how to be a successful phd student - google docsmdredze/publications/howtobeasuccessful...how to be...

13
How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins University) Hanna M. Wallach (Microsoft Research) Version 1.1, July 5, 2019 (fixed broken links) Version 1.0, March 20, 2012 Being a graduate student can be extremely rewarding and a lot of fun, but it’s also hard work. Like anything in life, there are ways to succeed and ways to fail. Hopefully, you will find a good support network that can help you learn what it takes to be a good graduate student. However, not everyone is able to find such a network, and it’s important to learn these lessons early. This guide was written based on our experiences as graduate students and our experiences advising graduate students, specifically, PhD students in NLP and machine learning; however, we hope many of the points are applicable to a wider audience. Where possible, we have indicated when advice is field-specific. Beyond that, the advice is listed in no particular order. No doubt there will be points with which you disagree. That’s totally fine -- there are certainly points about which even we had differing views; however, before you decide to ignore these points, it’s well worth taking the time to understand why we included them. Becoming a PhD Student Why Get a PhD? This question is dealt with extensively in other documents, so we will not discuss it in detail here; however, we want to emphasize the importance of asking this question. Getting a PhD will change your career path dramatically. Not only does getting a PhD mean you will spend years in school (with both personal and financial implications), it will also change the type of career you pursue. While we are both happy we got PhDs, it isn’t the right decision for everyone. Before applying to graduate school, you should think hard about whether this is the right path for you. Some specific points to consider are as follows: 1. You have to really want it. Getting a PhD takes a long time and a great deal of dedication and hard work. Unless you really want it, you will not finish the PhD. 2. It’s more than okay to do something else first. Many students go straight from their undergraduate degree to graduate school. That’s certainly fine -- in fact, that’s what we did. However, some of the best PhD students left academia for a bit before starting graduate school. Doing something else gives you a break from the academic world and

Upload: others

Post on 09-Jul-2020

1 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

HowtoBeaSuccessfulPhDStudent(inComputerScience(inNLP/ML))MarkDredze(JohnsHopkinsUniversity)HannaM.Wallach(MicrosoftResearch)Version1.1,July5,2019(fixedbrokenlinks)Version1.0,March20,2012

Beingagraduatestudentcanbeextremelyrewardingandalotoffun,butit’salsohardwork.Likeanythinginlife,therearewaystosucceedandwaystofail.Hopefully,youwillfindagoodsupportnetworkthatcanhelpyoulearnwhatittakestobeagoodgraduatestudent.However,noteveryoneisabletofindsuchanetwork,andit’simportanttolearntheselessonsearly.Thisguidewaswrittenbasedonourexperiencesasgraduatestudentsandourexperiencesadvisinggraduatestudents,specifically,PhDstudentsinNLPandmachinelearning;however,wehopemanyofthepointsareapplicabletoawideraudience.Wherepossible,wehaveindicatedwhenadviceisfield-specific.Beyondthat,theadviceislistedinnoparticularorder.Nodoubttherewillbepointswithwhichyoudisagree.That’stotallyfine--therearecertainlypointsaboutwhichevenwehaddifferingviews;however,beforeyoudecidetoignorethesepoints,it’swellworthtakingthetimetounderstandwhyweincludedthem.BecomingaPhDStudentWhyGetaPhD?Thisquestionisdealtwithextensivelyinotherdocuments,sowewillnotdiscussitindetailhere;however,wewanttoemphasizetheimportanceofaskingthisquestion.GettingaPhDwillchangeyourcareerpathdramatically.NotonlydoesgettingaPhDmeanyouwillspendyearsinschool(withbothpersonalandfinancialimplications),itwillalsochangethetypeofcareeryoupursue.WhilewearebothhappywegotPhDs,itisn’ttherightdecisionforeveryone.Beforeapplyingtograduateschool,youshouldthinkhardaboutwhetherthisistherightpathforyou.Somespecificpointstoconsiderareasfollows:

1. Youhavetoreallywantit.GettingaPhDtakesalongtimeandagreatdealofdedicationandhardwork.Unlessyoureallywantit,youwillnotfinishthePhD.

2. It’smorethanokaytodosomethingelsefirst.Manystudentsgostraightfromtheirundergraduatedegreetograduateschool.That’scertainlyfine--infact,that’swhatwedid.However,someofthebestPhDstudentsleftacademiaforabitbeforestartinggraduateschool.Doingsomethingelsegivesyouabreakfromtheacademicworldand

Page 2: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

canhelpyougainperspective.StudentswhodecidethattheywantaPhDenoughtoreturntoacademiaareoftenextremelymotivatedanddoverywell.

