Federated Learning Distributed Training for AI Excellence with Shared Model

By AIT News Desk On Jul 26, 2022

ADLINK’s Edge Server implements edge federated learning, disrupting the traditional centralized ML training model and solving privacy issues for personal information for application to privacy-critical financial, medical, retail and Internet verticals.

ADLINK’s Edge Server featured dual Intel Xeon Scalable Silver/Gold processors and provides high performance and efficiency platform for federated learning.
Federated learning is to accelerate applications in intensive computing such as privacy computing, AIoT, gene sequencing, financial business, medical, video processing, network security, etc.

ADLINK Technology, Inc., a global leader in edge computing, has implemented edge federated learning application with the MECS-7211 as an edge computing server to resolve the privacy of personal information and for accelerated applications in intensive computing, such as privacy computing, machine learning, gene sequencing, financial business, medical, video processing and network security.‎

With the rapid development of IoT and popularization of 5G networks, a large number of terminal devices are connected to the network and generating massive amounts of data. Traditional data computing and analysis are based on cloud computing. With the rapid increase in data quantities, the transmission from application terminals to cloud computing centers can cause delay and data leakage. Timely and effective processing of data becomes a major challenge for cloud computing centers. ‎‎A new computing model – edge computing – processes data at the network edge; providing intelligence services closer to the person or object requiring that data, and allowing network services to respond more efficiently and better meet the business needs of applications in IoT, Internet of Vehicles (IoV), industrial control, intelligent manufacturing and video processing.‎ ‎The introduction of edge computing technology has eased the network burden of cloud centers, but also raised security issues. Localization of data can easily hinder data interaction. In addition, data security and application specifications have been tightened in recent years. Topics such as GDPR data privacy and data protection have been given high priority.‎ ‎The centralized computation of data used in traditional machine learning algorithms cannot meet current and future data processing requirements and limits the development of artificial intelligence.‎

95 Percent of Retail Leaders Prioritize AI, but Only 40 Percent Feel Ready Due to Data Gaps

Nov 21, 2024

Baffle Announces Vector Database Protection to Enhance Data Security for GenAI Applications

Nov 21, 2024

H2O.ai Generative and Predictive AI Now Validated on the Dell AI Factory with NVIDIA

Nov 21, 2024

Prev Next 1 of 41,001

‎In this context, federated machine learning provides the solution to edge computing security problems.‎ Federated learning is a machine learning framework that uses encrypted private data for operations by participants, exchanging only the parameters, weights, and gradients of the encrypted model, without moving the raw data out of the local area or moving the encrypted raw data set. Multiple agencies can model data usage and implement machine learning while allowing multiple organizations to conduct data usage and machine learning modeling under the requirements of user privacy protection, data security and government regulations.‎ Federated learning, as a distributed machine learning paradigm, ensures that data are not leaked and allows enterprises to use more data training models, carry out joint modeling, realize AI collaboration, and provide strong support for the implementation of privacy protection computing solutions.‎

Recently, ADLINK and Clustar jointly launched an integrated machine for edge federated learning. ‎Using ADLINK’s MECS-7211 as an edge computing server and Clustar’s FPGA isomeric acceleration card, the system performs qualitative analysis and hardware optimization of commonly used complex operators in federated learning to facilitate user acceleration of distributed dense state machine learning tasks. ‎Efficient storage, computing and data transmission systems play a collaborative optimization role in the efficient operation of isomeric systems. Compared with traditional CPU architectures, performance is improved by 7 times, and improved by 2 times with power consumption reduced by 40% over CPU+GPU platforms. ‎This integrated machine for edge federated learning is suitable or financial, medical, and data center applications that require extensive data analysis and focus on privacy, and has already been deployed in many instances.

Julian Ye, Director of Network Communications and Public Sector of ADLINK, said, “ADLINK’s MECS series is a 5G-based edge computing platform. As one of the initiators of the Open Telecom IT Infrastructure (OTII) specification, MECS series products conform to OTII industry specifications, using isomeric architecture to flexibly support FPGA, GPU, and 5G acceleration cards and other expansion cards. ‎Compact size design coupled with an operating environment that supports a wide range of temperatures makes the MECS series suitable for distributed architecture applications and deployment at the edge and on the application side of the network. ‎ADLINK and Clustar jointly launched an integrated machine for edge federated learning to optimize computing systems collaboratively and expand the application of MECS series products. In the future, ADLINK will continue to cooperate with Clustar in the AI field to enrich application scenarios of edge computing.”

‎ADLINK is committed to edge computing and the AI industry, has more than two decades of R&D experience in telecommunications network computing; focuses on network security, 5G, edge computing, IoT and other infrastructure products and services; provides leading, hardened and reliable hardware and software solutions; and has become the driving force for artificial intelligence to change the world.

[To share your insights with us, please write to sghosh@martechseries.com]