ALTeGraD 2019-20AdvancedLearningforTextand

GraphsProf.M.Vazirgiannis

DataScienceandMininggroup(DaSciM)@LIX,Ecole Polytechnique

mvazirg@lix.polytechnique.frScholar:https://tinyurl.com/yash58y6

Tweeter:@mvazirg

DaSciM@X - ΜLforGraphsandText• GraphMining

• GraphDegeneracy• GraphofWords– bestpaperACMCIKM2013• D-coremetricforacademicimpact– adoptedbyAminer.org

• Graphsimilarity• KernelsandDL– distinguishedpaperawardIJCAI2018• InfluenceMaximization –Nature/Scientificreports2016

• DeepLearningforGraphsandTextdata• Nodeembeddings – SocialNets• Graphclassification

• Strongrelationshipsto• Industry(Google,Tencent, BNP,Airbus,AXA,Tradelab,Deezer,…)• Academia(Tsinghua,Jiaotong,CarnegieMellon,Columbia,NTUAAthens)

ALTEGRAD(since2014)

• Objectives• Providestateoftheartresearchresultsandhandsonexperiencefor

• TextMiningNLPincludingDLmethods• Categorization,opinionmining• Eventdetectionintwitter• Keywordextraction,automatedsummarization,recommendations

• Machinelearningforgraphsincluding• Graphdegeneracyforcommunitydetection• DeepLearningfornode/edge/(sub)graphclassification• Applicationsforsocialnetworks,frauddetection,biology,chemo-informatics

ALTEGRADSyllabusTEXT/NLP – Graph based Text Mining- Graph of Words - GoWvis- Keyword extraction (TFIDF, TextRank, ECIR'15, EMNLP'16)- extractive summarization (EMNLP'17)- Subevent detection in twitter streams (ICWSM'17)- graph based document classification: TW-IDF (ASONAM'15), TW-ICW, subgraphs (ACL'15)- abstractive summarization - ACL 2018 summarizationTEXT – NLP - Word & doc embeddings (P) - Word embeddings: word2vec-glove models, doc2vec, subword, Latent Semantic Indexing, context based embeddings- doc similarity metrics: Word Mover’s distance, shortest path kernels (EMNLP16)

Deep learning for NLP ICNNs, RNNs LSTMs for NLP, text classification

Deep learning for NLP IMeta-architectures- Sequence to Sequence: Attention (HAN),

Domains: summarization. Translation, image captioning

- Siamese – energy based learning

ALTEGRADSyllabusGraph kernels, community detectionGrakel python library - https://github.com/ysig/GraKeL/tree/develop.

Deep Learning for Graphs – node classification- node embeddings (deepwalk & node2vec) for node classification and link prediction - Supervised node embeddings (GCNN, …)

Deep Learning for Graphs – Graph classification- graph CNNs- message passing - Auto-encoders- Sets embeddings – point clouds

Altegrad Team

• M.Vazirgiannis – Prof,DaSciM leader,LIX@X

• Dr.A.Tixier – ML/DLforText

• Dr.G.Nikolentzos - DLforGraphsbestpaperawardinIJCAI2018

• JBRemy- MLforText/NLP

ALTEGRADCourseformatandlogistics• 7 sessions x 4 hours• 2h Lecture + 2h Lab• Data challenge (1 month …)

Evaluation20% lab assignments80% data challenge performance (report/creativity/leaderboard score/)

Moodle: http://moodle.lix.polytechnique.fr/moodle/

VERY IMPORTANT: Register/enroll at: https://tinyurl.com/ycsp2wcs- get accesstotheteaching/labmaterial- Receiveourannouncements

Schedule07/10/2018: 13- 17:00- Curie@Cachan22/10/2018: 13- 17:00- Curie@Cachan28/10/2018:13- 17:00- Curie@Cachan19/11- 13h-17h:Amphi Faurre@X16/12 13h-17h:Amhi Monge@X17/12 13h-17h:Amphi Monge@X14/01/2018:13- 17:00- Curie@Cachan

ALTEGRAD- Whychoosethiscourse• State of the art AI DL methods and software for the

dominant data formats: Graphs, Text/NLP• Acquire practical experience with large scale

relevant problems• Awesome applications: Web Marketing/advertising,

NLP, Search, fraud detection, social media• Research Internship and/or PhD with DaSciM

• Register/enroll at: https://tinyurl.com/ycsp2wcs

THANKYOU!!

Michalis Vazirgiannis,mvazirg@lix.polytechnique.frScholar: https://tinyurl.com/yash58y6

Tweeter:@mvazirgDataScienceandMiningGroup@LIX.Xhttp://www.lix.polytechnique.fr/dascim/

Thispresentation:https://tinyurl.com/y69ptlgr