Eye diagnosis is a method for inspecting systemic diseases and syndromes by observing the eyes.With the development of intelligent diagnosis in traditional Chinese medicine(TCM);artificial intelligence(AI)can improve ...Eye diagnosis is a method for inspecting systemic diseases and syndromes by observing the eyes.With the development of intelligent diagnosis in traditional Chinese medicine(TCM);artificial intelligence(AI)can improve the accuracy and efficiency of eye diagnosis.However;the research on intelligent eye diagnosis still faces many challenges;including the lack of standardized and precisely labeled data;multi-modal information analysis;and artificial in-telligence models for syndrome differentiation.The widespread application of AI models in medicine provides new insights and opportunities for the research of eye diagnosis intelli-gence.This study elaborates on the three key technologies of AI models in the intelligent ap-plication of TCM eye diagnosis;and explores the implications for the research of eye diagno-sis intelligence.First;a database concerning eye diagnosis was established based on self-su-pervised learning so as to solve the issues related to the lack of standardized and precisely la-beled data.Next;the cross-modal understanding and generation of deep neural network models to address the problem of lacking multi-modal information analysis.Last;the build-ing of data-driven models for eye diagnosis to tackle the issue of the absence of syndrome dif-ferentiation models.In summary;research on intelligent eye diagnosis has great potential to be applied the surge of AI model applications.展开更多
This paper takes the school-enterprise cooperation between the University of Electronic Science and Technology of China(UESTC)and Baidu(China)Co.,Ltd.as an example to build the Paddle Paddle(Sichuan)AI Education Innov...This paper takes the school-enterprise cooperation between the University of Electronic Science and Technology of China(UESTC)and Baidu(China)Co.,Ltd.as an example to build the Paddle Paddle(Sichuan)AI Education Innovation Center by enjoying the best of both UESTC and Baidu(China),and cooperating with 16 universities in Sichuan Province.With the support of this center,both the school and the enterprise successfully built a school-enterprise collaborative AI innovative talent training model,which mainly serves the universities,industries,and districts.Furthermore,this training model is able to facilitate the update of the industrial intelligence and the regional economic development in southwest China,and also provide a reference for deep integration of school-enterprise collaboration and the cultivation of innovative talents in electronic information fields.展开更多
针对边缘计算环境下人工智能(Artificial Intelligence,AI)模型训练效率低下的问题,提出了基于边缘云计算的混合并行训练框架(Edge-Cloud based Hybrid Parallel Training Framework,ECHPT).ECHPT能在终端设备、边缘服务器和云计算中心...针对边缘计算环境下人工智能(Artificial Intelligence,AI)模型训练效率低下的问题,提出了基于边缘云计算的混合并行训练框架(Edge-Cloud based Hybrid Parallel Training Framework,ECHPT).ECHPT能在终端设备、边缘服务器和云计算中心之间实现AI模型和数据样本的自适应调度.ECHPT将模型计算和数据任务调度问题建模为训练时间最小化的优化问题,设计了调度算法对优化问题进行求解.实现了由设备、边缘服务器和云服务器组成的硬件原型.实验结果表明,与现有框架相比,ECHPT可以有效缩短AI模型的训练时间.展开更多
针对目前人工智能开放平台普遍存在预训练模型开发成本高、行业数据标注成本高等问题,本文研究了一种使用弱监督训练范式且任务可解耦的人工智能算法平台(Weak Supervision&Task Decoupling AI Platform,简称为WSTD-AP)。不同于基...针对目前人工智能开放平台普遍存在预训练模型开发成本高、行业数据标注成本高等问题,本文研究了一种使用弱监督训练范式且任务可解耦的人工智能算法平台(Weak Supervision&Task Decoupling AI Platform,简称为WSTD-AP)。不同于基于传统强监督学习的人工智能算法平台,WSTP将大规模无标签数据用于不同规模的模型训练,在有限成本情况下充分利用“扩展法则”(Scaling Law),使模型在下游任务(如图像识别、目标检测等)达到更好的效果。展开更多
拥有千亿级别参数的大语言模型(large language model,LLM)已为今天的人工智能和云服务带来了巨大的技术和商业变革.然而,大模型训练与传统的通用云计算(例如,亚马逊EC2弹性计算服务)之间存在较多根本性的网络行为差异,从而带来了很多...拥有千亿级别参数的大语言模型(large language model,LLM)已为今天的人工智能和云服务带来了巨大的技术和商业变革.然而,大模型训练与传统的通用云计算(例如,亚马逊EC2弹性计算服务)之间存在较多根本性的网络行为差异,从而带来了很多新的挑战,主要包括流量模式差异造成负载难均衡(挑战1)、多训练任务通信竞争影响GPU利用率(挑战2),以及对网络故障的高敏感性(挑战3)等.因此,为通用云计算设计的数据中心网络技术(例如,网络架构、选路方法、流量调度,以及可靠性保障方法等)已不适合今天的大模型训练,这要求专门为大模型训练设计新型的数据中心网络以及配套的技术方案.介绍了阿里云专门为大模型训练设计的数据中心网络HPN以及多任务通信调度方法Crux解决上述3个挑战.HPN通过引入了一种2层、双平面(dual-plane)的网络架构,不但能够在一个Pod内高速互联15000个GPU,还能做到适用大模型训练的精准选路(解决挑战1).此外,HPN提出了一种新型的去堆叠双ToR(top-of-rack)设计来替代传统数据中心网络的单ToR交换机连接方式,根本性地避免了单点失效可靠性风险(部分解决挑战3).针对挑战2,Crux通过对GPU利用率优化问题的建模与证明,将该NP完全问题近似成GPU强度相关的流量调度问题.随后,Crux提出了一个方法优先处理具有高GPU计算强度的任务流,从而极大降低了多任务的通信竞争,优化了GPU利用率.与相关工作对比,Crux可以将GPU利用率提高多达23个百分点.HPN和Crux均已在阿里云生产环境规模化部署超过8个月,后续会持续演进迭代.在此基础上,进一步展望了大模型训练与推理领域可能的研究方向,为后续工作提供指导性建议.展开更多
基金National Natural Science Foundation of China(82274265 and 82274588)Hunan University of Traditional Chinese Medicine Research Unveiled Marshal Programs(2022XJJB003).
