算法创新Lab_研究方向

华为云算法创新Lab

Algorithm Powers Innovation

算法驱动创新

Algorithm Powers Innovation

算法驱动创新

研究方向

算法创新Lab秉承开发共赢的理念，愿意与学术界和工业界各位精英就以下研究方向（或历史方向）合作，如有意向请联系shijia@huawei.com

软硬件结合的高性能算子
大模型训推加速技术
新型大模型设计与训练

{"header":{"state":0,"promotionTag":"","style":"normal","mark":"","productIcon":"","titlePc":"软硬件结合的高性能算子","titleMb":"软硬件结合的高性能算子","titleLink":"","rowTabTitle":"软硬件结合的高性能算子","tip":"","subtitlePc":"","subtitleMb":"","smartLink":"","smartText":"","tags":[],"productImg":"","productVideo":"","productImgTitle":"","productImgDetail":"","productImgBtns":[],"tooltip":{"more":"","title":"","detail":"","images":""},"titleHasTooltip":false},"description":{"specs":{"defaultShowLength":1,"items":[{"title":"","detail":"算子是计算任务中最基础的执行单元，直接决定了整个系统或模型的效率和性能，本方向通过深入分析昇腾硬件特性，设计并优化适配昇腾硬件平台的高性能算子，使用并行计算优化、访存优化、多算子融合，算子自动化等技术实现硬件资源的充分利用，提升大模型整体运行性能。"}]},"advantages":{"list":[{"items":[{"content":"高性能算子优化","iconUrl":"","tooltip":"通过RoofLine分析技术，结合硬件特性，针对单个算子设计包括循环优化、访存优化与计算单元并行优化等算法，在保证计算精度的前提下，降低算子计算延迟，支撑大模型的高效训练与推理"},{"content":"多算子融合","iconUrl":"","tooltip":"通过模型计算图与算子计算图两个层次的分析与修改，将多个独立算子或计算逻辑合并为单一算子或并行计算的逻辑，提高模型的训推效率"}],"title":""}]},"links":{"title":"","items":[]},"image":{"url":"","description":""}},"collapse":{"title":"","subtitle":"","sellingPoint":{"title":"","list":[]},"content":""},"price":{"currency":"","accuracy":2,"priceError":"计算中...","origin":"","originTemplate":"","current":"","currentTemplate":"","currentStyle":"red","unit":"","tooltip":""},"button":{"mainBtnCheck":true,"mainBtnText":" ","mainBtnHref":"","mainBtnHrefMb":"","initDomainMainBtnText":"请先输入域名","subBtnText":"","subBtnHref":""},"flashSale":{"hasFlashSale":false,"prId":"","flashSaleTime":["09:00:00","23:59:59"],"beforeFlashSaleMainBtnText":"秒杀未开始","afterFlashSaleMainBtnText":"秒杀已结束","runoutFlashSaleMainBtnText":"已售罄","runoutFlashSaleMainBtnHref":"","flashSaleProgressTemplate":"已抢购${progress}%","runoutFlashSaleMainBtnIsBlank":false},"priceDetail":{"show":false,"type":"discount","promotionId":"","coupons":[],"title":"优惠明细","priceTitle":"券后预估到手","priceDetailLeftDescription":"原价","priceDetailRightDescription":"活动优惠","explainTitle":"活动优惠说明：","explain":"","footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准"},"promotion":{"packageUrl":"","synchronizedOptions":["数据中心","时长"],"selectAtLeast":1,"productSort":false,"productConfigs":[]},"verification":{"userTypeConfig":{"scope":"all","title":"温馨提示","message":"","messageMb":"","guidances":[]},"retCodeConfigs":[]},"coupon":{"title":"优惠明细","coupons":[],"footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准","subtitle":"活动优惠说明：","hasCoupon":false,"expiration":"有效期至：6月30日24:00","priceTitle":"券后预估到手","description":"商品代金券","promotionId":"","couponDetailTemplate":"满${threshold}减${amount}","couponPriceDescription":"代金券","originPriceDescription":"官网价"}}

软硬件结合的高性能算子

算子是计算任务中最基础的执行单元，直接决定了整个系统或模型的效率和性能，本方向通过深入分析昇腾硬件特性，设计并优化适配昇腾硬件平台的高性能算子，使用并行计算优化、访存优化、多算子融合，算子自动化等技术实现硬件资源的充分利用，提升大模型整体运行性能。