ApplyingtoGraduateSchoolInordertobeaPhDstudent,youhavetogetintograduateschool.Therearemanygoodresourcesdevotedtothissubject,suchashttps://conquer.cra.org/students.Guidanceonhowtoprepareanapplicationiscoveredhere:https://www.clsp.jhu.edu/apply-for-phd/phd-admissions-faq/.Wewanttoemphasizetheimportanceofapplyingforexternalfellowships,suchasNSF’sGraduateResearchFellowshipsandNDSEGFellowships.Havingsuchafellowshipcanmakeahugedifferencetoyourgraduateschoolexperience.Sinceyoucanusuallyapplymorethanonce,youhavenothingtolosebyapplyingforfellowshipswhileyouarealreadyapplyingforgraduateschool.Wealsohavesomegeneraladviceonapplyingandpickingaschool:

3. Thinkbeyondtheschool.Ofcourseyouwanttogotothebestschoolpossible.Butyoualsoneedtobehappy.Ifyouaren’thappy,youwon’tbesuccessful.Ifyoufindyourselfwithnosociallifeandnofriends,youwon’tbehappy.

4. Surroundyourselfwithsmartpeople.Youradvisorisimportant,butyouwillalsospendhugeamountsoftimewithotherstudents.Beingaroundsmartpeoplemakesyousmarter.Ifyouarethesmartestpersonintheroom,it’stimetofindanotherroom.

YouandYourAdvisorYourrelationshipwithyouradvisoristhesinglemostimportantrelationshipyouwillhaveingraduateschool.Agoodrelationshipwithyouradvisoriscriticaltoyoursuccess.Makesureyoupicktherightadvisorforyou.Youwillspendhundredsofhourswithyouradvisor,soitisabsolutelycriticalthatyouradvisorissomeoneyoulikeandwithwhomyouwillworkwell.Justbecauseyoulikesomeone’sresearchdoesn’tmeanyou’llworkwelltogether.Differentadvisorshavedifferentadvisingstyles.Askthemhowtheyruntheirresearchgroup.Talktotheirexistingstudentstofindoutabouttheiradvisingstyleandwhattheyexpectoftheirstudents.

5. Listentoyouradvisor.Theywon'talwaysberight,buttheyhavebeendoingresearchfor(insomecasesmuch)longerthanyouhave.(And,ifitwasn’tclear,youradvisorisyourboss--youworkforthem.Whenyouradvisortellsyoutodosomething,doit.)

6. Pingyouradvisor.Ifyouradviserhasn’trepliedtoanimportantemail,remindthemthatyouarewaitingfortheirreply.(Butfirstmakesurethatyouremailwasclear.Seehereforadvice:http://matt.might.net/articles/how-to-email/)Youradvisoris(probably)abusypersonandmaynothaveseenyouremailorrealizeditsimportance.

Page 3: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

7. Feedback.Makesureyouobtainfeedbackfromyouradvisor.Thisshouldhappenat

leastonceayear,probablyintheformofadepartmentalstudentreview.However,feedbackishelpfulanditdoesn’thurttoaskforitmoreoften,e.g.,onceasemester.

MeetingswithYourAdvisorThisisyourprimaryopportunitytogetfeedback,direction,advice,etc.Makethemostofit.

8. Meetregularly.Regularmeetingswithyouradvisorensurethatyouarebeingproductiveandenableyouradvisortogiveyoufeedback.Youshouldmeetwithyouradvisorregularly(e.g.,weekly)--ESPECIALLYifyoudon'twantto.

9. Makeanagenda.Makeanagendaforeverymeetingwithyouradvisor.Outlineeachofthetopicsthatyou'dliketocoverduringthemeetingandtheirrelativeimportance.Ifyoudon’tknowwhatyouwanttotalkabout,youradvisor’sunlikelytoknoweither.

10. Bringresults.Trytobringresults(e.g.,graphs,tables,figures)toeverymeeting.

11. Startwithasummary.Starteachmeetingbysummarizingthepreviousmeeting.Remindthemwhatyouagreedonasnextsteps,summarizewhatyou'vedone(andhaven’tdone)sincethen,andgooveryouragenda.Youradvisorhasmanyresearchprojects.Unlikeyou,theydidn’tspendthepreviousweekworkingonyourresearchprojectonly,andwillthereforeneedtocontextswitch.Summarizingyourpreviousmeetingwillenablethemtoswitchfasterandmakesureyou’rebothonthesamepage.

