[{"data":1,"prerenderedAt":503},["ShallowReactive",2],{"content-query-pQyjHFMhmY":3},{"_path":4,"_dir":5,"_draft":6,"_partial":6,"_locale":7,"title":8,"description":9,"date":10,"cover":11,"type":12,"body":13,"_type":497,"_id":498,"_source":499,"_file":500,"_stem":501,"_extension":502},"/version-updates/zh/246","zh",false,"","MindSpore新版本v0.7.0-beta发布！","大家好，经历了八月如火如荼的开发之后，MindSpore带来了本月的更新——v0.7.0-beta（后简称v0.7）版本。端侧大更新，四个SIG/WG成立，首个开源ISV对接完成！","2020-09-01","https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/21/189b4abca06a4b8ca12a41c7a1f1b5bf.png","version-updates",{"type":14,"children":15,"toc":494},"root",[16,24,30,36,41,49,54,59,66,71,82,87,92,97,104,109,118,123,128,133,138,143,150,154,163,168,173,178,183,188,193,200,205,210,215,222,227,232,237,242,249,254,265,276,286,291,300,308,313,318,323,328,333,338,343,348,353,358,363,368,373,378,383,394,405,416,427,439,450,455,466,477,482,487],{"type":17,"tag":18,"props":19,"children":21},"element","h1",{"id":20},"mindspore新版本v070-beta发布",[22],{"type":23,"value":8},"text",{"type":17,"tag":18,"props":25,"children":27},{"id":26},"端侧大更新四个sigwg成立首个开源isv对接完成",[28],{"type":23,"value":29},"端侧大更新，四个SIG/WG成立，首个开源ISV对接完成！",{"type":17,"tag":31,"props":32,"children":33},"p",{},[34],{"type":23,"value":35},"大家好，经历了八月如火如荼的开发之后，MindSpore带来了本月的更新——v0.7.0-beta（后简称v0.7）版本。老规矩先上视频，了解MindSpore v0.7版本最新特性。",{"type":17,"tag":31,"props":37,"children":38},{},[39],{"type":23,"value":40},"按照惯例，接下来给大家解读MindSpore又有了什么新的关键特性~",{"type":17,"tag":31,"props":42,"children":43},{},[44],{"type":17,"tag":45,"props":46,"children":48},"img",{"alt":7,"src":47},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/7ff5395a3f0b42908aef2f08484b0cbf.jpg",[],{"type":17,"tag":31,"props":50,"children":51},{},[52],{"type":23,"value":53},"MindSpore Lite",{"type":17,"tag":31,"props":55,"children":56},{},[57],{"type":23,"value":58},"本月最大的惊喜是MindSpore发布了针对手机及IoT场景，端到端的解决方案——MindSpore Lite。从模型开始到最后的编译发布，提供了包括转换工具、Runtime等工具链，帮助用户在手机以及后续的IoT设备上使能MindSpore的AI能力。",{"type":17,"tag":31,"props":60,"children":61},{},[62],{"type":17,"tag":45,"props":63,"children":65},{"alt":7,"src":64},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/f0c76399237343e893d5d08c09ac2c8f.jpg",[],{"type":17,"tag":31,"props":67,"children":68},{},[69],{"type":23,"value":70},"了解MindSpore Lite：",{"type":17,"tag":31,"props":72,"children":73},{},[74],{"type":17,"tag":75,"props":76,"children":80},"a",{"href":77,"rel":78},"https://www.mindspore.cn/lite",[79],"nofollow",[81],{"type":23,"value":77},{"type":17,"tag":31,"props":83,"children":84},{},[85],{"type":23,"value":86},"Fuzz测试工具",{"type":17,"tag":31,"props":88,"children":89},{},[90],{"type":23,"value":91},"在实际使用过程中因为一些原因，譬如使用图像分类应用时，当镜头模糊、光照不足、下雨大雾、拍摄角度等等情况下，实际输入的数据与训练模型时的标准数据集有一些不同。在这种情况下，模型可能出现模型精度答复下降。而在某些极端情况下，可能出现未知的推理失误，引入安全性问题。