The emergence of large-scale models, such as GPT-3, has become increasingly popular in the field of natural language processing (NLP) and also significantly advanced other artificial intelligence (AI) applications. Despite their many benefits, these models require massive amounts of computational resources and energy, making them difficult to deploy in real-world scenarios. According to Open AI, ChatGPT war trained on a dataset over 8 million web pages to allow it capture the important semantic link and hidden information for text generating tasks. Furthermore, training the GPT-3 model required 175 billion parameters and over 3 million GPU hours, which is beyond the reach of most individuals and organizations. Edge Artificial Intelligence (Edge AI), which refers to the practice of processing AI training tasks on local devices rather than in the cloud, has emerged as a promising solution to address these challenges. The most distinguished feature of Edge AI is it brings high-performance computing capabilities to the edge, where sensors and IoT devices are located. Under such settings, it reduces latency by processing data locally and expands computing power and data sources by integrating different end device. We argue that Edge AI could enable various end devices to not only perform inference but also continuously update the large-scale models to newly collected data, paving the way for lifelong on-device learning. Moreover, this technique could also preserve privacy by keeping data on the device or edge, which could be especially beneficial for training large-scale model when data are sensitive.
This call for papers invites researchers and practitioners to submit original contributions on the topic of giant model training based on Edge AI. We welcome papers that explore innovative approaches to training large-scale models on edge devices, as well as those that investigate the benefits and challenges of such approaches. Topics of interest include but are not limited to:
The authors are requested to submit their full research papers complying with the general scope of the journal. The submitted papers will undergo peer review process before they can be accepted. Notification of acceptance will be communicated as we progress with the review process.
Papers submitted to this journal for possible publication must be original and must not be under consideration for publication in any other journals. Prospective authors should submit an electronic copy of their completed manuscript to https://mc03.manuscriptcentral.com/bdma with manuscript type as “Special Issue on Edge AI Empowered Giant Model Training”. Further information on the journal is available at: https://ieeexplore.ieee.org/xpl/RecentIssue.jsp?punumber=8254253.
Deadline for submissions: 31 October 2023
Dongxiao Yu, Shandong University, China. E-mail: firstname.lastname@example.org.
Xu Chen, Sun Yat-sen University, China. E-mail: email@example.com.
Zhipeng Cai, Georgia State University, USA. E-mail: firstname.lastname@example.org.