- 快召唤伙伴们来围观吧
- 微博 QQ QQ空间 贴吧
- 文档嵌入链接
- 复制
- 微信扫一扫分享
- 已成功复制到剪贴板
5 Jina:云原生开源神经搜索框架_王楠_20210314
议题简介:
随着信息规模的爆炸式增长和数据类型的日益丰富,基于符号的传统搜索逐渐无法满足用户的需求。得益于深度学习技术的发展,神经搜索系统应运而生。但是,在搭建和维护神经搜索系统的过程中,工程团队不仅需要具备分布式架构的经验,更需要熟悉多个软件框架和理解不同AI算法。针对这个痛点,Jina提供覆盖搜索全链路的一站式云原生开源解决方案。在本次报告中,我们将分享Jina的设计思想和主要特点,并展示如何使用Jina搭建神经搜索系统。
嘉宾简介:
王楠,博士,Jina AI联合创始人兼CTO。
专注于机器学习和深度学习算法在NLP和搜索领域的实际应用。作为开源神经搜索框架jina的核心贡献者,热衷于开源软件和云原生技术。
展开查看详情
1 .Jina: 云原生开源神经搜索框架 王楠 2021-3-20
2 .Jina AI Founded in Feb. 2020 Jina AI has quickly grown as one of the most high valued AI COSS start-ups in the world.
3 . Our team: Async & Distributed c ○ Broad aperture: Europe, Asia, NA ○ Product + Engineering > 70% ○ Ex Microsoft, Google, Tencent, Adobe, Zalando, Soundcloud
4 .About me • 2020-now, Jina AI • 2017-20, Tencent • 2015-17, Zalando SE • 2009-14, Ph.D. at Ruhr-Universität Bochum, Germany
5 .Neural Search Searching unstructured data using unstructured query. • The full-stack ownership • Covers all the steps from end to end. • Includes indexing and querying • Search AI in production • Powered by AI models • Use vector search engine • Perform single-/multi-/cross-modal search
6 .Neural Search Searching unstructured data using unstructured query. • Search image via image • Search video via video • Search audios via audio • Search items via sketch • Search PDF via image • Search audio via image • Search map via image • …
7 . Neural Search Key-Value index index index .pdf Feature Encoding Vector search extraction index Text Text BERT Extractor Segmenter NER Word2Vec Rankin g .pdf Image Image Ranker search MobileNet .pdf Extractor Segmenter
8 . Neural Search Key-Value index index index .pdf Feature extraction Encoding Vector index search Text Text BERT Extractor Segmenter NER Word2Vec Ranking .pdf Image Image Ranker search MobileNet Extractor Segmenter .pdf
9 .Neural Search search Crafter Vector data Encoder Ranker index Indexer Segmenter Key- Value output Indexer
10 .Jina: a cloud native neural search framework Search Analytics Search as a Service Solution Architects Search Engineer Flow APIs Pea & Pod APIs Backend Engineer Drivers & Primitive Types Executors Crafter Encoder Vector Ranker Evaluator Indexer Segmenter Key- Value Classifier Indexer AI Engineer AI models & search algorithms
11 .Jina: a cloud native neural search framework Multi/Cross-Modality Search on Unstructured Data Search AI in Production Universal Full-stack Time-saver Plug & Play Ownership Cloud-Native
12 .Jina: PDF search system revisit
13 .Jina: Flow, Pod, Pea Po Pea Executor d BERT model BERT-0 Runtime Executor Interface Class BERT-1 for Driver transformers Message Queue
14 .Jina: Recursive Document Doc mod: mod: text image g=1 g=1 g=1 g=2 g=2 g=2 g=2 g=2 g=3
15 .Saving Human Time Shop the look, Ecommerce Chatbot, QA, Customer service Multimodal, Rich doc search LOC: ~150 LOC: ~50 LOC: ~250 50L Python + 93L YAML 21L Python + 27L YAML 9L Python + 237L YAML Save >1000 hours Save >500 hours Save >1500 hours
16 .Since 2020 May, LinkedIn Twitter 3285 680 Github Stars 2500+ Slack Community 446 Forks Used by 400+ 50+ Top30 Contributors Events Top50 Downloads 120+ 10 100K+ (<4% codebase)
17 . https://github.com/jina-ai/jina @JinaAI_ https://jina-ai.slack.com/ https://www.linkedin.com/company/jinaai/ https://www.youtube.com/c/jina-ai • 📚 Examples: https://github.com/jina-ai/examples • 👏 Join us at https://jina.ai/join
18 .关注我们 关注 “AI 检索技术博客”公众号, 获取更多重磅讲师技术文章、 相关领域资讯、以及线下分享活动信息 BigData+AI Meetup *S04 向量检索专场
19 .BigData+AI Meetup *S04 向量检索专场 Thanks