版权说明:本文档由用户提供并上传,收益归属内容提供方,若内容存在侵权,请进行举报或认领
文档简介
1、Pragmatic Deep Learning for image labelling- An application to a travel recommendation engine主讲人:Dataiku数据科学家 Alexandre HubertOutlineResultsMore images !Labeling ImagesPragmatic deep learning for dummiesIntroduction and ContextPost ProcessingAKA: Image for BI on steroidsIterative building of a recom
2、mender systemDataikuData Science Software Editor of Dataiku DSSFounded in 201390 + employees, 100 + clientsParis, New-York, London, San Francisco, SingaporeDESIGNPREPAREANALYSEMODELLoad and prepareVisualize and shareBuild your your datayour workmodelsPRODUCTIO NAUTOMATEMONITORSCORERe-execute yourFol
3、low your productionGet predictions workflow at easeenvironmentin real time Key FiguresE-business vacation retailerNegotiate the best prize for their clients Discount luxury18 Millions of clients.Hundreds of sales opened everydaySale Image is paramountPurchase is impulsiveSpecificitiesHighly temporar
4、y sales- Classical recommender system fail- Time event linked (Christmas, ski, summer)Expensive Product- Few recurrent buyers- Appearance counts a lotIterative Building of a Recommender SystemBasic Recommendation EnginesOther FactorsOne Meta Model to Rule Them AllDescribeRecommendRecommenders as fea
5、turesCombineMachine learning to optimizepurchasing probabilityRecommender system for Home Page Ordering+7% revenue(A/B testing)Optimization of home displayCustomer visits PurchasesMeta ModelRecommendation EnginesCleaning, combining and enrichment of dataSales informationEvery customer is shown the 1
6、0 sales he is the most likely to buyGeneration of recommendations based on user behaviourthe application automatically runs and compiles heterogeneous dataMetal model combine recommendations to directly optimize purchasing probabilitySales ImagesBatch Scoring every nightWhy use Image ?A picture is w
7、orth a thousand wordsWe want do distinguish Ski Sun and Beach Integrating Image InformationCONTENT BASEDSea + Beach +Forest + HotelPool + Palm Trees Hotel+ MountaPool + Forest + Hotel + SeaLabeling ModelSales ImagesSales descriptions vectorRecommender SystemImage Labelling For Recommendation EngineP
8、ragmatic Deep learning for “Dummies”Using Deep Learning modelsCommon Issues“I dont have a deep leaning expert”“I dont have GPUs server”“I dont have labelled data” (or too few)“I dont have the time to wait for model training ”I dont want to pay to pay for private apis” / “Im afraid their labelling wi
9、ll change over time”Solution 1 : Pre trained models“I dont have (or few) labelled data”- Is there similar data ?USPLACES DATABASESUN DATABASE205 categories2.5 M images307 categories110 K imagesSolution 1 : Pre trained modelsIf there is open data, there is an open pre trained model !Kudos to the comm
10、unityCheck the licensingExample with Places (Caffe Model Zoo) :swimming_pool/outdoor: 0.65inn/outdoor: 0.06tower: 0.53skyscraper: 0.26Solution 2 : Transfer LearningCredit : Fei-Fei Li & Andrej Karpathy & Justin Johnson /slides/winter1516_lecture11.pdfSolution 2 : Transfer Le
11、arningLeverage existing knowledge !OUR DATAPLACES DATABASESUN DATABASECaffe ModelZooTraining(optional)CPUGPUGPUtower: 0.53skyscraper: 0.26Re-TrainingTransferred Data :Last convolutional layer featuresRe-trained modelTensorFlow2 fully connected layersPre-trained modelVGG16Accuracy: 72%, Top-5 Acc: 90
12、 % state of the art on dataset alonePost Treatment & ResultsUsing Images information for BI on steroids(Or how we transfer the labelling information)Labels post-processingIssue with our approach:Redondant informationComplementary informationSolution : NMF Matrix FactorizationDimensionReductionSparsi
13、tyBalancednessExplicabilityImage content detectionTopic scores determine the importance of topics in an imageTOPICTOPIC SCORE (%)Tower Skyscraper Office building62Bridge River Viaduct38TOPICTOPIC SCORE (%)Golf course Fairway Putting green31Hotel Inn Apartment building outdoor30Swimming pool Lido Dec
14、k Hot tub outdoor22Beach Coast - Harbor17Results ?1) Visits : France and Morocco Pool displayed2) First Recommendation Mostly France & Mediterranean Fails to display pools3) Only Images recommendation Pool all around the world Does not respect budget4) Third column = Right Mix1)2)3)4)ConclusionDo it
15、erative data science !Start simple and grow Evaluate at each stepsImage labelling = BI on steroidsDeep LearningDont start from scratch ! Is there existing data ?Is there a pre-trained model ?Transfer LearningKick-start your project Gain time and moneyAny Data Scientist can do itWhats next ?Learned al
温馨提示
- 1. 本站所有资源如无特殊说明,都需要本地电脑安装OFFICE2007和PDF阅读器。图纸软件为CAD,CAXA,PROE,UG,SolidWorks等.压缩文件请下载最新的WinRAR软件解压。
- 2. 本站的文档不包含任何第三方提供的附件图纸等,如果需要附件,请联系上传者。文件的所有权益归上传用户所有。
- 3. 本站RAR压缩包中若带图纸,网页内容里面会有图纸预览,若没有图纸预览就没有图纸。
- 4. 未经权益所有人同意不得将文件中的内容挪作商业或盈利用途。
- 5. 人人文库网仅提供信息存储空间,仅对用户上传内容的表现方式做保护处理,对用户上传分享的文档内容本身不做任何修改或编辑,并不能对任何下载内容负责。
- 6. 下载文件中如有侵权或不适当内容,请与我们联系,我们立即纠正。
- 7. 本站不保证下载资源的准确性、安全性和完整性, 同时也不承担用户因使用这些下载资源对自己和他人造成任何形式的伤害或损失。
最新文档
- 鼻息肉的鼻内镜手术护理
- 小儿身高体重的百分位监测
- 腰椎病患者的偏食与护理
- 2026三亚市教师招聘考试题库及答案
- 2026曲靖市教师招聘考试题及答案
- 二次函数参考题目及答案
- 2026一年级上《加减法初步认识》知识闯关游戏
- 《认识信息安全》教案-2025-2026学年鲁教版(新教材)小学信息技术三年级下册
- 保障报价连续性应急预案
- 2026年幼儿园复学宣传
- 大学校医笔试试题及答案
- 2025年北京市西城区高考数学二模试卷
- 山东中烟招聘考试真题2025
- 扶贫助销协议书
- 高压线防护脚手架专项方案
- 南方电力安全培训教材课件
- 2025年空军文职技能岗考试保管员复习题及答案
- 花束包装课件制作
- 工程质保期内维修方案(3篇)
- 2025年四川省法院公开招聘聘用制审判辅助人员考试(面试)历年参考题库及答案
- 老年高血压患者的康复护理
评论
0/150
提交评论