Ore-flow blending in open pit mining is a complex real-time scheduling decision problem constrained by high dimensional spatiotemporal dynamics and stochastic uncertainties. Traditional methods often struggle to reconcile the need for long term global stability with the requirement for sub second response times to dynamic events. To address this, we propose a heterogeneous ant colony optimization framework with semantic manifold learning. The framework bridges the gap between static offline learning and dynamic online adaptation. In the offline phase, we introduce a semantic manifold learning mechanism combining genetic programming with quality-diversity optimization. By constructing a behaviorally diverse archive of elite heuristics, this method provides the online system with a "warm start" and diverse initial perspectives, significantly enhancing real-time responsiveness. In the online phase, a heterogeneous ant colony system is deployed where different ant sub populations are coupled with distinct elite rules from the offline archive. This architecture enables parallel exploration of the solution space from multiple semantic perspectives, granting the system intrinsic robustness and high adaptability. Extensive experiments on real-world mine demonstrate that this method significantly outperforms state-of-the-art techniques in ore flow stability control and exhibits superior robustness and scalability across varying road network topologies, equipment configurations, and dynamic scenarios.

Power capping is an important technique for high-density servers to safely oversubscribe the power infrastructure in a data center. However, power capping is commonly accomplished by dynamically lowering the server processors’ frequency levels, which can result in degraded application performance. For servers that run important machine learning (ML) applications with Service-Level Objective (SLO) requirements, inference performance such as recognition accuracy must be optimized within a certain latency constraint, which demands high server performance. To achieve the best inference accuracy under the desired latency and server power constraints, this article proposes OptimML, a multi-input-multi-output (MIMO) control framework that jointly controls both inference latency and server power consumption, by flexibly adjusting the ML model size (and so its required computing resources) when server frequency needs to be lowered for power capping. Our results on a hardware testbed with widely adopted ML framework (including PyTorch, TensorFlow, and MXNet) show that OptimML achieves higher inference accuracy compared with several well-designed baselines, while respecting both latency and power constraints. Furthermore, an adaptive control scheme with online model switching and estimation is designed to achieve analytic assurance of control accuracy and system stability, even in the face of significant workload or hardware variations.

Dynamic constrained multiobjective optimization problems (DCMOPs) require algorithms to quickly track the feasible Pareto optima under dynamic environments. The existing dynamic constrained multiobjective evolutionary algorithms (DCMOEAs) normally focus on the convergence speed, but cannot well guarantee distribution. To address this issue, a subspace sparsity driven knowledge transfer strategy based DCMOEA is developed in this article, called SSDKT. First, reference points are introduced to partition objective space into multiple subspaces. Subsequently, the feasibility of each subspace is determined by the distribution of all historical feasible optimal solutions in it, and defined as the sparsity of subspace. A predictor based on the gated recurrent unit (GRU) network is further constructed to estimate the sparsity under the future environment. Once a new environment appears, a subspace transfer strategy is designed to generate an initial population. In each feasible subspace, the GRU-based prediction method is developed and competed with Kalman filter to generate the initial solution under the new environment. Based on the predicted solution of the nearest feasible neighbor, a potential initial individual in each infeasible subspace is produced by transferring the corresponding knowledge. The experimental results on various benchmarks verify that, compared with several state-of-the-art DCMOEAs, the proposed algorithm achieves the most competitive performance in solving DCMOPs.

Today's data centers often need to run various machine learning (ML) applications with stringent SLO (Service-Level Objective) requirements, such as inference latency. To that end, data centers prefer to 1) over-provision the number of servers used for inference processing and 2) isolate them from other servers that run ML training, despite both use GPUs extensively, to minimize possible competition of computing resources. Those practices result in a low GPU utilization and thus a high capital expense. Hence, if training and inference jobs can be safely co-located on the same GPUs with explicit SLO guarantees, data centers could flexibly run fewer training jobs when an inference burst arrives and run more afterwards to increase GPU utilization, reducing their capital expenses. In this paper, we propose GPUColo, a two-tier co-location solution that provides explicit ML inference SLO guarantees for co-located GPUs. In the outer tier, we exploit GPU spatial sharing to dynamically adjust the percentage of active GPU threads allocated to spatially co-located inference and training processes, so that the inference latency can be guaranteed. Because spatial sharing can introduce considerable overheads and thus cannot be conducted at a fine time granularity, we design an inner tier that puts training jobs into periodic sleep, so that the inference jobs can quickly get more GPU resources for more prompt latency control. Our hardware testbed results show that GPUColo can precisely control the inference latency to the desired SLO, while maximizing the throughput of the training jobs co-located on the same GPUs. Our large-scale simulation with a 57-day real-world data center trace (6500 GPUs) also demonstrates that GPU Colo enables latency-guaranteed inference and training co-location. Consequently, it allows 74.9 % of GPUs to be saved for a much lower capital expense.

Dynamic constrained multiobjective optimization problems (DCMOPs) abound in real-world applications and gain increasing attention in the evolutionary computation community. To evaluate the capability of an algorithm in solving DCMOPs, artificial test problems play a fundamental role. Nevertheless, some characteristics of real-world scenarios are not fully considered in the previous test suites, such as time-varying size, location and shape of feasible regions, the controllable change severity, as well as small feasible regions. Therefore, we develop the generators of objective functions and constraints to facilitate the systematic design of DCMOPs, and then a novel test suite consisting of nine benchmarks, termed as DCP, is put forward. To solve these problems, a dynamic constrained multiobjective evolutionary algorithm with a two-stage diversity compensation strategy (TDCEA) is proposed. Some initial individuals are randomly generated to replace historical ones in the first stage, improving the global diversity. In the second stage, the increment between center points of Pareto sets in the past two environments is calculated and employed to adaptively disturb solutions, forming an initial population with good diversity for the new environment. Intensive experiments show that the proposed test problems enable a good understanding of strengths and weaknesses of algorithms, and TDCEA outperforms other state-of-the-art comparative ones, achieving promising performance in tackling DCMOPs.