为了更好的评估和提高模型的鲁棒性，同时尽可能激活神经元，减少未知的安全风险，MindArmour提供了Fuzz（模糊）测试工具。",{"type":17,"tag":31,"props":93,"children":94},{},[95],{"type":23,"value":96},"测试过程中，工具的Fuzzer Moduler（Fuzz模块）可以对数据集进行增强，如色差、翻转、模糊等，自动化产生测试样例，如下图右侧。在结果输出方面，用户可以看到模糊测试结果，评价了图像的覆盖率指标以及精度下降。下图左侧可以看出在模型出现扰动的情况下，模型精度（Accuracy）下降程度以及神经元激活程度（Coverage Rate）。",{"type":17,"tag":31,"props":98,"children":99},{},[100],{"type":17,"tag":45,"props":101,"children":103},{"alt":7,"src":102},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/5db869906412417194bbaec3ebb096ca.jpg",[],{"type":17,"tag":31,"props":105,"children":106},{},[107],{"type":23,"value":108},"使用教程：",{"type":17,"tag":31,"props":110,"children":111},{},[112],{"type":17,"tag":75,"props":113,"children":116},{"href":114,"rel":115},"https://www.mindspore.cn/mindarmour/docs/zh-CN/r1.7/security_and_privacy.html?highlight=Fuzzer",[79],[117],{"type":23,"value":114},{"type":17,"tag":31,"props":119,"children":120},{},[121],{"type":23,"value":122},"深度概率编程",{"type":17,"tag":31,"props":124,"children":125},{},[126],{"type":23,"value":127},"深度概率编程融合了深度学习和贝叶斯建模，深度概率模型相比于传统的深度神经网络模型，引入了不确定性，提升了模型的鲁棒性和可解释性。",{"type":17,"tag":31,"props":129,"children":130},{},[131],{"type":23,"value":132},"MindSpore在v0.7版本中，提供了概率模型的高效编程库，可以与深度学习模型无缝融合，以面向不同需求的开发者。",{"type":17,"tag":31,"props":134,"children":135},{},[136],{"type":23,"value":137},"对于熟悉深度学习的同学，支持深度神经网络一键转贝叶斯神经网络；也支持调用封装好的接口，像实现深度神经网络一样实现贝叶斯神经网络和生成网络；并提供不确定性估计工具，方便同学评估模型和数据。",{"type":17,"tag":31,"props":139,"children":140},{},[141],{"type":23,"value":142},"对于熟悉概率学习的同学，其提供丰富、通用、模块化的贝叶斯建模和推断方法库，包括概率采样库、贝叶斯神经网络，概率推断算法等，帮助用户更加快速、简洁地进行概率编程。",{"type":17,"tag":31,"props":144,"children":145},{},[146],{"type":17,"tag":45,"props":147,"children":149},{"alt":7,"src":148},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/e1920a4d3ba54896a7f0241399597570.jpg",[],{"type":17,"tag":31,"props":151,"children":152},{},[153],{"type":23,"value":108},{"type":17,"tag":31,"props":155,"children":156},{},[157],{"type":17,"tag":75,"props":158,"children":161},{"href":159,"rel":160},"https://gitee.com/mindspore/mindspore/tree/r0.7/mindspore/nn/probability",[79],[162],{"type":23,"value":159},{"type":17,"tag":31,"props":164,"children":165},{},[166],{"type":23,"value":167},"自动数据增强",{"type":17,"tag":31,"props":169,"children":170},{},[171],{"type":23,"value":172},"MindSpore数据处理组件新增了自动数据增强的能力，支持用户在处理数据过程中灵活定制及调整数据增强操作。主要有以下两种特性：",{"type":17,"tag":31,"props":174,"children":175},{},[176],{"type":23,"value":177},"• 支持基于概率动态调整数据增强策略",{"type":17,"tag":31,"props":179,"children":180},{},[181],{"type":23,"value":182},"根据用户预置策略列表，每次随机选择一个策略，同时同一策略各数据增强算子有不同的概率，按照概率选择算子执行。",{"type":17,"tag":31,"props":184,"children":185},{},[186],{"type":23,"value":187},"• 支持用户根据pipeline上下文信息或训练结果动态调整数据增强策略",{"type":17,"tag":31,"props":189,"children":190},{},[191],{"type":23,"value":192},"支持batch、 