Modeling and simulation (M&S) has long played an important role in developing tactics and evaluating the measure of effectiveness (MOE) for the underwater warfare system. In simulation-based acquisition, M&S technology facilitates decisions about future equipment procurements, such as a mobile decoy or a torpedo. In addition, assessment of submarine tactical development, during an engagement against a torpedo, can be conducted using M&S techniques. This paper presents a case study that applies discrete event systems specification-based M&S technology to develop a simulation of an underwater warfare system, specifically, an anti-torpedo combat system, to analyze the MOE of the system. The entity models required for M&S are divided into three sub-models: controller, maneuver, and sensor model. The developed simulation allows us to conduct a statistical evaluation of the overall underwater warfare system under consideration, an assessment of the anti-torpedo countermeasure's effectiveness, and an assessment of tactics development of the underwater vehicle. Moreover, it can be utilized to support the decision-making process for future equipment procurements. In order to analyze the system effectiveness, we performed extensive combat experiments by varying parameters, such as various tactics and weapon performance. The experimental results show how the factors influence the MOEs of the underwater warfare system.

The recent advance of self-supervised learning associated with the Transformer architecture enables natural language processing (NLP) to exhibit extremely low perplexity. Such powerful models demand ever-increasing model size, and thus, large amounts of computations and memory footprints. In this paper, we propose an efficient inference framework for largescale generative language models. As the key to reducing model size, we quantize weights by a non-uniform quantization method. Then, quantized matrix multiplications are accelerated by our proposed kernel, called nuQmm, which allows a wide trade-off between compression ratio and accuracy. Our proposed nuQmm reduces the latency of not only each GPU but also the entire inference of large LMs because a high compression ratio (by low-bit quantization) mitigates the minimum required number of GPUs. We demonstrate that nuQmm can accelerate the inference speed of the GPT-3 (175B) model by about 14.4 times and save energy consumption by 93%.

No abstract

Introduction: This study examined the perceived competence of Clinical Research Coordinators (CRCs) using several conceptual frameworks. Accurate self-assessment of one’s professional competence is a critical component in the career navigation process and contributes to (a) identifying and securing professional development (training), (b) leveraging professional strengths, and (c) integrating self-knowledge into a comprehensive career plan. Method: A survey design gathered responses from a sample of 119 CRCs in a southeastern region of the USA Two conceptual frameworks were used to represent aspects of CRC professional competence: the eight Joint Task Force (JTF) competence domains, and perceptions of strengths and training needs from a list of 12 task categories. Results: The JTF domain with the lowest competence level was Development and Regulations, while the highest was Communication. Perceived competence increased incrementally with years of experience. Top strengths involved direct patient interaction and data management. Tasks in need of training included project management and reporting issues. Variations in responses were based on years of experience as a CRC. Conclusion: Our results demonstrate an association between the self-reported strengths and training needs of CRCs and experience. This information can contribute to the self-directed career navigation of CRCs.

The deployment of widely used Transformer architecture is challenging because of heavy computation load and memory overhead during inference, especially when the target device is limited in computational resources such as mobile or edge devices. Quantization is an effective technique to address such challenges. Our analysis shows that for a given number of quantization bits, each block of Transformer contributes to translation quality and inference computations in different manners. Moreover, even inside an embedding block, each word presents vastly different contributions. Correspondingly, we propose a mixed precision quantization strategy to represent Transformer weights by an extremely low number of bits (e.g., under 3 bits). For example, for each word in an embedding block, we assign different quantization bits based on statistical property. Our quantized Transformer model achieves 11.8× smaller model size than the baseline model, with less than -0.5 BLEU. We achieve 8.3× reduction in run-time memory footprints and 3.5× speed up (Galaxy N10+) such that our proposed compression strategy enables efficient implementation for on-device NMT.

Transformer is being widely used in Neural Machine Translation (NMT). Deploying Transformer models to mobile or edge devices with limited resources is challenging because of heavy computation and memory overhead during inference. Quantization is an effective technique to address such challenges. Our analysis shows that for a given number of quantization bits, each block of Transformer contributes to translation accuracy and inference computations in different manners. Moreover, even inside an embedding block, each word presents vastly different contributions. Correspondingly, we propose a mixed precision quantization strategy to represent Transformer weights with lower bits (e.g. under 3 bits). For example, for each word in an embedding block, we assign different quantization bits based on statistical property. Our quantized Transformer model achieves 11.8× smaller model size than the baseline model, with less than -0.5 BLEU. We achieve 8.3× reduction in run-time memory footprints and 3.5× speed up (Galaxy N10+) such that our proposed compression strategy enables efficient implementation for on-device NMT.

No abstract

Simulation of a system-of-systems (SoS) model, which consists of a combat model and a network model, has been used to analyze the performance of network-centric warfare in detail. However, finding the combat model parameters satisfying the required combat power using simulation can take a long time for two reasons: (1) the prolonged execution time per simulation run and (2) the enormous number of simulation runs. This paper proposes a simulation-based optimization method for the SoS-based simulation model to overcome these problems. The method consists of two processes: (1) the transformation of the SoS-based model into an integrated model using the neural network to reduce the execution time and (2) the optimization of the integrated model using the genetic algorithm with ranking and selection to decrease the number of simulation runs. The experimental result reveals that the proposed method significantly reduced the time for finding the optimal combat parameters with an acceptable level of accuracy.