Ex­tend­ed Sem­i­nar in Ma­chine Learn­ing and Data Min­ing

The sem­i­nar page can be found here in TUCaN.

Who, when and where?

The sem­i­nar will be held as block sem­i­nar on Jan­uary 8th and 9th, in Room A126.

The group write-ups are due on Jan­uary 29, 2018, the stu­dents' re­views of these write-ups on Febru­ary 12, 2018.

The kick-off meet­ing was on Tues­day, Oc­to­ber 24, 2017, 17:10h in C110.

The sem­i­nar will be joint­ly held by Profs. Carsten Bin­nig, Jo­hannes Fürnkranz, and Kris­tian Ker­st­ing.

In­struc­tions for the write-ups are now avail­able.

Pre­req­ui­sites

It is not nec­es­sary to have prior knowl­edge in ar­ti­fi­cial in­tel­li­gence, but prior knowl­edge in data min­ing and ma­chine learn­ing is helpful.​ Partic­i­pa­tion is lim­it­ed to 20 students.​ In case we have more stu­dents, stu­dents with prior knowl­edge in data min­ing and knowl­edge dis­cov­ery will be preferred.​ The se­lec­tion will be made at kick-off meet­ing.

For fur­ther ques­tions feel free to send an email to ml-sem@​ke.​tu-darmstadt.​de. No prior reg­is­tra­tion is need­ed, how­ev­er, please stlll send us an email so that we are able to es­ti­mate be­fore­hand the num­ber of par­tic­i­pants, and have your E-mail ad­dress for pos­si­ble announcements.​ Also make sure that you are reg­is­tered in TUCaN.

Con­tent

This year's topic of the sem­i­nar is In­ter­ac­tive Ma­chine Learn­ing, i.​e.​ machine learn­ing al­go­rithms that are meant to be used in­ter­ac­tive­ly or co-ac­tive­ly with a human user.​ We will con­cen­trate on re­cent pa­pers pub­lished in work­shops, jour­nals, and conferences.​ A list of top­ics is avail­able below.​ The top­ics will be as­signed based on an on-line bid­ding pro­cess, which will be opened after the kick-off.​ The final as­sign­ment will be made a week later.

Ex­tend­ed Sem­i­nar?

What is "Ex­tend­ed" about this sem­i­nar? Stu­dents are not only ex­pect­ed to give a short talk, but also to pre­pare a small write-up.​ The write-up will be pre­pared in groups, each group will cover one theme, con­sist­ing of four topics.​ The final write-up must be con­cise and short, and should give a short overview of the theme (not nec­es­sar­i­ly lim­it­ed to the stud­ied pa­pers).

In ad­di­tion, we will also do a peer re­view­ing pro­cess, as it is usu­al­ly done at sci­en­tif­ic conferences.​ This means that you also have to read (some) of the other write-ups and pro­vide feed­back by fill­ing out a re­view form.

Be­cause they are more work for stu­dents, stu­dents re­ceive 4 CPs for Ex­tend­ed Sem­i­nars (in­stead of 3 CPs for reg­u­lar sem­i­nars).

Talks

Al­though each topic is typ­i­cal­ly as­so­ci­at­ed with a sin­gle paper, the point of the talk is not to ex­act­ly re­pro­duce the en­tire con­tents of the paper, but to com­mu­ni­cate the key ideas of the meth­ods that are in­tro­duced in the paper.​ Thus, the con­tent of the talk should ex­ceed the scope of the paper, and demon­strate that a thor­ough un­der­stand­ing of the ma­te­ri­al was achieved.​ See also our gen­er­al ad­vices on giv­ing talks.

Stu­dents are ex­pect­ed to give a 20 (!) minute talk on the ma­te­ri­al they are as­signed, fol­lowed by 10 min­utes of questions.​ Note that the com­pa­ra­bly short pe­ri­od of time forces you to get the most im­por­tant points of your topic across.​ You are not ex­pect­ed to pre­sent ev­ery­thing.

The talks are ex­pect­ed to be ac­com­pa­nied by slides.​ In case you do not own a lap­top, please send us the slides in ad­vance, so that we can pre­pare and test the slides.​ The talk and the slides should be in En­glish.

Write-Up

The talks are or­ga­nized in top­i­cal groups.​ Each group must pre­pare one short write-up of their work.

Con­tent: The pa­pers are re­lat­ed to each other.​ Your task is to use these pa­pers to cre­ate a mi­ni-sur­vey that com­bines the re­sults of all pa­pers, and pos­si­bly other papers.​ The con­tri­bu­tion of each in­di­vid­u­al paper can be lim­it­ed to the most im­por­tant points that are con­tribut­ed by this paper to the topic.​ There must be a clear "red thread" with­in each sur­vey, a con­cate­na­tion of in­di­vid­u­al paper sum­maries is not enough.​ A pos­si­ble out­line can con­sist of an in­tro­duc­tion to set the stage and out­line the cross-cut­ting themes of all pa­pers, mul­ti­ple sec­tions on in­di­vid­u­al con­tri­bu­tions w.​r.​t.​ cross-cut­ting themes and com­par­i­son of dif­fer­ent ap­proach­es, a joined re­lat­ed work sec­tion, and a sum­ma­ry and out­look.

For­mat: The for­mat for the write-up is pre­de­fined, and fol­lows con­ven­tions that are typ­i­cal­ly used for pub­li­ca­tions in com­put­er science.​ In par­tic­u­lar, we re­quire each paper to be for­mat­ted ac­cord­ing to the Tem­plate for Con­tri­bu­tions in IEEE Trans­ac­tions. Each paper should have no more than 5 pages in this for­mat (the bib­li­og­ra­phy is not count­ed, and can be as long as nec­es­sary). The for­mat must not be changed in order to gen­er­ate more space.​ Each paper also must, of course, have a title, au­thors, and an abstract.​ The tem­plates are avail­able in Word and LaTeX, but we strong­ly rec­om­mend that you try to use LaTeX.​ Environ­ments such as MiK­TeX and TeXs­tu­dio make local La­TeX-edit­ing quite easy, and web-sites like Over­leaf offer col­lab­o­ra­tive work­ing en­vi­ron­ments for LaTeX.

Dead­line: The write-ups are due on Jan­uary 29, 2018.

Re­view­ing

Re­view­ing as­sign­ments will be made in Jan­uary based on your bids.​ A review­ing form will be pro­vid­ed by then.​ The dead­line of the stu­dents' re­views will be Febru­ary 12, 2018.

Top­ics and Sched­ule

All pa­pers should be avail­able on the in­ter­net or in the ULB.​ Note that Springer link often only works on cam­pus net­works (some­times not even via VPN). If you can­not find a paper, con­tact us.

Ses­sion 1: Ac­tive Learn­ing (Jan. 8th, 13:00h)

• Luna A.:
  Jason Baldridge, Miles Os­borne: Ac­tive learn­ing and log­a­rith­mic opin­ion pools for HPSG parse se­lec­tion. Nat­u­ral Lan­guage En­gi­neer­ing 14(2): 191-222 (2008)
• Jakob W.:
  Sud­heen­dra Vi­jaya­narasimhan, Kris­ten Grau­man: Large-Scale Live Ac­tive Learn­ing: Train­ing Ob­ject De­tec­tors with Crawled Data and Crowds. In­ter­na­tion­al Jour­nal of Com­put­er Vi­sion 108(1-2): 97-114 (2014)
• Her­rmann L.:
  Zhou S, Chen Q, Wang X (2014) Ac­tive Se­mi-Su­per­vised Learn­ing Method with Hy­brid Deep Be­lief Net­works. PLoS ONE9(9): e107122.
• Felix H.:
  Jamieson, K.​G.,​ Jain, L.,​ Fernan­dez, C.,​ Glat­tard, N.​J.,​ Nowak, R.:​ NEXT: A Sys­tem for Re­al-world De­vel­op­ment, Eval­u­a­tion, and Ap­pli­ca­tion of Ac­tive Learn­ing. In: Cortes, C.,​ Lawrence, N.,​ Lee, D.,​ Sugiya­ma, M.,​ Garnett, R.​ (eds.​) Advances in Neu­ral In­for­ma­tion Pro­cess­ing Sys­tems 28 (NIPS). pp. 2638–2646 (2015)

Ses­sion 2: Hy­per­pa­ram­e­ter Op­ti­miza­tion (Jan 8th, 15:30h)

• Jia H.:
  Lin­dauer, M.​T.,​ Hoos, H.​H.,​ Hutter, F.,​ Schaub, T.:​ Aut­o­Fo­lio: An Au­to­mat­i­cal­ly Con­fig­ured Al­go­rithm Se­lec­tor. Jour­nal of Ar­ti­fi­cial In­tel­li­gence Re­search 53, 745–778 (2015)
• Yan­tao S.:
  Lars Kot­thoff, Chris Thorn­ton, Hol­ger H.​ Hoos, Frank Hut­ter, Kevin Ley­ton-Brown: Auto-WEKA 2.0: Au­to­mat­ic model se­lec­tion and hy­per­pa­ram­e­ter op­ti­miza­tion in WEKA. Jour­nal of Ma­chine Learn­ing Re­search 18: 25:1-25:5 (2017).
  Aaron Klein, Ste­fan Falkn­er, Simon Bar­tels, Philipp Hen­nig, Frank Hut­ter:Fast Bayesian Op­ti­miza­tion of Ma­chine Learn­ing Hy­per­pa­ram­e­ters on Large Datasets. AIS­TATS 2017: 528-536
• Yimin X.:
  Yu­tian Chen, Matthew W.​ Hoffman, Ser­gio Gomez Col­menare­jo, Misha Denil, Tim­o­thy P.​ Lilli­crap, Matthew Botvinick, Nando de Fre­itas: Learn­ing to Learn with­out Gra­di­ent De­scent by Gra­di­ent De­scent. ICML 2017: 748-756.
• Jianyang T.:
  Lisha Li, Kevin Jamieson, Giu­lia De­Sal­vo, Af­shin Ros­tamizadeh, Ameet Tal­walkar, Hy­per­band: A Novel Ban­dit-Based Ap­proach to Hy­per­pa­ram­e­ter Op­ti­miza­tion (only on Arxiv so far)

Ses­sion 3: In­ter­ac­tive Ma­chine Learn­ing (Jan 9th, 10:00h)

• Matthias B.:
  Dzyu­ba, V.,​ van Leeuwen, M.,​ Nijssen, S.,​ De Raedt, L.:​ In­ter­ac­tive Learn­ing of Pat­tern Rank­ings. In­ter­na­tion­al Jour­nal of Ar­ti­fi­cial In­tel­li­gence Tools 23(6) (2014)
• Max­i­m­il­ian O.:
  Pan­na­ga Shiv­aswamy, Thorsten Joachims: Coac­tive Learn­ing. Jour­nal of Ar­ti­fi­cial In­tel­li­gence Re­search 53: 1-40 (2015)
• Ma­ciej M.:
  Odom, P.,​ Natara­jan, S.:​ Ac­tive­ly in­ter­act­ing with ex­perts: A prob­a­bilis­tic logic ap­proach. In: ECML PKDD 2016, Lec­ture Notes in Com­put­er Sci­ence, vol. 9852, pp. 527–542.​ Springer, Cham (2016)
• Sascha S.:
  Ste­fano Teso, Paolo Drag­one, An­drea Passeri­ni: Coac­tive Cri­tiquing: Elic­i­ta­tion of Pref­er­ences and Fea­tures. AAAI 2017: 2639-2645

Ses­sion 4: In­ter­ac­tion with Ex­pert Knowl­edge (Jan 9th, 13:30h)

• Alexan­der L.:
  Hu, Z.,​ Ma, X.,​ Liu, Z.,​ Hovy, E.,​ Xing, E.:​ Har­ness­ing Deep Neu­ral Net­works with Logic Rules. In: 54th An­nu­al Meet­ing of the As­so­ci­a­tion for Com­pu­ta­tion­al Lin­guis­tics (ACL). pp. 2410–2420.​ ACL  (2016)
• Erich W.:
  Ste­fano Teso, Rober­to Se­bas­tiani, An­drea Passeri­ni: Struc­tured learn­ing mod­u­lo the­o­ries. Ar­ti­fi­cial In­tel­li­gence 244: 166-187 (2017)
• Leon Ch.:
  Marco Túlio Ribeiro, Sameer Singh, Car­los Guestrin: "Why Should I Trust You?": Ex­plain­ing the Pre­dic­tions of Any Clas­si­fi­er. Pro­ceed­ings KDD 2016: 1135-1144
• Emil A.:
  Caru­a­na, R.,​ Lou, Y.,​ Gehrke, J.,​ Koch, P.,​ Sturm, M.,​ Elhadad, N.:​ In­tel­li­gi­ble mod­els for health­care: Pre­dict­ing pneu­mo­nia risk and hos­pi­tal 30-day read­mis­sion. In: Pro­ceed­ings of the 21th ACM SIGKDD In­ter­na­tion­al Con­fer­ence on Knowl­edge Dis­cov­ery and Data Min­ing, Syd­ney, NSW, Aus­tralia, Au­gust 10-13, 2015.​ pp.​ 1721–1730 (2015)
• David L.:
  Son N.​ Tran; Artur S.​ d'Avila Garcez: Deep Logic Net­works: In­sert­ing and Ex­tract­ing Knowl­edge From Deep Be­lief Net­works.  IEEE Trans­ac­tions on Neu­ral Net­works and Learn­ing Sys­tems, 2016.

Bid­ding

Pa­pers were dis­tribut­ed via bidding.​ Students could bid for pa­pers using this form. All stu­dents re­ceived a paper in one of the two most pre­ferred categories.​ A better choice was often not pos­si­ble (e.​g.,​ if you bid only for a sin­gle paper and set all oth­ers to "Don't Want It", it is un­like­ly that you re­ceived your paper).

Grad­ing