This paper discusses the application of DIKW model in medical field, and introduces how the model transforms data into information and knowledge, and finally realizes the purpose of improving hospital performance and service quality. This paper innovatively proposes the application of TIF model in the construction of pharmaceutical platform, including information technology chain, credit chain, industrial chain and financial chain. Finally, the paper establishes the performance evaluation index system, and puts forward the conclusion and prospect. Through these methods, hospitals can better understand their operations, improve service quality and efficiency, and thus provide better medical services to patients.
In recent years, embedded platforms became attractive targets to deploy machine learning (ML) -empowered control and processing systems. However, securing a high performance executability to satisfy the hard real-time constraints of such systems on resource-limited platforms, such as FPGAs, is a challenging task. This paper introduces a methodology for deploying hardware-accelerated, adaptable convolutional neural networks (CNN) on embedded FPGA platforms. It enables automated synthesis of hardware IPs, instantiation of CNN architectures and mapping of the CNN layers processing to hardware IPs. This will result in iteration-free deployment reducing the expensive cost of HW synthesis, so that the acceleration hardware IPs are synthesized once and configured at runtime following the CNN architecture. To demonstrate the applicability and assess the performance of our deployment model, we implemented and deployed a segment-based acceleration of a image classification CNN on Xilinx ZYBO Z7 FPGA board. Among others, we have analyzed the computation performance, accuracy and trade-offs between CNN size, image segmentation size, resources utilization and scalability.
This paper studies GPU-based implementations of CRYSTALS-Dilithium (Dilithium) algorithm, which is one of four post-quantum algorithms that NIST has standardized. We focus on the scenarios of batch processing and explore low latency implementations by parallelizing the critical components of Dilithium with the consideration of overhead due to inter-thread data sharing. We also investigate the techniques of using GPU memories and warp-level instructions for data sharing in GPU-based implementations of Dilithium. All of the efforts lead to effective low latency implementations that are suited to cloud applications. Finally, we conduct evaluations on two testbeds with NVIDIA GeForce RTX 3060 and RTX A4000, respectively. The experimental results show that our low latency implementations save up to 41 percent (resp. 33 percent) latency at low task loads on the former testbed (resp. the latter testbed) when compared with the baseline implementation. We also provide engineers with a practical strategy to achieve better overall performance for applications that the number of arrived tasks is not predictable.
Job scheduling is a critical component of workload management systems that can significantly influence system performance, e.g., in HPC clusters. The scheduling objectives are often mixed, such as maximizing resource utilization and minimizing job waiting time. An increasing number of researchers are moving from heuristic-based approaches to Deep Reinforcement Learning approaches in order to optimize scheduling objectives. However, the job scheduler's state space is partially observable to a DRL-based agent because the job queue is practically unbounded. The agent's observation of the state space is constant in size since the input size of the neural networks is predefined. All existing solutions to this problem intuitively allow the agent to observe a fixed window size of jobs at the head of the job queue. In our research, we have seen that such an approach can lead to “window staleness” where the window becomes full of jobs that can not be scheduled until the cluster has completed sufficient work. In this paper, we propose a novel general technique that we call split window, which allows the agent to observe both the head and tail of the queue. With this technique, the agent can observe all arriving jobs at least once, which completely eliminates the window staleness problem. By leveraging the split window, the agent can significantly reduce the average job waiting time and average queue length, alternatively allowing the use of much smaller windows and, therefore, faster training times. We show a range of simulation results using HPC job scheduling trace data that supports the effectiveness of our technique.
This paper proposes a visualization function for developers in volunteer computing. In recent years, information systems such as e-Learning systems and groupware have been introduced in companies and educational institutions to promote DX, but the introduction and operation of these systems incur a large cost burden. Therefore, we are developing a distributed information system that can be introduced and operated at low cost by utilizing surplus computers (idle computers) in an organization. Idle computers are unstable depending on the operations of their original users and have different performances. In order to use idle computers efficiently, it is necessary to improve the performance of the system and modify it to add new functions, taking into account the differences in behavior and performance of each machine. In this paper, visualization for developers to easily check the behavior and performance of a computer in an idle state is developed, and its usefulness is evaluated through experiments using a real system.
Inverse materials design can be viewed as a reverse engineering process of material constituents from a user given set of targets characterizing material properties. Machine learning techniques could be used to map a feature space $x$ of material properties to a contextual latent space $z$ over an object space $O$ , and then reconstruct a distinct object space $O^{\prime}$ over the latent space $z$ to maximize the context $c$ . In this paper, we introduce a new machine learning tool, called MatFlow, for inverse material design using machine learning. In MatFlow, we first learn the latent space $z$ characterizing a context feature set $c$ . MatFlow then helps identify novel materials $O^{\prime}$ over the latent space $z$ with a potential to exceed the contextual threshold $\theta$ without destabilizing the latent space. We explain MatFlow features and its capabilities using an application in quantum dye material discovery. The main focus of this paper is developing a computational strategy to identify $z$ in the context of $c$ , and inform the characteristics of $z$ space values of $O^{\prime}$ most likely to meet the contextual threshold $\theta$ .