{"header":{"state":0,"promotionTag":"","style":"normal","mark":"","productIcon":"","titlePc":"大模型训推加速技术","titleMb":"大模型训推加速技术","titleLink":"","rowTabTitle":"大模型训推加速技术","tip":"","subtitlePc":"","subtitleMb":"","smartLink":"","smartText":"","tags":[],"productImg":"","productVideo":"","productImgTitle":"","productImgDetail":"","productImgBtns":[],"tooltip":{"more":"","title":"","detail":"","images":""},"titleHasTooltip":false},"description":{"specs":{"defaultShowLength":1,"items":[{"title":"","detail":"大模型训练和推理加速技术是解决计算资源瓶颈、降低计算成本、缩短训练时间、降低部署门槛的关键，我们希望通过研究大模型分布式训推策略，改进训练和推理框架，利用先进的并行计算，模型压缩与量化等技术，降低训练和推理成本，提升吞吐量。"}]},"advantages":{"list":[{"items":[{"content":"大模型训推框架优化","iconUrl":"","tooltip":"针对大规模深度学习模型的训练和推理过程，通过分布式训练加速、混合精度计算、数据加载优化、推理延迟降低等技术，旨在减少计算资源消耗、提高系统吞吐和稳定性，并支持大规模模型的高效部署和应用。"},{"content":"大模型分布式并行算法","iconUrl":"","tooltip":"研究高维TP，Dualpipe，异步数据并行等新型分布式计算技术，将大规模模型的训练或推理任务分配到多个计算节点上，解决单机资源不足的问题。优化通信开销和负载均衡，加速训练过程，支持超大规模模型的高效训练和推理。"},{"iconUrl":"","content":"模型压缩与量化","tooltip":"通过剪枝、量化、知识蒸馏等技术，减少模型参数和计算复杂度，降低推理时的内存占用和计算开销，同时保持较高的精度。"}],"title":""}]},"links":{"title":"","items":[]},"image":{"url":"","description":""}},"collapse":{"title":"","subtitle":"","sellingPoint":{"title":"","list":[]},"content":""},"price":{"currency":"","accuracy":2,"priceError":"","origin":"","originTemplate":"","current":"","currentTemplate":"","currentStyle":"red","unit":"","tooltip":""},"button":{"mainBtnCheck":true,"mainBtnText":" ","mainBtnHref":"","mainBtnHrefMb":"","initDomainMainBtnText":"请先输入域名","subBtnText":"","subBtnHref":""},"flashSale":{"hasFlashSale":false,"prId":"","flashSaleTime":["09:00:00","23:59:59"],"beforeFlashSaleMainBtnText":"秒杀未开始","afterFlashSaleMainBtnText":"秒杀已结束","runoutFlashSaleMainBtnText":"已售罄","runoutFlashSaleMainBtnHref":"","flashSaleProgressTemplate":"已抢购${progress}%","runoutFlashSaleMainBtnIsBlank":false},"priceDetail":{"show":false,"type":"discount","promotionId":"","coupons":[],"title":"优惠明细","priceTitle":"券后预估到手","priceDetailLeftDescription":"原价","priceDetailRightDescription":"活动优惠","explainTitle":"活动优惠说明：","explain":"","footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准"},"promotion":{"packageUrl":"","synchronizedOptions":["数据中心","时长"],"selectAtLeast":1,"productSort":false,"productConfigs":[]},"verification":{"userTypeConfig":{"scope":"all","title":"温馨提示","message":"","messageMb":"","guidances":[]},"retCodeConfigs":[]},"coupon":{"title":"优惠明细","coupons":[],"footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准","subtitle":"活动优惠说明：","hasCoupon":false,"expiration":"有效期至：6月30日24:00","priceTitle":"券后预估到手","description":"商品代金券","promotionId":"","couponDetailTemplate":"满${threshold}减${amount}","couponPriceDescription":"代金券","originPriceDescription":"官网价"}}

大模型训推加速技术

大模型训练和推理加速技术是解决计算资源瓶颈、降低计算成本、缩短训练时间、降低部署门槛的关键，我们希望通过研究大模型分布式训推策略，改进训练和推理框架，利用先进的并行计算，模型压缩与量化等技术，降低训练和推理成本，提升吞吐量。