epoch粒度调整数据增强策略。",{"type":17,"tag":31,"props":194,"children":195},{},[196],{"type":17,"tag":45,"props":197,"children":199},{"alt":7,"src":198},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/9aba6a456f004df5867ac62c85e8a21f.jpg",[],{"type":17,"tag":31,"props":201,"children":202},{},[203],{"type":23,"value":204},"—————————————————————————————————————————————————————————————————————————————————————",{"type":17,"tag":31,"props":206,"children":207},{},[208],{"type":23,"value":209},"MindSpore迎来首个开源ISV对接！",{"type":17,"tag":31,"props":211,"children":212},{},[213],{"type":23,"value":214},"Jina.ai，AI开源新势力",{"type":17,"tag":31,"props":216,"children":217},{},[218],{"type":17,"tag":45,"props":219,"children":221},{"alt":7,"src":220},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/9c15a0431fd54aac920973f03a1a2e6f.jpg",[],{"type":17,"tag":31,"props":223,"children":224},{},[225],{"type":23,"value":226},"Jina.ai是由MindSpore技术治理委员会成员肖涵博士主导的初创公司，致力于提供面向云原生的开源神经网络搜索能力。让我们看一下Jina如何描述他们对AI开源的看法：",{"type":17,"tag":31,"props":228,"children":229},{},[230],{"type":23,"value":231},"得益于深度学习框架和现成的预训练模型，端到端应用程序市场中有许多人可以做出重大贡献的机会。随着每天涌现出新的AI算法，如何跟上步伐？项目范围是什么？面对社区的要求时，如何维护项目？这些困难的问题需要一个明确的答案，比如在此前开发bert-as-service时，有很多要求将其扩展到AlBERT，DistilBERT，BioBERT等。我们优先考虑了适合我们的路线图的那些需求进行落地。最好的开源也许不是最闪亮的，但是可以持续生长的开源。",{"type":17,"tag":31,"props":233,"children":234},{},[235],{"type":23,"value":236},"可以看到Jina团队对开源开发有着非常成熟的思考，那么与MindSpore的对接是否属于适合Jina路线图的需求呢？",{"type":17,"tag":31,"props":238,"children":239},{},[240],{"type":23,"value":241},"Jina助力MindSpore的云原生野望",{"type":17,"tag":31,"props":243,"children":244},{},[245],{"type":17,"tag":45,"props":246,"children":248},{"alt":7,"src":247},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/83496936cc2f4030936579b1e496b364.jpg",[],{"type":17,"tag":31,"props":250,"children":251},{},[252],{"type":23,"value":253},"答案是肯定是！就在8月份，Jina社区团队完成了MindSpore框架的初步对接支持，成为了首个对接MindSpore应用集成商。Jina+MindSpore提供了业界少有的完整开源解决方案，Jina.ai将在近期推送详细的技术介绍博文，我们也会在第一时间转载推送，心急的同学可以先来关注几个主要的PR：",{"type":17,"tag":31,"props":255,"children":256},{},[257,259],{"type":23,"value":258},"•",{"type":17,"tag":75,"props":260,"children":263},{"href":261,"rel":262},"https://github.com/jina-ai/jina/pull/836",[79],[264],{"type":23,"value":261},{"type":17,"tag":31,"props":266,"children":267},{},[268,270],{"type":23,"value":269},"• ",{"type":17,"tag":75,"props":271,"children":274},{"href":272,"rel":273},"https://github.com/jina-ai/examples/pull/155",[79],[275],{"type":23,"value":272},{"type":17,"tag":31,"props":277,"children":278},{},[279,280],{"type":23,"value":269},{"type":17,"tag":75,"props":281,"children":284},{"href":282,"rel":283},"https://github.com/jina-ai/examples/issues/154",[79],[285],{"type":23,"value":282},{"type":17,"tag":31,"props":287,"children":288},{},[289],{"type":23,"value":290},"—————————————————————————————————————————————————————————————————————————————————",{"type":17,"tag":31,"props":292,"children":293},{},[294],{"type":17,"tag":295,"props":296,"children":297},"strong",{},[298],{"type":23,"value":299},"One