AI services necessitates a fresh view on the architecture, algorithms and services at the network edge in order to meet stringent performance requirements. We present recent work addressing these challenges. In order to meet the requirement for low-latency, the execution of computing tasks moves form the cloud to the network edge, closer to the end-users. We considered the joint optimization of service placement and request routing in dense mobile edge computing networks with multidimensional constraints, that capture the storage requirements of the vast amounts of data needed. Furthermore the asymmetric bandwidth requirements exhibited by many emerging services is captured as well. We proposed an algorithm that achieves close-to-optimal performance using a randomized rounding technique and demonstrated through extensive evaluation that our approach can effectively utilize available storage, computation, and communication resources to maximize the number of requests served by low-latency edge cloud servers. Recent advances in network virtualization and programmability enable realization of services as chains, where flows can be steered through a pre-defined sequence of functions deployed at different network locations. We considered the problem of optimal deployment of such service chains where storage is a stringent constraint in addition to computation and bandwidth and proposed an approximation algorithm with provable performance guarantees that exhibit significant improvement on resource utilization in a realistic evaluation setting. Finally we considered the problem of traffic flow classification as it arises in firewalls and intrusion detection applications. We proposed an approach for realizing such functions based on a novel two-stage deep learning method for attack detection. Leveraging the high level of data plane programmability in modern network hardware, we demonstrated the realization of these mechanisms at the network edge.
Leandros Tassiulas is the John C. Malone Professor of Electrical Engineering at Yale University. His research interests are in the field of computer and communication networks with emphasis on fundamental mathematical models and algorithms of complex networks, architectures and protocols of wireless systems, sensor networks, novel internet architectures and experimental platforms for network research. His most notable contributions include the max-weight scheduling algorithm and the back-pressure network control policy, opportunistic scheduling in wireless, the maximum lifetime approach for wireless network energy management, and the consideration of joint access control and antenna transmission management in multiple antenna wireless systems. Dr. Tassiulas is a Fellow of IEEE (2007) and of ACM (2020). His research has been recognized by several awards including the IEEE Koji Kobayashi computer and communications award (2016), the ACM SIGMETRICS achievement award (2020), the inaugural INFOCOM 2007 Achievement Award “for fundamental contributions to resource allocation in communication networks,” several best paper awards including the INFOCOM 1994, 2017 and Mobihoc 2016, a National Science Foundation (NSF) Research Initiation Award (1992), an NSF CAREER Award (1995), an Office of Naval Research Young Investigator Award (1997) and a Bodossaki Foundation award (1999). He holds a Ph.D. in Electrical Engineering from the University of Maryland, College Park (1991).