【2】亚马逊AWSAI Services Overview

2020-03-01 135浏览

2.围绕数据的“飞轮” 点击流用户活动内容生成购买点击喜好传感器数据更多的数据更多的用户更好的分析更好的产品对象存储数据库数据仓库数据流分析商业智能 Map/Reduce 内存数据库数据检索机器学习深度学习人工智能
3.围绕数据的“飞轮” 大数据更多的数据更多的用户更好的分析更好的产品机器学习& 人工智能
4.图像理解深度学习的爆发算法语音识别程序模型数据自然语言处理 GPUs & 计算加速机器自主
5.AWS 之上的人工智能应用 Zillow • Zestimate (using Apache Spark) Howard Hughes Corp • Lead scoring for luxury real estate purchase predictions FINRA • Anomaly detection, sequence matching, regression analysis, network/tribe analysis Netflix • Recommendation engine Pinterest • Image recognition search Fraud.net • Detect online payment fraud DataXu • Leverage automated & unattended ML at large scale (Amazon EMR + Spark) Mapillary • Computer vision for crowd sourced maps Hudl • Predictive analytics on sports plays Upserve • Restaurant table mgmt & POS for forecasting customer traffic TuSimple • Computer Vision for Autonomous Driving Clarifai • Computer Vision APIs
6.AWS 上的 AI 应用 • Pinterest Lens • Netflix 推荐引擎
7.Amazon 的人工智能应用数千名员工致力于人工智能领域发现& 搜索执行 &物流现有产品的增强定义新的产品分类将机器学习拓展更广领域
8.在Amazon 最初的人工智能应用 (1995)
9.
10.
11.
12.
13.
14.AWS 可以帮助客户把人工智能应用于每个应用程序的核心&业务之中吗?
15.Amazon AI 构建于深度学习之上的智能服务
16.AmazonAI:新的深度学习服务 Polly 生活化的语音服务 Rekognition 图像分析 Lex 对话引擎深度学习框架 MXNet, TensorFlow, Theano, Caffe, Torch
17.控制力可用性& 简易性 AmazonAI:新的深度学习服务 Polly Rekognition 深度学习框架 MXNet, TensorFlow, Theano, Caffe, Torch Lex 人工智能的托管的 API服务为客户模型定制的深度学习框架
18.AWS 深度学习AMI 高达 ~40k CUDA cores MXNet TensorFlow Theano Caffe Torch 预配置的 CUDA 驱动 Anaconda, Python3 + CloudFormation 模版 + 容器镜像文件一键获得的GPU 加速的深度学习
19.全新的 EC2 P2 实例高达16 块 GPUs ▪ 这款新实例类型包含了高达 8个 NVIDIA Tesla K80 Accelerators, 每个运行一对 NVIDIA GK210 GPUs. ▪ 每块GPU 提供 12 GiB 内存 (内存存取带宽达到240 GB/秒), 以及 2,496 个并行处理核心 Instance Name GPU Count vCPU Count Memory Parallel Processing Cores GPU Memory Network Performance p2.xlarge 1 4 61 GiB 2,496 12 GiB High p2.8xlarge 8 32 488 GiB 19,968 96 GiB 10 Gigabit p2.16xlarge 16 64 732 GiB 39,936 192 GiB 20 Gigabit
20.深度学习框架 – MXNet 概述
21.MXNet
22.为什么选择 MXNet ？ • 节省以及资源效率 • 工程中廉价的GPUs、较小的内存以及网络的限制 • 速度 • 线性的扩展能力 • 简单 • 混合了声明式（declarative）和命令式()代码的特点
23.MXNet:可扩展的深度学习框架
24.MXNet 框架的特点
25.MXNet：博采众家之长命令式 NDArray API 声明式 Symbolic Executor
26.3D Image Constructionhttps://github.com/piiswrong/deep3d100行Python代码
27.在 TX1 无人机上运行 TX1 with customized board Drone Realtime detection and tracking on TX1 ~10 frame/sec with 640x480 resolution
28.处处可部署 Beyond Runs in browser with Javascript Amalgamation Fit the core library with all dependencies into a single C++ source file Easy to compile on … The first image for search “dog” at images.google.com Outputs “beagle” with prob = 73% within 1 sec BlindTool by Joseph Paul Cohen, demo on Nexus 4
29.Deep RL Playing Flappy Birds • Reinforcementlearning:Observe environment  Take Action  Achieve Reward  Repeat. Goal is to maximize rewards over time. • There are threeinterfaces:• • • getInitState() for initialization getAction() setPerception(nextObservation,action,reward,termina l) •Resources:• •http://ww1.sinaimg.cn/mw690/8708cad7jw1f8naomrmweg209n0fo7wj.gifhttps://github.com/li-haoran/DRL-FlappyBird
30.MXNet 演示：https://35.161.116.218/notebooks/money_predict.ipynb
31.Polly:生活化的语音服务将文本转化为生活化语音全托管 Voice Quality & Pronunciation 1. 自动化、精准的文本处理 2. 智能化的且易于理解 3. 将语义加入文本当中 4. 定制化的发音 47 种语音 24 种语言低延迟、实时文章、博客训练材料 Chatbots (Lex) 公告
32.人-机交互会话的发展第一代: 面向机器的交互第二代: 面向控制& 翻译第三代: 意图导向
33.Lex:构建自然的通过语音和文本的会话交互语音 & 文本 “Chatbots” Alexa 应用在手机、Web以及设备上的语音交互不断提升的人性化交互… • 联系、服务支持中心的接口 (文本 + 语音) • 员工工作效率和协同 (分钟级别到秒级) 在Slack & Messenger 上的文本交互企业应用 Salesforce Microsoft Dynamics Marketo Zendesk Quickbooks Hubspot
34.Book Flight booking Flight London “Book a flight to London” London Heathrow Automatic Speech Recognition Natural Language Understanding Intent / Slot model Utterances Flight Booking Origin Destination Departure Date London Heathrow
35.Book Flight booking Flight London “Book a flight to London” Location London Heathrow Automatic Speech Recognition Natural Language Understanding Intent / Slot model Location Utterances Flight Booking Origin Seattle Destination London Heathrow Departure Date
36.Book Flight booking Flight London “Book a flight to London” Location London Heathrow Automatic Speech Recognition Natural Language Understanding Intent / Slot model Location Utterances Flight Booking Origin Seattle Destination London Heathrow Departure Date Polly “When would you like to fly?” “When would you like to fly?” Prompt
37.“Next Friday” Flight Booking Origin Seattle Destination London Heathrow Departure Date Polly “When would you like to fly?” “When would you like to fly?” Prompt
38.Flight booking Next “Next Friday” Friday 02 / 24 / 2017 Automatic Speech Recognition Natural Language Understanding Intent / Slot model Utterances Flight Booking Origin Seattle Destination London Heathrow Departure Date 02/24/2017
39.Flight booking Next “Next Friday” Friday 02 / 24 / 2017 Automatic Speech Recognition Natural Language Understanding Intent / Slot model Utterances Flight Booking Polly “Your flight is booked for next Friday” Origin Seattle Destination London Heathrow Departure Date 02/24/2017 “Your flight is booked for next Friday” Confirmation
40.Flight booking Next “Next Friday” Friday 02 / 24 / 2017 Automatic Speech Recognition Natural Language Understanding Intent / Slot model Utterances Flight Booking Origin Seattle Destination London Heathrow Departure Date 02/24/2017 Hotel Booking
41.与 AWS Mobile Hub 集成 Authenticate users Store and share media Synchronize data Conversational Bots AWS Mobile SDKs Lex Analyze user behavior AWS Mobile Hub Track retention More ….
42.Mobile Hub 实现企业应用的连接 Mobile Hub SaaS Connector Amazon Lex Amazon API AWS Gateway Lambda 1: Understand user intent 2: Invoke a SaaS application or an existing business application User Input 3: Translate REST response into natural language Mobile App Amazon API AWS Gateway Lambda Mobile Hub Custom Connector Firewall Business Application
43.应用案例: Capital One “As a heavy user of AWS, Amazon Lex’s seamless integration with other AWS services like AWS Lambda and AWS DynamoDB is really appealing.” “A highly scalable solution, it also offers potential to speed time to market for a new generation of voice and text interactions such as our recently launched Capital One skill for Alexa.”
44.Amazon Rekognition 基于深度学习的图像识别服务目标和场景检测面部分析人脸比对集成了 S3, Lambda, Polly, Lex 人脸识别
45.对象和场景识别为成千上万的对象、场景和概念生成标签，并配有可信度的数字 • 检索、过滤并对图片库去粗取精 • 对用户生成的内容进行智能检索 • 摄影、旅游、房地产、度假以及租赁等应用场景 Maple Villa Tree Plant Potted Plant Garden Swimming Pool Water Backyard
46.面部分析在图像中定位人脸并分析面部的情绪，检测情感、姿势、地标等特性 • 剪裁图片和重叠广告时需要避开的面孔 • 获得人口学以及情感的数据推荐最佳照片 • 提高在线约会匹配的推荐 • 动态的个性化广告
47.人脸比对测量两张图片中同一个人的可能性 • 为应用和设备添加人脸验证 • 扩展了物理安全控制的应用领域 • 客人对VIP 设施的使用 • 在线考试以及民意调查时的用户验证
48.人脸识别通过针对存储的面部向量的集合找到输入面部图像的最接近的匹配来识别图像中的人 • 社交应用、消息类应用中加入朋友标签 • 协助找到始终人口 • 确定可以访问敏感区域的员工 • 在历史和媒体的档案中找到“名人”
49.
50.
51.
52.应用案例：公共安全领域的智能应用
53.
54.人工智能的时代已经到来
55.Amazon AI 服务 • 充分利用了 Amazon 内部在 AI / Ml领域的经验 • 全托管的API 服务，嵌入的AI服务提供了最大的可访问性和简单性 • 完整的深度学习堆栈，包含了专业的平台、引擎应用程序
56.Thank you! lianghong@amazon.com