The key task in dynamic multiobjective optimization problems (DMOPs) is to find Pareto-optima closer to the true one as soon as possible once a new environment occurs. Previous dynamic multiobjective evolutionary algorithms (DMOEAs) normally focus on DMOPs with regular environmental changes, but neglect widespread random one, limiting their applications in real-world fields. To address this issue, a knowledge guided transfer strategy (KTS)-based DMOEA is proposed in this article. First, knowledge described as a two-tuple is extracted under each historical environment and preserved to a knowledge pool. Redundant knowledge is recognized and adaptively removed so as to guarantee the diversity of the pool. Second, a knowledge matching strategy is developed to re-evaluate the representative of each stored knowledge under a new environment, with the purpose of finding the most valuable one to promote positive knowledge transfer. Third, an improved knowledge transfer mechanism based on subspace alignment is introduced. By integrating it with the knowledge reuse mechanism, a hybrid transfer strategy is constructed to adaptively select the most suitable one in terms of the similarity degree of selected knowledge on the current environment, and then generate a new initial population. Experiments on 20 benchmark problems demonstrate that the KTS outperforms five state-of-the-art algorithms, achieving good versatility in solving DMOPs with both regular and random changes.

Power capping is an important technique for high-density servers to safely oversubscribe the power infrastructure in a data center. However, power capping is commonly accomplished by dynamically lowering the server processors’ frequency levels, which can result in degraded application performance. For servers that run important machine learning (ML) applications with Service-Level Objective (SLO) requirements, inference performance such as recognition accuracy must be optimized within a certain latency constraint, which demands high server performance. In order to achieve the best inference accuracy under the desired latency and server power constraints, this paper proposes OptimML, a multi-input-multi-output (MIMO) control framework that jointly controls both inference latency and server power consumption, by flexibly adjusting the machine learning model size (and so its required computing resources) when server frequency needs to be lowered for power capping. Our results on a hardware testbed show that OptimML achieves higher inference accuracy compared with several well-designed baselines, while respecting both latency and power constraints.

Aiming at the problems of color deviation and poor visibility of underwater images, an underwater image enhancement method based on color correction and transmission estimation (CCTE) is proposed. The method combines the global information of the attenuation channel with the pixel point information to adaptively correct the color deviation of the image so that the histogram distribution of the three channels is similar; the transmission is estimated based on the image blur to remove the blur effect of the underwater image; the multi-scale fusion process is combined Achieve artifact-free fusion of color-corrected images with sharp images. The qualitative and quantitative experimental results show that the enhanced images obtained by the CCTE method in this paper have good color perception and sharper details, and the UIQM and UCIQE scores on the RUIE dataset are 18.43% and 7.00% higher, respectively, compared to the optimal values in other methods, which are highly competitive.

MOTIVATION: Sequencing-based 3D genome mapping technologies can identify loops formed by interactions between regulatory elements hundreds of kilobases apart. Existing loop-calling tools are mostly restricted to a single data type, with accuracy dependent on a predefined resolution contact matrix or called peaks, and can have prohibitive hardware costs. RESULTS: Here, we introduce cLoops ('see loops') to address these limitations. cLoops is based on the clustering algorithm cDBSCAN that directly analyzes the paired-end tags (PETs) to find candidate loops and uses a permuted local background to estimate statistical significance. These two data-type-independent processes enable loops to be reliably identified for both sharp and broad peak data, including but not limited to ChIA-PET, Hi-C, HiChIP and Trac-looping data. Loops identified by cLoops showed much less distance-dependent bias and higher enrichment relative to local regions than existing tools. Altogether, cLoops improves accuracy of detecting of 3D-genomic loops from sequencing data, is versatile, flexible, efficient, and has modest hardware requirements. AVAILABILITY AND IMPLEMENTATION: cLoops with documentation and example data are freely available at: https://github.com/YaqiangCao/cLoops. SUPPLEMENTARY INFORMATION: Supplementary data are available at Bioinformatics online.

Combinatorial effects of epigenetic modifications on transcription activity have been proposed as "histone codes". However, it is unclear whether there also exist inter-nucleosomal communications among epigenetic modifications at single nucleosome level, and if so, what functional roles they play. Meanwhile, how clear nucleosome patterns, such as nucleosome phasing and depletion, are formed at functional regions remains an intriguing enigma. To address these questions, we developed a Bayesian network model for interactions among different histone modifications across neighboring nucleosomes, based on the framework of dynamic Bayesian network (DBN). From this model, we found that robust inter-nucleosomal interactions exist around transcription start site (TSS), transcription termination sites (TTS) or around CTCF binding sites; and these inter-nucleosomal interactions are often involved in transcription regulation. In addition to these general principles, DBN also uncovered a novel specific epigenetic interaction between H2A.Z and H4K20me1 on neighboring nucleosomes, involved in nucleosome free region (NFR) and nucleosome phasing establishment or maintenance. The level of negative correlation between neighboring H2A.Z and H4K20me1 strongly correlate with the size of NFR and the strength of nucleosome phasing around TSS. Our study revealed inter-nucleosomal communications as important players in signal propagation, chromatin remodeling and transcription regulation.