To compare the safety, effectiveness and effect of the AI early warning system and the traditional fetal monitoring system. Methods: Establish and deploy AI remote fetal monitoring system; From January 1, 2022 to December 31, 2022, 440 pregnant women in the third trimester of pregnancy admitted to the Maternal and Child Health Hospital of Haikou City and meeting the admission criteria were selected and randomly assigned to the artificial intelligence early warning remote fetal visit group 220 cases and the traditional group 220 cases. In the AI group, remote fetal monitoring system was used for examination and treatment, while in the traditional group, pregnant women went to the hospital for examination and delivery according to the usual standard procedures. The effectiveness and safety of the AI-early warning impromptu fetal heart test system were observed, and the rates of perinatal adverse outcomes (low birth weight, perinatal death, premature delivery, neonatal asphyxia, cord blood oxygen saturation and fetal distress) of pregnant women in the two groups were compared. Results: In the test of artificial intelligence early warning fetal monitoring system, the AI group detected 332 normal fetal cardiogram cases of pregnant women, 206 suspected pathological changes in 18 cases; In the traditional group, 421 cases of normal fetal cardiogram were detected, 117 cases of suspected pathological changes were detected in 16 cases. The sensitivity and specificity of normal pregnant women (true positive was normal group, false negative was suspicious + lesion) were 81.55% and 96.98% respectively. The sensitivity and specificity of pregnant women with abnormal fetal monitoring map were 100% and 82%. The low birth weight rate was 6/220 in the AI group and 13/220 in the control group, with a total value of 2.695 (p value 0. 101). Perinatal death was 0 in both groups. The rate of preterm infants in AI group was 7/220, compared with 15/220 in control group, and peason2 was 3.062,p value 0.08; Neonatal asphyxia rate was 1/220 in the AI group and 0/220 in the control group, and the exact detection rate by Fisher was p 0. 3 72. The low oxygen saturation rate of umbilical cord blood was 3/220 in the AI group and 2/220 in the control group. The p value of Fisher's accurate detection was 1.0. The fetal distress rate was 50/220 in the AI group and 76/220 in the control group, with a total of 7.518 (p value 0.006). Among the monitored perinatal adverse outcome indicators, only the fetal distress rate was significantly different between the two groups (p =0.05). Conclusions: The AI early-warning remote fetal care system is as safe and effective as the traditional fetal care system, which is helpful to break the time and space limitations of the traditional fetal care system and to sinking high-quality medical resources to the primary care. Remote fetal heart monitoring with ai warning can improve the adverse perinatal outcomes of fetal distress.
Efficient allocation of funds is crucial for enhancing the quality of postgraduate education. The DIKWP model, integrating Data, Information, Knowledge, Wisdom, and Practice, offers a comprehensive framework for decision-making in fund allocation. Through data collection, information processing, knowledge synthesis, and the application of wisdom, this model aids in making informed decisions regarding fund distribution. The study discusses the allocation of funds across key areas such as faculty development, research support, student scholarships, and infrastructure enhancement. By utilizing the DIKWP model, universities and educational institutions can optimize the allocation of funds to maximize the impact on postgraduate education quality. This research contributes to the advancement of fund allocation strategies in the context of postgraduate education, fostering a more efficient and transparent system. It offers a holistic approach to decision-making, ensuring that available resources are used effectively to support the development of academic faculty, encourage research excellence, provide financial aid to students, and improve infrastructure.
Nowadays, networks are increasingly reliant on software frameworks and virtualization. To obtain relevant pre-dictions of the behavior of their protocols, network emulation tools play a crucial role. However, as networks grow in size and complexity, the emulation task demands more physical resources such as CPU and memory. Consequently, a single physical machine is no longer sufficient to handle large-scale network emulation. Instead, distributed network emulation tools must be employed for resource-intensive network emulation tasks. In resource-intensive distributed network emulation, the physical resource allocation is NP-hard to perform appropriately. In this paper, we propose SCBG (spectral clustering based greedy node placement algorithm) as a means to enhance the bandwidth utilization between physical nodes in distributed emulation. SCBG aims to address the communication bottleneck in distributed network simulation, surpassing the algorithm proposed by Distrinet, while also improving the algorithm's execution speed. Additionally, SCBG incorporates optimization for time dilation when physical bandwidth resources are severely insufficient. This optimization leads to a reduction in the additional time required for network simulation after applying time dilation. Experimental results demonstrate that SCBG surpasses the algorithm in Distrinet across a diverse range of intricate large-scale networks.
