New Ideas in Networked Systems — 2026 |
||||||||
| Home | Attending NINeS | Pods | Papers | Call for Talks | CfP | FAQ | Statement of Purpose | Organizers |
The inaugural NINeS conference will be happening on February 10th. We are a global conference, so you can participate from anywhere in the world, and at a time that is convenient for you! Please join us.
There are two main ways you can participate on February 10:
Go to one of our pods, where you can join in the fun with other people. These pods require registration, so please sign up early.
Watch our live stream. The live stream will include three live keynote talks (see schedule below), and in between those talks will be showing streams from various pods over the course of the day. We will post a link and details closer to the actual date.
We are using Discord for our discussions, so join our server to talk to other participants, and to discuss the papers with the authors and others. People are already on the Discord server, so join and start participating today!
In addition, the full set of taped talks will be available on our web site starting on February 10th. These includes:
| Time (UTC) | Speaker |
|---|---|
| 3 - 4am | Henry Hong Xu: Hyper-Scale AI Infrastructure: Efficiency, Reliability, and Beyond |
| 3 - 4pm | Michael Schapira: Adventures in Analog Computing |
| 8 - 8:40pm | Bruce Maggs: Engineering Lessons Learned the Hard Way: War Stories from Akamai Technologies and Emerald Innovations |
We have witnessed extraordinary development of AI over the recent years. One driving force behind this stride is the hyper-scale infrastructure that empowers the large machine learning models with tens of thousands of accelerators like GPU and high-speed interconnects. Utilizing and managing AI infrastructure is often the bottleneck to real-world use of AI. In this talk, I will share some recent systems we have built (and deployed in production in some cases) that help people train and serve their models faster, and identify faults and root causes more easily.
Growing computational demands have renewed interest in unconventional hardware, including quantum and analog systems, for solving complex problems. The utility of such hardware depends on its expressiveness, i.e., the range of problems it can efficiently address. Quantum hardware is restricted to binary solution spaces, limiting its practical scope. In contrast, recent advances in analog hardware and algorithms enable far greater expressiveness by supporting both continuous and binary variables. This review introduces the Analog Computing Expressiveness (ACE) framework, a hierarchical classification of problems suited to analog computing, and shows that recent progress enables solutions across all levels of the hierarchy. We further demonstrate how core problems in optimization, game theory, and control theory naturally fit within ACE, underscoring the broader applicability of analog computing. Finally, we discuss extensions of the ACE hierarchy to emerging applications, including control methods based on learned world models.
Bruce Maggs was the first head of engineering at both Akamai Technologies, the operator of the world’s leading content delivery network, and Emerald Innovations, the developer of a wireless health monitoring platform used by pharmaceutical companies in clinical trials. This talk describes the technologies underlying the services provided by both companies and shares some amusing anecdotes about incidents that occurred during the rollout of these services, noting important engineering lessons learned along the way.