Det(Com)<sup>2</sup>: Deterministic Communication and Computation Integration Toward AIGC Services
Weiting Zhang, Nian Tang, Dong Yang, Ruibin Guo, Hongke Zhang, Xuemin Shen
Abstract
As an emerging intelligence paradigm, artificial intelligence generated content (AIGC) is envisioned to be a key technique for Internet of intelligence, which inevitably puts forward higher requirements for the network capability from both the forwarding and computing perspectives. This article proposes a novel deterministic communication and computation integration architecture, that is, Dot(Com) <sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> . for future networks to effectively support large AI model services such as distributed training, rapid deployment, and collaborative inference. Deep reinforcement learning-based solutions are developed to achieve cross-domain computation resource orchestration and deterministic transmission scheduling. The proposed learning-based solutions can efficiently schedule computing tasks of large AI models among multiple geographically dispersed computing domains while guaranteeing bounded latency and near-zero packet loss, thus facilitating integrated resource management and supporting large AI model services across their life cycles. Finally, we present a case study on communication and computation integration and discuss open research issues.