摘要
Robotics&Machine Learning Daily News Daily News的新闻记者兼工作人员新闻编辑根据来自弗吉尼亚州亚历山大市Rx记者的新闻报道，发明者杜拉科维奇，S Enad(加利福尼亚州帕洛阿尔托)，纳拉马拉普，戈帕尔(加利福尼亚州圣克拉拉)，索达尼，阿维纳什(加利福尼亚州圣何塞)于2021年4月22日申请的专利于2024年5月专利号11995463转让给Marvell Asia Pte Ltd.(新加坡，新加坡E)。以下引文由新闻编辑从发明人提供的背景信息中获得：“应用机器学习(ML)是一个蓬勃发展的领域，它利用非线性处理单元和算法的级联来进行特征提取和转换，具有广泛的用途和应用。ML通常包括两个阶段，训练，它使用丰富的训练数据来训练多个机器学习模型和推理，并将训练好的机器学习模型应用到实际应用中。这两个阶段中的每一个都对其底层基础结构提出了不同的要求。可以使用各种基础结构，例如图形处理单元(GPU)、中央处理单元(CPU)、现场可编程门阵列(FPGA)、具体地说，作为非限制性示例，训练阶段侧重于与训练的模型和再训练频率相匹配的GPU或ASIC基础设施，其中训练阶段的关键目标是实现高性能并减少训练时间。另一方面，推理阶段侧重于随应用程序、用户和数据扩展的基础设施，推理阶段的关键目标是实现能源（如性能功耗比）和资本（如投资回报率）效率。
Abstract
By a News Reporter-Staff News Editor at Robotics & Machine Learning Daily News Daily News-A patent by the inventors Durakovic, S enad (Palo Alto, CA, US), Nalamalapu, Gopal (Santa Clara, CA, US), Sodani, Avina sh (San Jose, CA, US), filed on April 22, 2021, was published online on May 28, 2024, according to news reporting originating from Alexandria, Virginia, by News Rx correspondents. Patent number 11995463 is assigned to Marvell Asia Pte Ltd. (Singapore, Singapor e). The following quote was obtained by the news editors from the background informa tion supplied by the inventors: "Applied Machine Learning (ML) is a booming fiel d that utilizes a cascade of layers of nonlinear processing units and algorithms for feature extraction and transformation with a wide variety of usages and app lications. ML typically involves two phases, training, which uses a rich set of training data to train a plurality of machine learning models, and inference, wh ich applies the trained machine learning models to actual applications. Each of the two phases poses a distinct set of requirements for its underlying infrastru ctures. Various infrastructures may be used, e.g., graphics processing unit (GPU ), a central processing unit (CPU), a Field Programmable Gate Array (FPGA), an A pplication Specific Integrated Circuit (ASIC), etc. Specifically, the training p hase focuses on, as a non-limiting example, GPU or ASIC infrastructures that sca le with the trained models and retraining frequency, wherein the key objective o f the training phase is to achieve high performance and reduce training time. Th e inference phase, on the other hand, focuses on infrastructures that scale with the applications, user, and data, and the key objective of the inference phase is to achieve energy (e.g., performance per watt) and capital (e.g., return on i nvestment) efficiency.
NETL
NSTL