ManagingYourDay-to-DayWorkLife/BeingProductiveGraduateschoolisdifferentfromyourothereducationalexperiencesandanyjobyou’vehad.Successmeansbeingproductive,creative,andindependent.Thisdoesn’tcomenaturallytoeveryone,sofigureoutquicklyhowyoucanbestsucceed.Remembertoothatasuccessfulpersoncanfail.Alot.Butyouaren'tjudgedbyyourfailures,youarejudgedbyyoursuccesses.Youdon'thavetobethesmartestpersonintheroom,butyouhavetoworkveryhard.

12. Researchis#1.Noteveryoneisgoodatresearch:somepeopleexcelintheclassroom,whileothersexcelatindependentresearch.However,aPhDisaresearchdegree--thepurposeofgraduateschoolisresearch,nottakingclasses.Althoughtakingclassesispartofgraduateschool,whenitcomestosuccess,it’sallaboutresearch.Dowellenoughinyourclassesbutfocusonpublishinghighqualityresearchpapers.

13. Talktootherstudents.Talktootherstudentsregularly,bothwithinandoutsideyourlab.Thisisagoodwaytolearnwhetheryourexpectationsofyourselfandgraduate

Page 4: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

schoolarerealistic,aswellastolearnaboutimplementationdetails,interestingproblemsandsolutions,waysofthinkingandproblem-solving,and"tricksofthetrade".

14. Everyoneworksdifferently.Figureouthowyouliketoworkandwhatmakesyoumostproductive.(Doyouworkbestintheearlymorningsorlateevening?Doyoulikeworkingwithothersordoyouprefertoworkbyyourself?Doyouworkbestwhenyouhavemultipleprojectsorjustone?)It’sagoodideatodiscussthesepreferenceswithyouradvisorsotheyunderstandyoubetterandcanworkwithyouaseffectivelyaspossible.

15. Keeptoaregularschedule.Thiswillhelpyoumakeprogressevenwhenyou'refeelingunmotivated.It'sgenerallyagoodideatoworkinthelabatleast20hoursaweek.Youwillbenefitfromhavingafocusedworkenvironment,beingaroundcolleagues,andbeingaccessibleifyouradvisorneedstofindyou.Ifyoufeelyoucan’tworkinthelab,trytofigureoutwhy,anddosomethingaboutit(e.g.,talktoyouradvisor).

16. Prioritize.Youwillfindthatyouhavemanyopportunities,andnotenoughtimetopursueallofthem.Trytofigureoutyourpriorities(e.g.,research,classes,service,sociallife,etc.)andmakesureyouarespendingyourtimeaccordingly.It’sagoodideatodoatimeauditifyoufeelthingsaren’tmatchingup.Learntodistinguishtheimmediatefromimportant--don’tfocusonimmediatedeadlines(e.g.,homework)totheexclusionordetrimentoflonger-term,moreimportantdeadlines(e.g.,conferences).Havingasmallnumberofhighqualitypapersisbetterthanmanylowqualitypapers.

17.Makeyourown“nextactions".Ifyou'resittingaroundwaitingforyouradvisortotellyouwhattodonext,somethingiswrong.Findsomethingtodo,e.g.,readpapers.

18. Keepalog.Keepadailylogofeverythingyoudoandeverythingyouthink.It'sagoodideatomakesureyourlogissearchable(e.g.,plaintextoruseanotetakingprogram).

19. Gettingthingsdone.Read(andpreferablyimplement)DavidAllen’s"GettingThingsDone".Itdoesn’tworkforeveryone,buttheideasareverygoodandworthconsidering.

20. Asociallife.Youneedtobehappytobeproductiveandmanageyourworklifeeffectively.Beinghappyusuallyinvolveshavingasociallife.Takethetimetoensureyouarehappyandhealthy,oryouwon’tbeasproductiveoreffectiveinyourworklife.

21. It’sokaytogetstuck.RememberthatEVERYONEgetsstuck/demoralized/etc.No,really.Evensuper-famous,successful,seemingly-perfectresearchersgetstuck/demoralized/etc.Whatmakesthemsuccessful,however,isthattheyfigureouthowtomovepasttheselowpointstothenextgreatidea.

22. Learnfromyourmistakes.Failingisfine(andarguablyanimportantkeytosuccess).Failingwillalsohappenoften.Thequestioniswhatyoudoafterfailing.Takenotes.

Page 5: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

Understandwhyyoufailedandthinkaboutwhatyou’ddodifferentlynexttime.Manyawesomeresearchideascameaboutbecausesomeonefailedandthenasked“why?”

