The network has become a bottleneck for generative artificial intelligence (GAI) jobs. Accelerating GAI jobs in edge data centers using hybrid electrical/optical switch is considered a promising solution. This architecture optimizes bandwidth utilization by enabling demand-aware topology reconfiguration through flexible configuration of optical circuit switche optical circuit switches (OCS). However, frequent topology reconfiguration may increase latency. Therefore, there is a balanced relationship between latency and bandwidth utilization. In this article, we propose a multigranularity adaptive interleaved algorithm for service scheduling in edge data centers. First, different degrees of time slot shifts are introduced based on the latency sensitivity of jobs, where large bandwidth GAI jobs are transmitted in a single hop by configuring a demand-aware topology. Additionally, when the reconfiguration threshold is met, low-priority ports are prioritized for reconfiguration to ensure latency requirements are met. This approach effectively resolves the tradeoff between bandwidth utilization and latency by decoupling them from each other. Simulation results show that this approach can effectively reduce the latency and improve the network throughput.

This paper presents an automated point-of-care semen analysis method that uses smartphone imaging to visually measure sperm concentration and motility of semen samples. The proposed method follows the same visual tracking scheme as laboratory semen analysis systems, aiming to match clinical standards while being suitable for point-of-care use. A boundary-sensitive segmentation network is developed to identify and distinguish sperm from impurities in raw semen. A novel occlusion-aware multi-sperm tracking algorithm is proposed to tackle challenges posed by smartphone imaging and undiluted raw semen samples. For automated motility measurement, an occlusion-awareness module is proposed to robustly track multiple sperm during frequent sperm crossover/occlusion. The module combines the segmented contour and kinematic-based probabilistic modeling to determine the occlusion status of both targets and measurements, facilitating fundamental improvement to feasible joint event enumeration to enable robust data association. The proposed method achieved a high success rate of 95.14% for tracking occluded sperm, with low mean errors for sperm concentration (2.03 million/ml) and motility (1.58%), outperforming existing multi-sperm tracking methods. In clinical tests involving 50 participants, our method exhibited good agreement with clinical standards (Spearman rank correlation coefficients of 0.94 for concentration and 0.89 for motility) even when used by inexperienced users.Note to Practitioners—Semen analysis is the gold standard method for assessing male reproductive capacity. Clinical semen analysis routinely uses professional computer-assisted semen analysis (CASA) systems to examine sperm concentration and motility; however, clinical visits for semen analysis are not always feasible due to the unavailability of such professional systems and the mental stress brought by clinical visits. This work provides a point-of-care semen analysis method. For hardware, a smartphone microscopic imaging modality was developed to enable clear visualization of sperm with the built-in smartphone camera. For software, an occlusion-aware multi-sperm tracking algorithm was proposed to automatically measure sperm concentration and motility. In addition to intensively validating the proposed point-of-care method against clinical CASA systems, this work also analyzed different types of tracking failure and quantified their effects on the automated evaluation of sperm concentration and motility. The techniques pave the way for further improvement in both point-of-care and clinical semen analysis. Overall, this work offers an accessible and reliable tool for automated male fertility evaluation.

We propose to personalize a 2D human pose estimator given a set of test images of a person without using any manual annotations. While there is a significant advancement in human pose estimation, it is still very challenging for a model to generalize to different unknown environments and unseen persons. Instead of using a fixed model for every test case, we adapt our pose estimator during test time to exploit person-specific information. We first train our model on diverse data with both a supervised and a self-supervised pose estimation objectives jointly. We use a Transformer model to build a transformation between the self-supervised keypoints and the supervised keypoints. During test time, we personalize and adapt our model by fine-tuning with the self-supervised objective. The pose is then improved by transforming the updated self-supervised keypoints. We experiment with multiple datasets and show significant improvements on pose estimations with our self-supervised personalization. Project page with code is available at https://liyz15.github.io/TTP/.

In this paper, an uncalibrated dynamic visual servoing algorithm is proposed and analyzed. No calibration or robot model is needed. After a brief introduction of the development of uncalibrated visual servoing, the theoretical backgrounds and mathematical requirements of recursive least square (RLS) are stated respectively. Then the core uncalibrated visual servoing algorithm, in RLS form, or more technically, VS-RLS as well as its performance analysis, is investigated. After that, the experimental 6DOF Puma560 simulation of static and moving target tracking is demonstrated. Finally the weak and strength of the algorithm as well as the potential and promising improvements are discussed.