In this paper, we study a simultaneously transmitting and reflecting reconfigurable intelligent surface (STAR-RIS)enabled uplink mobile edge computing (MEC) network with rate-splitting multiple access (RSMA). We aim to minimize the sum energy consumption of all users by jointly optimizing the transmission and reflection beamforming at the STAR-RIS, the transmit power at the users and the number of offloading bits. The non-convex problem can be efficiently solved by an alternating optimization (AO) algorithm. Specifically, we apply the successive convex approximation (SCA) and semidefinite relaxation (SDR) to obtain a feasible rank-one solution of STAR-RIS beamforming, and then we utilize sequential convex programming (SCP) to solve the subproblem of offloading optimization. Simulation results demonstrate that our proposed RSMA+STAR-RIS scheme further reduce the energy consumption of users compared to other schemes.

The rapid progress in Deep Learning (DL) and Large Language Models (LLMs) has exponentially increased demands of computational power and bandwidth. This, combined with the high costs of faster computing chips and interconnects, has significantly inflated High Performance Computing (HPC) construction costs. To address these challenges, we introduce the Fire-Flyer AI-HPC architecture, a synergistic hardware-software co-design framework and its best practices. For DL training, we deployed the Fire-Flyer 2 with 10,000 PCIe A100 GPUs, achieved performance approximating the DGX-A100 while reducing costs by half and energy consumption by $40 \%$. We specifically engineered HFReduce to accelerate allreduce communication and implemented numerous measures to keep our Computation-Storage Integrated Network congestion-free. Through our software stack, including HaiScale, 3FS, and HAI-Platform, we achieved substantial scalability by overlapping computation and communication. Our system-oriented experience from DL training provides valuable insights to drive future advancements in AI-HPC.

This study proposes three‐node cooperative computing schemes for wireless sensor networks (WSNs), including parallel offload cooperative computing (POCC) and serial offload cooperative computing (SOCC). The cooperation system includes a task node with execution tasks and two peer help nodes. Specifically, the authors fully mobilise the computing and communication resources in the network by means of cooperative computing to minimise the energy consumption of processing tasks in the WSN while satisfying the delay constraints. The authors characterise these problems as optimal problems and find the optimal solutions using corresponding iterative algorithms, respectively. The simulation results show the superior performance of the proposed schemes in terms of energy consumption and node fairness. For different cases the superiority of POCC and SOCC is verified separately, and the SOCC is more suitable for the case where task node is located at the edge of WSN. Furthermore, the authors introduce a trade‐off scheme between POCC and SOCC, which can always achieve the optimal scheme in different cases.