ResearchYouareaPhDstudentandaPhDisaresearchdegree.Thatmeansthatyourgoalistolearnhowtodohighqualityresearch.APhDislikeanapprenticeship--youlearnbywatchingothersanddoingyourself.Forsome,doingresearchcomesnaturally.Formost,it’snotsomethingthatjusthappens;ittakeshardworkandcarefulplanning.Thereisalottolearnabouthowtodogoodresearch,themajorpointsofwhichwewilloutlinehere.Ingeneral,looktothosearoundyou,particularlyyouradvisor,formodelsofproductiveandsuccessfulresearchers.ReadingPapersYoucan’texpecttojumpinandcontributetoanactivefieldwithoutknowingwhatotherpeoplehavealreadydoneandaredoing,whatthemainchallengesare,andhowpeopletendtothinkabouttheseproblems.Allofthisinformationcomesfromreadingpapers.

23. Read,read,read!Readmultiplepapersaweek--askyouradvisor(orothergraduatestudents)forsuggestionsifyou'reunsureofwhattoread.Youhavealottolearn(especiallyinthefirstfewyears)andyou’llprimarilydothislearningbyreading.

24. Takenotes.Makenotesabouteverypaperyouread.Makenotesatmultiplelevelsofgranularity(e.g.,onesentencesummaryoftheentirepaperallthewaydowntosentence-levelnotes).Findanotetakingsystemthatmakessenseforyou.

25. Readdeeply.Whenreadinganimportantpaper,trytounderstandwhereeveryequationcomesfrom--inotherwords,derivethemyourself.Ifyougetstuck,don'tbeafraidtoasksomeoneelse(e.g.,anothergraduatestudentoryouradvisor)forhelp.

26. Breadthvs.depth.Noteverypaperneedsbereadindetailfromstarttofinish.Sometimesyouonlyneedahighlevelview.Whenreadingapaper,figureoutwhatyouneed(i.e.,whatyouhopetoachievebyreadingthispaper)andreaditaccordingly.

PickingaResearchTopicYou’lleventuallyneedtocomeupwithideasforyourownresearch.Thisisaprocessforwhichyouradvisorwillbeinvaluable.Wehavesomegeneraladviceonpickingaresearchtopic:

27. Knowtheliterature.Youneedtoknowwhat’sbeenpreviouslyinordertomakesureyourcontributionsareactuallynovelanduseful.Knowwhat'sbeendonebyotherssoyoudon'twasteyour(andyouradvisor's)timereplicatingwell-studiedideas.

Page 6: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

28. Knowthecommunity.Papersarenotpublishedinavacuum,rathertheyarepartofanongoingdialoguewithinacommunity.Knowthatcommunity,i.e.,theparticipants,priorwork,terminology,etc.Youhavetoknowwhoyouaretalkingtoandhowtotalktothem.

29. Thinkbig.Youshouldn’tbetryingtosolveeasily-solvableproblems.Focusonsolvingbigproblems,evenifyouenduptakingsmallstepstowardsasolution.Therearemanyideasthatmakeforobviouspapersthatconstitutesimpleextensionsofpreviouswork.Whilethesepapersmaygetpublished,theydon’tusuallyamounttomuch.Trytofocusonbigproblemsratherthanmakingincrementalimprovementstopreviouswork.

30. Ittakestime.Goodresearchideasdon’thappenalongeveryday.Itmaytaketimeforyoutocomeupwithabigidea,andthat’sperfectlyfine.Ultimately,youwillbejudgedonwhatyoupublish,nothowlongittookyoutocomeupwiththeidea.

31. Don’tmakeupproblemsthatdon’texist.Comingupwithanewproblemisgreat,butmakesureit’srealproblem.Ifyoucan’tthinkofatleastthreeexamplesofyourproblem,thenitsprobablynotarealproblem.Putdifferently,don’tcomeupwithhammersandthengolookingfornails.Learnhowtodesign/createhammersforactualnails.

TheResearchProcess

32. FirstYear.Asafirstyear,findaseniorstudentwithaninterestingresearchprojectandoffertohelp.Thiscangiveyouthechancetoworkonagoodideaandlearnfromsomeonewhoisknowledgeableaboutthearea.Thiswillalsohelpyouworktowardsapublicationearlyoninyourcareerandlearnfirst-handhowtodogoodresearch.

33. Startwithwriting.Whenyouhaveanidea,startbywritingitdown.Workoutthedetailsonpaperfirstbeforeyouwriteanycode.Thiswillhelpexposeproblemsandfleshoutthedetails.Thisisespeciallyhelpfulwhenworkingonmathematics.Whenworkingonapaper,writeanoutlinebeforewritinganytextsoyouknowwhatyouaretryingtodo.