More Thing：",{"type":17,"tag":31,"props":301,"children":302},{},[303],{"type":17,"tag":295,"props":304,"children":305},{},[306],{"type":23,"value":307},"MindSpore 8月新增了2个新的SIG和2个新的WG！",{"type":17,"tag":31,"props":309,"children":310},{},[311],{"type":23,"value":312},"什么是SIG和WG?",{"type":17,"tag":31,"props":314,"children":315},{},[316],{"type":23,"value":317},"特殊兴趣组（Special Interest Group，简称SIG）以及工作组（Working Group，简称WG）是开源社区开发中常见的项目组织形式。由于MindSpore社区采取开放治理架构[0]，除了把握大的发展方向的技术治理委员会（Technical Steering Committee，简称TSC）之外，日常的模块开发为了兼顾开放协作和迭代效率，通常通过SIG的形式由开发者自由组队，并通过推举SIG Lead的方式来产生模块开发的负责人，管理日常工作，有关注过MindSpore邮件列表的同学应该会留意到经常会有SIG负责人换届的公开邮件。工作组WG则主要是负责一些涉及到多个SIG的需求传递、设计讨论等工作。",{"type":17,"tag":31,"props":319,"children":320},{},[321],{"type":23,"value":322},"是不是SIG/WG的数目越多越好？",{"type":17,"tag":31,"props":324,"children":325},{},[326],{"type":23,"value":327},"当然不是！SIG/WG的创建总是为了应对具体的开发或协作的问题而产生的，一个开源社区的SIG/WG的数目，在客观上不说明任何社区健康度或者活跃度的问题。关键还是要看SIG/WG本身的内容和质量，比如在之前MindSpore社区开源后成立的第一个新SIG-AKG，就为开发者带来了图算融合这样非常黑魔法的特性。",{"type":17,"tag":31,"props":329,"children":330},{},[331],{"type":23,"value":332},"那么这次新成立的4个社区组织有什么特色？",{"type":17,"tag":31,"props":334,"children":335},{},[336],{"type":23,"value":337},"8月份社区迎来的这4个新组织都非常各具特色，我们将会在接下来通过4篇独立的文章分别介绍：",{"type":17,"tag":31,"props":339,"children":340},{},[341],{"type":23,"value":342},"SIG-MSLITE",{"type":17,"tag":31,"props":344,"children":345},{},[346],{"type":23,"value":347},"MindSpore社区8月份端侧特性大版本更更新，现在大家使用的华为手机中支撑HMS Core里面AI应用的MindSpore Lite模块，已经在社区所见即所得了！立项文档可参见[1]，我们将会在9月1日的公众号文章详细介绍SIG-MSLITE。",{"type":17,"tag":31,"props":349,"children":350},{},[351],{"type":23,"value":352},"SIG-DPP",{"type":17,"tag":31,"props":354,"children":355},{},[356],{"type":23,"value":357},"由清华大学开源项目珠算团队[2]带来的深度概率编程支持！通过开发MindSpore深度概率编程库，为广大开发者提供与深度学习无缝融合的高效贝叶斯学习编程库。立项文档可参见[3]，我们将会在9月2日的公众号文章详细介绍SIG-DPP。",{"type":17,"tag":31,"props":359,"children":360},{},[361],{"type":23,"value":362},"WG-MM",{"type":17,"tag":31,"props":364,"children":365},{},[366],{"type":23,"value":367},"由深圳湾实验室分子动力学团队带来的非常令人兴奋的需求性工作组。分子动力学如今被广泛应用于广泛应用于物理、化学、生物、材料、医学等领域，可以说是当之无愧的基础核心技术。这个新工作组将聚焦在MindSpore如何帮助加快深度分子模型的创新实现。立项文档可参见[4]，我们将会在9月3日的公众号文章详细介绍WG-MM",{"type":17,"tag":31,"props":369,"children":370},{},[371],{"type":23,"value":372},"WG-Research",{"type":17,"tag":31,"props":374,"children":375},{},[376],{"type":23,"value":377},"是由华为团队发起的基于MindSpore面向未来的十大课题，通过开源社区面向开发者征集好的想法，以及对开发进行必要的资助。