By Casimer DeCusatis
While you might not associate OFC with topics such as machine learning (ML) or artificial intelligence (AI), these technologies are actually playing an increasingly important role in many types of optical communication research. In this blog, we’ll take a look at two of the OFC workshops on these topics that you won’t want to miss.
Will Machine Learning Replace QoT/Performance Estimation and Has it Reached the Stage of Commercial Deployment?
In optical fiber networks, the quality of transmission (QoT) degrades due to a number of physical layer effects, including both linear and nonlinear noise and interference. Predicting the QoT beforehand provides useful guidance for network resource management and deployment, provided that the predictions are accurate. Traditionally, network performance is estimated using deterministic models and algorithms. These algorithms can be difficult to derive, computationally intensive, and sometimes do a poor job estimating typical or worst case results. In recent years, thanks in part to improved data acquisition on software defined networks (SDN), new models based on machine learning have demonstrated advantages over conventional approaches. This is especially helpful in cases where a clear analytical solution may not be available, or when such a solution is computationally impractical. But how advanced are current solutions, and are they ready to make the transition from research to commercial applications? The OFC workshop on machine learning for QoT brings together a panel of experts to address this question. This workshop will consider the requirements for successful deployment of machine learning based QoT estimation, the obstacles to large scale deployment, and related issues such as scalability, universality, and trust for such models. Generation of reliable, standardized training data sets is a key issue, as well as testing machine learning models based on labeled or unlabeled data.
What Will the Future Machine Learning and Artificial Intelligence Systems Look Like?
Also at this year’s OFC, there will be a lively discussion on the future of AI in optical networks at a workshop devoted to this topic. The future of AI looks promising, but still faces significant challenges. Machine learning workloads are doubling every three to four months. We used to rely on successive generations of technology to build smaller devices which would consume less power, in line with Dennard’s Law. However, this is no longer the case; modern computer systems may be able to demonstrate predictive capabilities on par with human experts in some areas, but at the cost of extremely high-power consumption. The first part of the OFC workshop in this area will discuss scalability issues such as these, along with opportunities for fundamental changes in computing system architectures enabled by intelligent programmable photonics.
In the second part of this workshop, emerging technologies for accelerated machine learning training and inference workloads will be discussed. For example, trends in neuromorphic computing aren’t just for high performance computing; they are moving closer to the network edge. These methods currently suggest a potential 3-6 orders of magnitude improvement in energy-efficiency and throughput compared to the traditional von Neumann computer architectures. Further, advances in quantum computing using photonics may enable currently unattainable increases in performance at a fraction of the power expenditure. The use of Ising machines and photonic tensor processing ASICs may open up new opportunities for QoT prediction and other applications. This session will address the role of photonics in computing for future hyperscale data centers handling AI and machine learning workloads.
These are just two of the AI sessions planned for this year’s OFC, which provides a great opportunity for you to hear from experts in the field about emerging technologies that will change how we configure optical networks in the coming years. What’s the most annoying problem in your network design that you wish could be solved with AI? Drop me a line on Twitter (@Dr_Casimer), and maybe we’ll explore your suggestions in a future blog. For cited sources, please contact @Dr_Casimer on Twitter.
Learn more about workshops at OFC. Register today to attend OFC in person or virtually.
Posted: 1 March 2022 by
| with 0 comments