34. Learnwhentoquit.Youcanlearnfromfailure,butonlyifyoumoveon.Ifyoufindyourselfrepeatedlyloweringyourgoalsandexpectations,youaren’tmakingprogress.Learnwhentoquitandrelyonyouradvisor,whoprobablyknowsbetterthanyoudo.

35. Don’tbedeadlinefocused.Thereisalwaysanotherdeadlinearoundthecorner.Publishinterestingworkwhenit’sreadytobepublished.It’swonderfultousedeadlinesasmotivation,butyoushouldn’tpublishjustbecauseyoureceivedaCFP.

36. Don’tleavethewritingtotheend.Startwritingthepaperasearlyaspossibleandaimtogetresultswellbeforethedeadline.Writingwillhelpyouplanyourworkandthinkthroughyourhypothesesandarguments.Evenonceyouhaveadraftwithinitialresults,thereisplentyofworktobedoneinordertoturnthisdraftintoahighqualitypaper.You

Page 7: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

won’tfinishthisworkifyouleavethewritingtothenightbeforethedeadline.

37. Askquestions.Don'tbeafraidtoaskquestionsand/oraskforhelp--you'lllearnfasterandbemoreproductivethanifyoutrytofigureabsolutelyeverythingoutforyourself.

38. Implement.Youunderstandbestwhenyouimplement(understanding=intuition+math+code).Ifyoucan,implementthingsmorethanonce(e.g.,usingtwodifferentmethods,orintwodifferentlanguages)andcheckyourimplementationsgiveidenticalresults.

39. Versioncontrol.Useversioncontrolforeverything(notes,code,papers,etc.).No,really.Itmaybetime-consumingtosetup,butitwillsaveyoutimeinthelongrun.

Getting(andPresenting)GoodResults

40. Knowyourdata.Knowyourdatareallywell.Makesureitexhibitsthepropertiesyouthinkitdoes.Performexploratorydataanalysis--plot/visualizeitinvariousdifferentways--sometimesyou'llfinditexhibitsinterestingpropertiesthatyouhadn'texpected.

41. Knowyoursoftware.Makesureyouunderstandwhatthesoftwarepackagesyou'reusingaredoing.Ifyou'republishingresultsobtainedusingsomeoneelse'ssoftware,youshouldknowforSUREthattheirsoftwaredoeswhatyouthinkitisdoing.There'san“easy”waytodothis:readthesourcecode.Ifthere’snosourcecode,bewary.

42. Goodbaselines.Beatingbaselinesisgood,butonlyiftheyareworthbeating.Additionally,yourideaisonlyusefulifitisanimprovementoverexistingorsimplerapproaches.Learnhowtocomeupwithconvincing,effective,andSIMPLEbaselines.Alwaysaskyourself,"What'sthesimplestexperimentIcoulddoto(in)validatemyhypothesis?”Talentedresearchershaveaknackforcomingupwithsimplebaselines.

43. Understandyourresults.It'snotsufficienttoknowthatyourmethodgets95%accuracyonyourdata.Youalsoneedtoknowexactlywhat'shappeningonthe5%ofdatapointsforwhichyourmethodDOESN'Twork.Lookatactualdatapointsthatyourmethodishandling(in)correctly,plot/visualizeyourresultsinvariousdifferentways,etc.Thisexercisewillbeusefulwhenpresentingyourworkandwhenimprovinguponit.

44.Makeyourresultsaccessible.Learnhowtopresentresultssuchthattheyareaccessible,useful,andconvincing.Atableisrarelythemostappropriatewaytopresentnumericalinformation--it’sagoodideatoreadatleastonebookabout(quantitative)informationvisualization.Yourresultsareonlyconvincingiftheyareunderstandable.

Publishing

45. Keeptrackofdeadlines.Don'trelyonyouradvisortodothisforyou--they'realready

Page 8: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

keepingtrackofazillionotherdeadlines.Ifyouthinkyou'renotontracktomeetadeadline,DOtelltoyouradvisor--they'dratherknowsoonerthanlater.

46. Finishwritingearly.Finishyourpaper(atleast)aweekbeforethedeadline.Notonlywillthisgiveyoutimetopolishyourwriting,getfeedbackfromothers,andrunanyexperimentstheysuggest,butitmakesitmorelikelythatyou'llactuallygetanyusefulfeedbackfromyouradvisor,whoprobablyhasseveralotherpaperstoreadbeforethedeadline.It’snothelpfultogetgreatfeedbackfromacolleagueabouthowtoframeyourideaswhenyouonlyhaveafewhoursleftbeforethepaperisdue.

47. Learnhowtowritewell.Ifyoueverthoughtthatyoudidn’tneedtowritewellbecauseyouwereanengineerorscientist,youwere100%wrong.Asascientist,it'syourjobtocommunicateyourideastoothers.Itdoesn'tmatterhowamazingyourworkis,it’sunlikelytohaveanyimpactifnoonecanunderstandyourexplanations.Ifyoufindyourselfthinking“thereviewersdidn’tunderstandmypaper,”maybeyoudidn’texplainyourideasclearlyenough.Awell-writtenpapercanmakethedifferencebetweenarejectandanaccept,orbetweenagoodpaperandanaward-winningpaper.Ifyouwanttoimproveyourwriting,you’llneedtogetearlyfeedbackonyourdrafts(seeabove.)

48. Reproduceyourresults.Partofpublishingisattestingtotheaccuracyofyourpublishedresults.Thatmeansyoumustbeabletoreproducethem.Makesureyoukeepdetailednotesonhowyouobtainedtheresultsinyourpapersothatyoucanrepeattheexperiments.Attheminimum,thiswillbehelpfulforrunningadditionalexperimentsbasedonreviewerfeedback(orwhenyouneedtowriteyourthesis.)

49. Reorganizeaftersubmission.Organizeanddocumentyourcode,results,etc.IMMEDIATELYafterapaperdeadline.Don’tkidyourself--ifyoudon'tdoitthen,it'snevergoingtohappen.Again,youwillbegratefulyoutookthetimetodothiswhenyoufindyouneedtorerunyourexperiments(e.g.,foryourthesis).

50. Releasecodeanddata.Releaseanycodeordatathatiscentraltoyourpublication.Thisensuresthatotherscanreproduceorbuildonyourideas.Furthermore,ifothersusethecode/data,theywillciteyou.Whatmorecouldyouwant?!

51. Qualityandnotquantity.Youwillbejudgedbasedonthequality,andnotthequantityofyourpublications.YouwillfindendlessCFPsforconferences,books,journals,etc.Mostofthesewon'tmattertoyourcareer.Nooneisimpressedbypublicationsinsomerandomjournal.Takethetimetogetahighqualitypaperintoaconferencethatmatters.

52. Publishorperishdoesn’tstartinyourfirstyear.Withincreasingpressuretopublish,somestudentsthinktheyneedapaperintheirfirstyear.Youdonot.Itisunlikelythatafterlessthanayearyouwillhaveanimportantandnovelideatocontributetothecommunity.Learnsomethingfirstsoyourfirstcontributionisworthwhile.

Page 9: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

TalksWecannotoverestimatetheimportanceofgivinggoodtalks.Agoodtalkcanmakethedifferencebetweenpeoplereading/citingyourconferencepaperandpeopledismissingit.Furthermore,knowinghowtogiveagoodtalkwillhelpyougetagoodjobaftergraduateschool.AsaPhDstudent,youmustlearnhowtogivegoodtalks,sostartearly.

53. Practice.Thesinglebestwaytolearnhowtogivegoodtalksistopractice.Practiceinfrontofthemirror,infrontoffriends,colleagues,etc.Itisnotunusualtopracticeanimportanttalkseveraltimes.Also,findopportunitiestogivetalks.Ifyourschoolhasastudentseminar,volunteertospeak.Theseforumswillallowyoutoimproveyourpresentationskillswellbeforeyouhavetogiveconferenceandjobtalks.

54. Askforfeedback.Ifyougiveatalk(eitherapracticetalkorarealtalk)askyouraudienceforfeedbackonclarity,style,content,presentation,etc.

55. Spendtimeoncontent.WithPowerPoint,Keynote,etc.,it’seasytospendlotsoftimeonthelookandfeeloftheslides.However,thisisfarlessimportantthanhavingaclearoutlineandclearwaysofpresentingyourcontent.Spendyourtimeonwhatyouwanttosayandhowyouwanttosayitbeforeyouworkonfancyanimations.

56. Timing.Therightamountofmaterialintherightamountoftimeisthehallmarkofagoodtalk.Themomentaspeakersays“Idon’thavetimesoIamgoingtorushthroughthis,”halftheaudiencetunesout.Ifyoudon’thavetimetocoversomething,skipitentirely(orbetteryet,don’tputitinthetalkinthefirstplace).Itisbettertocoverasingletopicwellthanmanytopicspoorly.Timeyourselfwhenyoupracticeyourtalk.Usethesetimingstomakenoteswhichwillhelpyoudecideifyouaremovingtooslowlyorquicklyduringtheactualtalk.Doyoutalkfasterorslowerduringpracticeascomparedtotheactualpresentation?Figurethisoutsoyouknowhowtoadjustyourtimingsaccordingly.

PostersPosterpresentationscanbeanimportantpartofthepublicationprocess.Therearemanydifferentapproachestomakinggoodposters,sowe’lloutlineafewhighlevelpoints.

57. Startaconversation.Aposterissupposedtoassistyouinstartingandmaintainingaconversation.Yourpostershouldthereforeadvertiseyourworkandpersuadepeoplethattheywanttotalkwithyou.Itshouldalsocontaininformationthatwillhelpyoutalkaboutyourwork.Aposterisnotareplacementforyourpaper.It’sokaytoomitdetail.

58. Visuallyattractive.Peoplevisitpostersthatlookniceandaren’tjustamessofmath.

Page 10: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

Youdon’tneedtobecomeanartist,butlearnhowtousegraphicsandcolors.

59. Tellastory.Justasatalkshouldhaveaclearstoryline,soshouldyourposter.Youwilluseyourposterasaguidewhentalkingtovisitorsaboutaboutyourwork.Makesuresureithastherightcontentintherightorder(seebelow).

60. Practice.Justasit’simportanttopracticetalks,it’simportanttopracticepresentingyourposter.Practicingyourpresentationwillhelpyouworkoutwhetheryourpostercontainstherightcontent,attherightlevelofdetail,arrangedintherightorder.

ProfessionalDevelopmentProfessionaldevelopment,networking,and(ultimately)findingajobareimportant.

61. Dointernships.Industrial(andacademic)internshipsaregreatwaystolearnnewskills,meetnewpeople,workonnewproblems,andmakesomeextramoney.Therearedifferentschoolsofthoughtontiming,however.Oneapproachistodoaninternshipsearlyoningraduateschool,whenyoudon’tyethavearesearchtopicandhaven’tbuiltupmomentumonaspecificproblem.Anotherapproachistodoaninternshiplater,whenyouhaveaspecializedsetofskills/interestsandareclosertolookingforajob.Botharegoodapproachesanditdoesn’thurttodomorethanoneinternship.

62. Reviewpapers.Startreviewingpapersinyourresearcharea.Offertohelpyouradvisorwithpaperreviews--theywillalmostalwaystakeyouuponyouroffer.Askyouradvisorforfeedbackonyourreviewssoyoucanimproveyourreviewingskills.

63. Givetalks.Learningtogivegoodpresentationsisveryimportant(seeabove).Onebenefitofgivingtalksisthatdoingsoadvertisesyourworkandmakessurepeopleknowwhoyouare.Beingwell-knownwillpayoffwhenyouarelookingforajob.

ProgressingThroughGraduateSchoolAsyoulookaroundatotherstudents,you’llprobablywonderhowyouaredoing.Don’tjustspeculateandtrytocompareyourselftoothers--askyouradvisor.WhileCSPhDsusuallytake5-6years,plentyofpeopletakelonger.It’snotnecessarilybadtotake7years.Rememberthatyouwon’tbejudgedonhowlongyouspentinyourPhDprogram,butbywhatyouproduced.Roughlyspeaking,hereiswhattoexpectintermsofprogress.

64. Stage1:Takeclasses,findanadvisor,watchothersdoresearch,readpapers,completeaprojectwithresults(thoughnotnecessarilyapaper).Usually2-3years.

65. Stage2:Pickageneralareaofinterest,leadresearchonaprojectinthisarea,publisha

Page 11: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

paper,makeacontributiontothisarea.Usually2-4years.

66. Stage3:Evaluateyourwork,comeupwithastory,writeathesis.Usually1year.Networking

67. Tutorials.Writetutorials/annotatedbibliographies/technicalnotes.Ifthey’regood,thiscanbeahighlyeffectivewaytomakesureyournameisknownwithinyourcommunity.Thinkofallthetutorialsyou’vereadbywell-knownacademics.

68. Bignames.Knowwhothe"bignames"areinyourareaandfollowtheirworkclosely.

69. Gotoconferences.Evenifyouaren’tpresentingapaper,trytogotoaconferenceeveryyear.(Rememberthatyouradvisormaynotalwayshavemoneytosendyou,however.)Ifyoufeeluncomfortableaboutnotpresentingapaper,rememberthatyoucantellpeople,"Oh,I'mnotpresentingapaperhereTHISyear."

70. Introduceyourself.Don’tbeshyaboutintroducingyourselftonewpeople(juniorandsenior)atconferences,evenifyoufindthishard.Ifyou’renotsurewhattosay,askthemtotellyouabouttheirwork.Ifithelps,askyouradvisortomakeintroductions.