{"header":{"state":0,"promotionTag":"","style":"normal","mark":"","productIcon":"","titlePc":"新型大模型设计与训练","titleMb":"新型大模型设计与训练","titleLink":"","rowTabTitle":"新型大模型设计与训练","tip":"","subtitlePc":"","subtitleMb":"","smartLink":"","smartText":"","tags":[],"productImg":"","productVideo":"","productImgTitle":"","productImgDetail":"","productImgBtns":[],"tooltip":{"more":"","title":"","detail":"","images":""},"titleHasTooltip":false},"description":{"specs":{"defaultShowLength":1,"items":[{"title":"","detail":"大模型结构设计是人工智能技术的核心，决定了模型的性能、效率和应用潜力。通过不断创新和优化结构设计，可以推动大模型在更广泛的领域中发挥作用。本方向探索如MLA、NSA、MOBA等新型模型架构及其预训练、后训练、压缩等技术，提升大模型在复杂场景下的的准确性、泛化能力和效率。"}]},"advantages":{"list":[{"items":[{"content":"新型大模型架构设计","iconUrl":"","tooltip":"创新性提出类似MLA,NSA,Moba等新型模型结构，在极大减少训推成本的同时能在较少程度或者不影响原始模型的性能效果。"},{"content":"大模型后训练技术","iconUrl":"","tooltip":"参数高效微调（如LoRA、Adapter）降低训练成本；指令精调（Instruction Tuning）提升任务泛化能力；人类反馈强化学习（RLHF）对齐人类价值观；知识增强通过外部数据注入领域知识。"}],"title":""}]},"links":{"title":"","items":[]},"image":{"url":"","description":""}},"collapse":{"title":"","subtitle":"","sellingPoint":{"title":"","list":[]},"content":""},"price":{"currency":"","accuracy":2,"priceError":"","origin":"","originTemplate":"","current":"","currentTemplate":"","currentStyle":"red","unit":"","tooltip":""},"button":{"mainBtnCheck":true,"mainBtnText":" ","mainBtnHref":"","mainBtnHrefMb":"","initDomainMainBtnText":"请先输入域名","subBtnText":"","subBtnHref":""},"flashSale":{"hasFlashSale":false,"prId":"","flashSaleTime":["09:00:00","23:59:59"],"beforeFlashSaleMainBtnText":"秒杀未开始","afterFlashSaleMainBtnText":"秒杀已结束","runoutFlashSaleMainBtnText":"已售罄","runoutFlashSaleMainBtnHref":"","flashSaleProgressTemplate":"已抢购${progress}%","runoutFlashSaleMainBtnIsBlank":false},"priceDetail":{"show":false,"type":"discount","promotionId":"","coupons":[],"title":"优惠明细","priceTitle":"券后预估到手","priceDetailLeftDescription":"原价","priceDetailRightDescription":"活动优惠","explainTitle":"活动优惠说明：","explain":"","footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准"},"promotion":{"packageUrl":"","synchronizedOptions":["数据中心","时长"],"selectAtLeast":1,"productSort":false,"productConfigs":[]},"verification":{"userTypeConfig":{"scope":"all","title":"温馨提示","message":"","messageMb":"","guidances":[]},"retCodeConfigs":[]},"coupon":{"title":"优惠明细","coupons":[],"footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准","subtitle":"活动优惠说明：","hasCoupon":false,"expiration":"有效期至：6月30日24:00","priceTitle":"券后预估到手","description":"商品代金券","promotionId":"","couponDetailTemplate":"满${threshold}减${amount}","couponPriceDescription":"代金券","originPriceDescription":"官网价"}}

新型大模型设计与训练

大模型结构设计是人工智能技术的核心，决定了模型的性能、效率和应用潜力。通过不断创新和优化结构设计，可以推动大模型在更广泛的领域中发挥作用。本方向探索如MLA、NSA、MOBA等新型模型架构及其预训练、后训练、压缩等技术，提升大模型在复杂场景下的的准确性、泛化能力和效率。

AI智能体
面向AI的新型算力基础设施的调度优化
云存储算法

{"header":{"state":0,"promotionTag":"","style":"normal","mark":"","productIcon":"","titlePc":"AI智能体","titleMb":"AI智能体","titleLink":"","rowTabTitle":"AI智能体","tip":"","subtitlePc":"","subtitleMb":"","smartLink":"","smartText":"","tags":[],"productImg":"","productVideo":"","productImgTitle":"","productImgDetail":"","productImgBtns":[],"tooltip":{"more":"","title":"","detail":"","images":""},"titleHasTooltip":false},"description":{"specs":{"defaultShowLength":1,"items":[{"title":"","detail":"通过使用大模型和AI算法来模拟人类智能和行为的软件程序。它们需要理解自然语言、感知环境信息、自主决策规划、执行指令、与人类或环境进行自然而有效的交互，并不断反馈学习改善性能。因此，AI智能体通常基于GPT/Llama/盘古等大语言模型，通过分析、记忆、规划、动作等模块构建更智能的系统。所设计的算法包括但不限于自然语言处理、多模态学习、强化学习、深度学习等。"}]},"advantages":{"list":[{"items":[{"content":"算子自动化开发","iconUrl":"","tooltip":"通过数学抽象、编译优化与 AI Agent 等技术，针对特定的硬件架构，实现算子的快速生成与验证，提升算子开发效率。"},{"content":"自动工作流生成","iconUrl":"","tooltip":"AIAgent自动工作流生成技术通过自然语言处理（NLP）解析用户需求，结合知识图谱与深度学习模型，动态构建任务逻辑链,减少人工预定义的工作量或者降低学习门槛。"}],"title":""}]},"links":{"title":"","items":[]},"image":{"url":"","description":""}},"collapse":{"title":"","subtitle":"","sellingPoint":{"title":"","list":[]},"content":""},"price":{"currency":"","accuracy":2,"priceError":"计算中...","origin":"","originTemplate":"","current":"","currentTemplate":"","currentStyle":"red","unit":"","tooltip":""},"button":{"mainBtnCheck":true,"mainBtnText":" ","mainBtnHref":"","mainBtnHrefMb":"","initDomainMainBtnText":"请先输入域名","subBtnText":"","subBtnHref":""},"flashSale":{"hasFlashSale":false,"prId":"","flashSaleTime":["09:00:00","23:59:59"],"beforeFlashSaleMainBtnText":"秒杀未开始","afterFlashSaleMainBtnText":"秒杀已结束","runoutFlashSaleMainBtnText":"已售罄","runoutFlashSaleMainBtnHref":"","flashSaleProgressTemplate":"已抢购${progress}%","runoutFlashSaleMainBtnIsBlank":false},"priceDetail":{"show":false,"type":"discount","promotionId":"","coupons":[],"title":"优惠明细","priceTitle":"券后预估到手","priceDetailLeftDescription":"原价","priceDetailRightDescription":"活动优惠","explainTitle":"活动优惠说明：","explain":"","footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准"},"promotion":{"packageUrl":"","synchronizedOptions":["数据中心","时长"],"selectAtLeast":1,"productSort":false,"productConfigs":[]},"verification":{"userTypeConfig":{"scope":"all","title":"温馨提示","message":"","messageMb":"","guidances":[]},"retCodeConfigs":[]},"coupon":{"title":"优惠明细","coupons":[],"footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准","subtitle":"活动优惠说明：","hasCoupon":false,"expiration":"有效期至：6月30日24:00","priceTitle":"券后预估到手","description":"商品代金券","promotionId":"","couponDetailTemplate":"满${threshold}减${amount}","couponPriceDescription":"代金券","originPriceDescription":"官网价"}}

AI智能体

通过使用大模型和AI算法来模拟人类智能和行为的软件程序。它们需要理解自然语言、感知环境信息、自主决策规划、执行指令、与人类或环境进行自然而有效的交互，并不断反馈学习改善性能。因此，AI智能体通常基于GPT/Llama/盘古等大语言模型，通过分析、记忆、规划、动作等模块构建更智能的系统。所设计的算法包括但不限于自然语言处理、多模态学习、强化学习、深度学习等。

{"header":{"state":0,"promotionTag":"","style":"normal","mark":"","productIcon":"","titlePc":"面向AI的新型算力基础设施的调度优化","titleMb":"面向AI的新型算力基础设施的调度优化","titleLink":"","rowTabTitle":"面向AI的新型算力基础设施的调度优化","tip":"","subtitlePc":"","subtitleMb":"","smartLink":"","smartText":"","tags":[],"productImg":"","productVideo":"","productImgTitle":"","productImgDetail":"","productImgBtns":[],"tooltip":{"more":"","title":"","detail":"","images":""},"titleHasTooltip":false},"description":{"specs":{"defaultShowLength":1,"items":[{"title":"","detail":"为了打造ChatGPT/AIGC时代的云竞争力，需要最大化利用新型AI算力资源，提高AI算法的执行效率与性能。随着大模型以及AI应用的普及，用户对算力的消耗与需求逐渐从传统的CPU为主过渡到GPU/NPU等异构算力为主。针对新型AI算力基础设施，需要综合设计分布式并行计算、异构资源分配调度、多任务调度、负载均衡调度等算法。"}]},"advantages":{"list":[{"items":[{"content":"多业务混部","iconUrl":"","tooltip":"在AI集群中，存在多个任务与多个业务同时部署的场景，需要探索计算、存储、网络资源的解耦复用、训推混部、动态扩缩容，多时间片复用、抢占式调度等技术，提高资源利用率。"},{"content":"异构算力编排调度","iconUrl":"","tooltip":"综合考虑CPU/GPU/NPU等多样化算力的性能、成本、可用性、负载均衡、集群组网结构等因素，选择最合适的算力进行编排和调度，充分调度使用多资源池的计算能力，提高计算任务的效率和性能。"}],"title":""}]},"links":{"title":"","items":[]},"image":{"url":"","description":""}},"collapse":{"title":"","subtitle":"","sellingPoint":{"title":"","list":[]},"content":""},"price":{"currency":"","accuracy":2,"priceError":"","origin":"","originTemplate":"","current":"","currentTemplate":"","currentStyle":"red","unit":"","tooltip":""},"button":{"mainBtnCheck":true,"mainBtnText":" ","mainBtnHref":"","mainBtnHrefMb":"","initDomainMainBtnText":"请先输入域名","subBtnText":"","subBtnHref":""},"flashSale":{"hasFlashSale":false,"prId":"","flashSaleTime":["09:00:00","23:59:59"],"beforeFlashSaleMainBtnText":"秒杀未开始","afterFlashSaleMainBtnText":"秒杀已结束","runoutFlashSaleMainBtnText":"已售罄","runoutFlashSaleMainBtnHref":"","flashSaleProgressTemplate":"已抢购${progress}%","runoutFlashSaleMainBtnIsBlank":false},"priceDetail":{"show":false,"type":"discount","promotionId":"","coupons":[],"title":"优惠明细","priceTitle":"券后预估到手","priceDetailLeftDescription":"原价","priceDetailRightDescription":"活动优惠","explainTitle":"活动优惠说明：","explain":"","footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准"},"promotion":{"packageUrl":"","synchronizedOptions":["数据中心","时长"],"selectAtLeast":1,"productSort":false,"productConfigs":[]},"verification":{"userTypeConfig":{"scope":"all","title":"温馨提示","message":"","messageMb":"","guidances":[]},"retCodeConfigs":[]},"coupon":{"title":"优惠明细","coupons":[],"footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准","subtitle":"活动优惠说明：","hasCoupon":false,"expiration":"有效期至：6月30日24:00","priceTitle":"券后预估到手","description":"商品代金券","promotionId":"","couponDetailTemplate":"满${threshold}减${amount}","couponPriceDescription":"代金券","originPriceDescription":"官网价"}}

面向AI的新型算力基础设施的调度优化

为了打造ChatGPT/AIGC时代的云竞争力，需要最大化利用新型AI算力资源，提高AI算法的执行效率与性能。随着大模型以及AI应用的普及，用户对算力的消耗与需求逐渐从传统的CPU为主过渡到GPU/NPU等异构算力为主。针对新型AI算力基础设施，需要综合设计分布式并行计算、异构资源分配调度、多任务调度、负载均衡调度等算法。

{"header":{"state":0,"promotionTag":"","style":"normal","mark":"","productIcon":"","titlePc":"云存储算法","titleMb":"云存储算法","titleLink":"","rowTabTitle":"云存储算法","tip":"","subtitlePc":"","subtitleMb":"","smartLink":"","smartText":"","tags":[],"productImg":"","productVideo":"","productImgTitle":"","productImgDetail":"","productImgBtns":[],"tooltip":{"more":"","title":"","detail":"","images":""},"titleHasTooltip":false},"description":{"specs":{"defaultShowLength":1,"items":[{"title":"","detail":"云存储服务是华为云的三大基础服务之一，其体量大，存有EB级别的数据，并且还在快速增长中。存储服务会用到存储、网络、计算等多种资源，并且全球化部署，复杂度高。同时数据存储对可靠性有非常高的要求。其对性能和高可靠的极致追求给算法研究留下很大空间。我们希望通过数据驱动的算法优化，在为客户提供高可靠性的存储服务的同时，实现资源的高效利用，降本增效。"}]},"advantages":{"list":[{"items":[{"content":"数据资源优化","iconUrl":"","tooltip":"运用大数据分析与人工智能技术，依托跨集群、跨区域的云存储资源调度能力，构建涵盖资源数据本体的全生命周期模型，实现数据和机器学习驱动的海量数据挖掘与智能决策，从性能、成本与可靠性维度，提升云存储资源利用效率，推动云存储技术的持续突破与行业应用。"},{"content":"存储基础算法","iconUrl":"","tooltip":"结合传统和新兴存储设备，深耕数据分布与负载均衡、数据冗余与容错、缓存替换与预取、数据压缩与去重等云存储核心基础算法，支撑EB级别的云存储系统高效、可靠、安全的运行，推动云存储核心技术的持续创新与发展。"}],"title":""}]},"links":{"title":"","items":[]},"image":{"url":"","description":""}},"collapse":{"title":"","subtitle":"","sellingPoint":{"title":"","list":[]},"content":""},"price":{"currency":"","accuracy":2,"priceError":"","origin":"","originTemplate":"","current":"","currentTemplate":"","currentStyle":"red","unit":"","tooltip":""},"button":{"mainBtnCheck":true,"mainBtnText":" ","mainBtnHref":"","mainBtnHrefMb":"","initDomainMainBtnText":"请先输入域名","subBtnText":"","subBtnHref":""},"flashSale":{"hasFlashSale":false,"prId":"","flashSaleTime":["09:00:00","23:59:59"],"beforeFlashSaleMainBtnText":"秒杀未开始","afterFlashSaleMainBtnText":"秒杀已结束","runoutFlashSaleMainBtnText":"已售罄","runoutFlashSaleMainBtnHref":"","flashSaleProgressTemplate":"已抢购${progress}%","runoutFlashSaleMainBtnIsBlank":false},"priceDetail":{"show":false,"type":"discount","promotionId":"","coupons":[],"title":"优惠明细","priceTitle":"券后预估到手","priceDetailLeftDescription":"原价","priceDetailRightDescription":"活动优惠","explainTitle":"活动优惠说明：","explain":"","footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准"},"promotion":{"packageUrl":"","synchronizedOptions":["数据中心","时长"],"selectAtLeast":1,"productSort":false,"productConfigs":[]},"verification":{"userTypeConfig":{"scope":"all","title":"温馨提示","message":"","messageMb":"","guidances":[]},"retCodeConfigs":[]},"coupon":{"title":"优惠明细","coupons":[],"footnote":"* 以上优惠仅为初步预估，不代表最终价格，具体结算金额请以支付页实际抵扣金额为准","subtitle":"活动优惠说明：","hasCoupon":false,"expiration":"有效期至：6月30日24:00","priceTitle":"券后预估到手","description":"商品代金券","promotionId":"","couponDetailTemplate":"满${threshold}减${amount}","couponPriceDescription":"代金券","originPriceDescription":"官网价"}}

云存储算法

云存储服务是华为云的三大基础服务之一，其体量大，存有EB级别的数据，并且还在快速增长中。存储服务会用到存储、网络、计算等多种资源，并且全球化部署，复杂度高。同时数据存储对可靠性有非常高的要求。其对性能和高可靠的极致追求给算法研究留下很大空间。我们希望通过数据驱动的算法优化，在为客户提供高可靠性的存储服务的同时，实现资源的高效利用，降本增效。