立项文档可参见[5]，我们将会在9月4日的公众号文章详细介绍WG-Research里面的十大课题都有哪些。",{"type":17,"tag":31,"props":379,"children":380},{},[381],{"type":23,"value":382},"参考文献：",{"type":17,"tag":31,"props":384,"children":385},{},[386,388],{"type":23,"value":387},"[0]",{"type":17,"tag":75,"props":389,"children":392},{"href":390,"rel":391},"https://gitee.com/mindspore/community/blob/master/governance.md",[79],[393],{"type":23,"value":390},{"type":17,"tag":31,"props":395,"children":396},{},[397,399],{"type":23,"value":398},"[1]",{"type":17,"tag":75,"props":400,"children":403},{"href":401,"rel":402},"https://gitee.com/mindspore/community/blob/master/design/meps/mep-mslite/MEP-MSLITE.md",[79],[404],{"type":23,"value":401},{"type":17,"tag":31,"props":406,"children":407},{},[408,410],{"type":23,"value":409},"[2]",{"type":17,"tag":75,"props":411,"children":414},{"href":412,"rel":413},"https://github.com/thu-ml/zhusuan",[79],[415],{"type":23,"value":412},{"type":17,"tag":31,"props":417,"children":418},{},[419,421],{"type":23,"value":420},"[3]",{"type":17,"tag":75,"props":422,"children":425},{"href":423,"rel":424},"https://gitee.com/mindspore/mindspore/issues/I1RG6C",[79],[426],{"type":23,"value":423},{"type":17,"tag":31,"props":428,"children":429},{},[430,432],{"type":23,"value":431},"[4",{"type":17,"tag":75,"props":433,"children":436},{"href":434,"rel":435},"//gitee.com/mindspore/community/blob/master/design/meps/mep-mm/mep-mm",[79],[437],{"type":23,"value":438},"]https://gitee.com/mindspore/community/blob/master/design/meps/mep-mm/MEP-MM.md",{"type":17,"tag":31,"props":440,"children":441},{},[442,444],{"type":23,"value":443},"[5]",{"type":17,"tag":75,"props":445,"children":448},{"href":446,"rel":447},"https://gitee.com/mindspore/community/blob/master/design/meps/mep-research/MEP-RESEARCH.md",[79],[449],{"type":23,"value":446},{"type":17,"tag":31,"props":451,"children":452},{},[453],{"type":23,"value":454},"MindSpore官方资料",{"type":17,"tag":31,"props":456,"children":457},{},[458,460],{"type":23,"value":459},"GitHub:",{"type":17,"tag":75,"props":461,"children":464},{"href":462,"rel":463},"https://github.com/mindspore-ai/mindspore",[79],[465],{"type":23,"value":462},{"type":17,"tag":31,"props":467,"children":468},{},[469,471],{"type":23,"value":470},"Gitee:",{"type":17,"tag":75,"props":472,"children":475},{"href":473,"rel":474},"https://gitee.com/mindspore/mindspore",[79],[476],{"type":23,"value":473},{"type":17,"tag":31,"props":478,"children":479},{},[480],{"type":23,"value":481},"官方QQ群: 871543426",{"type":17,"tag":31,"props":483,"children":484},{},[485],{"type":23,"value":486},"扫描下方二维码关注",{"type":17,"tag":31,"props":488,"children":489},{},[490],{"type":17,"tag":45,"props":491,"children":493},{"alt":7,"src":492},"https://obs-mindspore-file.obs.cn-north-4.myhuaweicloud.com/file/2020/09/01/b9237d78a29e4c5da4f63904def8e57a.jpg",[],{"title":7,"searchDepth":495,"depth":495,"links":496},4,[],"markdown","content:version-updates:zh:246.md","content","version-updates/zh/246.md","version-updates/zh/246","md",1776506145036]