Green Edge AI: A Contemporary Survey
(2312.00333)Abstract
AI technologies have emerged as pivotal enablers across a multitude of industries, including consumer electronics, healthcare, and manufacturing, largely due to their resurgence over the past decade. The transformative power of AI is primarily derived from the utilization of deep neural networks (DNNs), which require extensive data for training and substantial computational resources for processing. Consequently, DNN models are typically trained and deployed on resource-rich cloud servers. However, due to potential latency issues associated with cloud communications, deep learning (DL) workflows are increasingly being transitioned to wireless edge networks near end-user devices (EUDs). This shift is designed to support latency-sensitive applications and has given rise to a new paradigm of edge AI, which will play a critical role in upcoming 6G networks to support ubiquitous AI applications. Despite its potential, edge AI faces substantial challenges, mostly due to the dichotomy between the resource limitations of wireless edge networks and the resource-intensive nature of DL. Specifically, the acquisition of large-scale data, as well as the training and inference processes of DNNs, can rapidly deplete the battery energy of EUDs. This necessitates an energy-conscious approach to edge AI to ensure both optimal and sustainable performance. In this paper, we present a contemporary survey on green edge AI. We commence by analyzing the principal energy consumption components of edge AI systems to identify the fundamental design principles of green edge AI. Guided by these principles, we then explore energy-efficient design methodologies for the three critical tasks in edge AI systems, including training data acquisition, edge training, and edge inference. Finally, we underscore potential future research directions to further enhance the energy efficiency of edge AI.
Overview
-
The paper surveys efforts to implement energy-efficient edge AI, particularly with the advent of 6G networks, and the challenges posed by deep neural networks' energy demands.
-
It explores energy-conscious approaches for edge training, data acquisition, and inference to maintain optimal performance with limited resources.
-
Techniques discussed include adaptive sampling, learning-centric communications, gradient sparsification, and knowledge distillation to minimize energy usage.
-
The research addresses local training adaptation, model quantization, dynamic device selection, and data offloading—all aimed at reducing energy expenditure in edge AI.
-
Potential future research areas are identified, such as hardware-software co-design, neuromorphic computing, and using green energy to power edge AI systems.
Efforts to embed AI at the edge of networks are rapidly gaining traction, yet they bring with them a significant energy challenge. As applications proliferate across consumer electronics, healthcare, and manufacturing, leveraging deep neural networks (DNNs) has become common. These AI models, often requiring large amounts of data, have typically been processed on cloud servers. However, the latency in communication and potential privacy issues have begun pushing deep learning tasks closer to users on wireless edge networks.
Edge AI, particularly in support of upcoming sixth-generation (6G) networks, promises ubiquitous AI applications with critical performance. But the limited resources of wireless edge networks and the energy-intensive nature of DNNs present substantial challenges. AI's transformative power hinges on the need to balance between resource limitations and intensive computation requirements. Thus, an energy-conscious approach to edge AI that ensures optimal and sustainable performance is imperative.
The reviewed paper provides a survey on green edge AI, focusing on energy-efficient design methodologies for training data acquisition, edge training, and edge inference—three critical tasks in edge AI systems. It addresses efficient data acquisition for centralized edge learning by considering data sampling and transmission methods while ensuring minimal energy expenditure. Novel strategies are proposed, including adaptive sampling rates and learning-centric communications that prioritize important data and adapt to system dynamics.
For distributed edge model training, the paper discusses methods to minimize on-device model updates and computations. Techniques such as model quantization, gradient sparsification, and knowledge distillation are suggested for conserving energy. Additionally, resource management strategies like local training adaptation, dynamic device selection, and data offloading are critical for reducing energy usage in edge AI systems.
Finally, the paper explores potential future research directions, suggesting interests in integrated sensing and communication (ISAC), hardware-software co-design for edge AI platforms, and neuromorphic computing with spiking neural networks and compute-in-memory techniques. It also looks at the potential of harnessing green energy to power edge AI systems without incurring carbon emissions.
In summary, the paper articulates the energy challenges associated with edge AI and outlines a comprehensive set of strategies and methodologies to enhance energy efficiency. By doing so, it lays out a road map for future sustainable developments in edge AI systems within the context of emerging 6G networks.
Create an account to read this summary for free: