In the context of green manufacturing, energy consumption issues have attracted widespread attention from all walks of life, especially in the manufacturing industry. In actual industrial production, the distributed blocking flow shop scheduling problem is a typical manufacturing production scenario, but its energy consumption problem has not been effectively solved. In this study, a two-stage cooperative discrete differential evolution with Q-learning (QTCDDE) is proposed to solve the energy-efficient distributed blocking flow shop scheduling problem (EEDBFSP) with total energy consumption (TEC) and total tardiness (TTD). An initialization strategy that considers both TEC and TTD is proposed to obtain an initial population with rich search space. In the first stage, a discrete differential evolution is designed to improve the quality of the solution. In the second stage, three types of local search operators are designed, and they are adaptively selected based on historical information and Q-learning. In addition, during the iterative process, the two stages cooperate and complement each other. Finally, each strategy of QTCDDE is effectively verified. Furthermore, QTCDDE is compared with state-of-the-art algorithms in the benchmark suite. Experimental results show that QTCDDE significantly outperforms state-of-the-art algorithms at the 95% confidence interval and effectively solves EEDBFSP.

We propose a novel framework for Russian Roulette and Splitting (RRS) tailored to wavefront path tracing, a highly parallel rendering architecture that processes path states in batched, stage-wise execution for efficient GPU utilization. Traditional RRS methods, with unpredictable path counts, are fundamentally incompatible with wavefront's preallocated memory and scheduling requirements. To resolve this, we introduce a normalized RRS formulation with a bounded path count, enabling stable and memory-efficient execution. Furthermore, we pioneer the use of neural networks to learn RRS factors, presenting two models: NRRS and AID-NRRS. At a high level, both feature a carefully designed RRSNet that explicitly incorporates RRS normalization, with only subtle differences in their implementation. To balance computational cost and inference accuracy, we introduce Mix-Depth, a path-depth-aware mechanism that adaptively regulates neural evaluation, further improving efficiency. Extensive experiments demonstrate that our method outperforms traditional heuristics and recent RRS techniques in both rendering quality and performance across a variety of complex scenes.

Recent research on learnable neural representations has been widely adopted in the field of 3D scene reconstruction and neural rendering applications. However, traditional feature grid representations often suffer from a substantial memory footprint, posing a significant bottleneck for modern parallel computing hardware. In this paper, we present neural vertex features, a generalized formulation of learnable representation for neural rendering tasks involving explicit mesh surfaces. Instead of uniformly distributing neural features throughout 3D space, our method stores learnable features directly at mesh vertices, leveraging the underlying geometry as a compact and structured representation for neural processing. This not only optimizes memory efficiency, but also improves feature representation by aligning compactly with the surface using task-specific geometric priors. Additionally, neural vertex features offer improved feature representation by compactly aligning with the surface using task-specific geometric priors. We validate our neural representation across diverse neural rendering tasks, with a specific emphasis on neural radiosity. Experimental results demonstrate that our method reduces memory consumption to only one-fifth (or even less) of grid-based representations, while maintaining comparable rendering quality and lowering inference overhead.

Peak carbon emissions and carbon neutrality have become important initiatives for the country to solve outstanding problems of resource and environmental constraints and promote green and low-energy development, and have attracted widespread attention from the industry. The distributed flow shop scheduling problem (DPFSP) is a typical problem that mainly works by consuming energy. However, DPFSP rarely considers energy efficiency and blocking constraints. In this study, an excellent bi-population cooperative discrete differential evolution (BCDDE) is proposed, aiming to address the energy-efficient distributed blocking flow shop scheduling problem (EEDBFSP) with total energy consumption (TEC) and total tardiness (TTD) as two objectives. A bi-population cooperative strategy is constructed to enhance the diversity of BCDDE, while utilizing it to initialize the population to enhance the quality of the initial solution. An adaptive local search operator strategy is developed to improve the BCDDE convergence. Critical and noncritical paths are devised to further optimize TEC and TTD objectives. The efficiency of each strategy related to BCDDE is verified and compared with state-of-the-art algorithms in the benchmark suite. Numerical results show that BCDDE becomes an efficient optimizer for the EEDBFSP, significantly outperforming the state-of-the-art algorithms at the 95% confidence interval.