文摘Eye diagnosis is a method for inspecting systemic diseases and syndromes by observing the eyes.With the development of intelligent diagnosis in traditional Chinese medicine(TCM);artificial intelligence(AI)can improve the accuracy and efficiency of eye diagnosis.However;the research on intelligent eye diagnosis still faces many challenges;including the lack of standardized and precisely labeled data;multi-modal information analysis;and artificial in-telligence models for syndrome differentiation.The widespread application of AI models in medicine provides new insights and opportunities for the research of eye diagnosis intelli-gence.This study elaborates on the three key technologies of AI models in the intelligent ap-plication of TCM eye diagnosis;and explores the implications for the research of eye diagno-sis intelligence.First;a database concerning eye diagnosis was established based on self-su-pervised learning so as to solve the issues related to the lack of standardized and precisely la-beled data.Next;the cross-modal understanding and generation of deep neural network models to address the problem of lacking multi-modal information analysis.Last;the build-ing of data-driven models for eye diagnosis to tackle the issue of the absence of syndrome dif-ferentiation models.In summary;research on intelligent eye diagnosis has great potential to be applied the surge of AI model applications.
基金supported by the Ministry of Education’s 2022 Second Batch of Industry-university Cooperation Collaborative Education Project"PaddlePaddle Artificial Intelligence Education Innovation Center Practice Base"(Grant No.220700001065953,220700001121532)the Ministry of Education’s Second Batch of Supply-demand Docking Employment Education Project"University of Electronic Science and Technology of China-Baidu Online Network Technology(Beijing)AI Technology Talent Training Project"(Grant No.20230103592)。
文摘This paper takes the school-enterprise cooperation between the University of Electronic Science and Technology of China(UESTC)and Baidu(China)Co.,Ltd.as an example to build the Paddle Paddle(Sichuan)AI Education Innovation Center by enjoying the best of both UESTC and Baidu(China),and cooperating with 16 universities in Sichuan Province.With the support of this center,both the school and the enterprise successfully built a school-enterprise collaborative AI innovative talent training model,which mainly serves the universities,industries,and districts.Furthermore,this training model is able to facilitate the update of the industrial intelligence and the regional economic development in southwest China,and also provide a reference for deep integration of school-enterprise collaboration and the cultivation of innovative talents in electronic information fields.
文摘针对边缘计算环境下人工智能(Artificial Intelligence,AI)模型训练效率低下的问题,提出了基于边缘云计算的混合并行训练框架(Edge-Cloud based Hybrid Parallel Training Framework,ECHPT).ECHPT能在终端设备、边缘服务器和云计算中心之间实现AI模型和数据样本的自适应调度.ECHPT将模型计算和数据任务调度问题建模为训练时间最小化的优化问题,设计了调度算法对优化问题进行求解.实现了由设备、边缘服务器和云服务器组成的硬件原型.实验结果表明,与现有框架相比,ECHPT可以有效缩短AI模型的训练时间.
文摘拥有千亿级别参数的大语言模型(large language model,LLM)已为今天的人工智能和云服务带来了巨大的技术和商业变革.然而,大模型训练与传统的通用云计算(例如,亚马逊EC2弹性计算服务)之间存在较多根本性的网络行为差异,从而带来了很多新的挑战,主要包括流量模式差异造成负载难均衡(挑战1)、多训练任务通信竞争影响GPU利用率(挑战2),以及对网络故障的高敏感性(挑战3)等.因此,为通用云计算设计的数据中心网络技术(例如,网络架构、选路方法、流量调度,以及可靠性保障方法等)已不适合今天的大模型训练,这要求专门为大模型训练设计新型的数据中心网络以及配套的技术方案.介绍了阿里云专门为大模型训练设计的数据中心网络HPN以及多任务通信调度方法Crux解决上述3个挑战.HPN通过引入了一种2层、双平面(dual-plane)的网络架构,不但能够在一个Pod内高速互联15000个GPU,还能做到适用大模型训练的精准选路(解决挑战1).此外,HPN提出了一种新型的去堆叠双ToR(top-of-rack)设计来替代传统数据中心网络的单ToR交换机连接方式,根本性地避免了单点失效可靠性风险(部分解决挑战3).针对挑战2,Crux通过对GPU利用率优化问题的建模与证明,将该NP完全问题近似成GPU强度相关的流量调度问题.随后,Crux提出了一个方法优先处理具有高GPU计算强度的任务流,从而极大降低了多任务的通信竞争,优化了GPU利用率.与相关工作对比,Crux可以将GPU利用率提高多达23个百分点.HPN和Crux均已在阿里云生产环境规模化部署超过8个月,后续会持续演进迭代.在此基础上,进一步展望了大模型训练与推理领域可能的研究方向,为后续工作提供指导性建议.