71. Actprofessionally.Youractionsreflectnotonlyyourself,butalsoyourlabandyouradvisor.Additionally,it'spossiblethatyouwillcontinuetointeractwiththeresearchersinyourcommunityfordecadestocome--makingsureyouestablishareputationassomeonewhoactsprofessionallyisthereforeaninvestmentinyourfuture.

FindingaJobFindingajobisacomplexprocessandbeyondthescopeofthisdocument.Therearemanygoodguidesonthistopic.We’llleaveyouwithtwothingstokeepinmind,however:

72. Youarenotalone.Donotexpecttofindajoballbyyourself.Partofyouradvisor’sjobistohelpyoufindemployment.Starttalkingwithyouradviserearlyaboutyourcareergoalsandyourjobsearch.Workwithyouradvisortodevelopaplan.Youcanalsorelyonyourothercontacts,suchasyourcommitteemembersandyourinternshiphosts.

73. Successmeansmanythings.Manystudentsthinkthatanacademicresearchjobisthetruemeasureofsuccess:ifyoudon’tbecomeaprofessor,youhaven’tsucceeded.Thisistotalnonsense.Noreally.YouaregettingaPhDbecauseofthetrainingitprovides.Academicresearchisoneareathatbenefitsfromorrequiresthistraining,buttherearemanymanyothers:teaching,industrialresearch,governmentresearch,sciencepolicy,financialinvestment,entrepreneurship,etc.DuringyourPhD,trytoworkoutwhatyoulovetodo.Then,afteryourPhD,findajobthatwillletyoudothat.For

Page 12: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

some,thisisteachingandacademicresearch,butitisridiculoustothinkthatistherightcareerforeveryonewithaPhD.Weknowmanybrilliantpeoplewhochooseanon-academiccareerpathbecauseit’swhattheywantedtodoanditmakesthemhappy.Ultimately,youneedtodowhatwillmakeyouhappy.

ThingstoLearnMasteringthefollowingtechnicaltoolswillhelpyoueffectivelyconductsuccessfulresearch:

1. Bashconfiguration(e.g.,.bashrc,environmentvariables,aliases)2. PATH,CLASSPATH,PYTHONPATH3. Unixbasics(e.g.,cd,ls,mv,rm,rmdir,man,history,ctrl+r,4. ctrl+a,ctrl+e,ctrl+←andctrl+→,killringvs.clipboard)--see

http://freeengineer.org/learnUNIXin10minutes.htmlforastartingpoint5. Basicbashscripting(e.g.,sort,uniq,cut,tr,wc,xargs,find,grep,awk,sed)6. ctrl+c,psauxandgrep,kill-9,ctrl+zandbg,fg,top7. diff8. Howtoinstallsoftwarelocally9. Emacsorvi10. EclipseorIntellij11. Screen(andwhyyouneedit)12. Make13. BasicRplottingcommands14. Gridengine(e.g.,qsub,qstat,qlogin)15. Versioncontrol(e.g.,git)16. Howtouseadebugger(e.g.,jdb)

OtherResourcesAfewexcellentarticlesonrelatedtopics:

1. LawrenceSaul’sadvicefornewgraduatestudents:http://www.cs.ucsd.edu/~saul/start_research.pdf

2. 3qualitiesofsuccessfulPhDstudents:http://matt.might.net/articles/successful-phd-students/

3. 10reasonsPhDstudentsfail:http://matt.might.net/articles/ways-to-fail-a-phd/4. HOWTO:Sendandreplytoemail:http://matt.might.net/articles/how-to-email/.Seeother

articlesfromhttp://matt.might.net/articles/5. AdvicefromJasonEisner:http://www.cs.jhu.edu/~jason/advice/--includinghowtoread

apaper(http://www.cs.jhu.edu/~jason/advice/how-to-read-a-paper.html),howtofindresearchproblems(http://www.cs.jhu.edu/~jason/advice/how-to-find-research-problems.html),andwhyyou

Page 13: How to Be a Successful PhD Student - Google Docsmdredze/publications/HowtoBeaSuccessful...How to Be a Successful PhD Student (in Computer Science (in NLP/ML)) Mark Dredze (Johns Hopkins

shouldwritethepaperfirst(http://www.cs.jhu.edu/~jason/advice/write-the-paper-first.html)

6. Adviceforseniorgradstudents:http://www.cs.umbc.edu/~mariedj/papers/advice-summary.html

7. Howtowritewell:http://www.amazon.com/Manual-Writers-Research-Dissertations-Seventh/dp/0226823377/ref=sr_1_1?s=books&ie=UTF8&qid=1316904475&sr=1-1

8. Howtobeagoodgraduatestudent:http://www.cs.indiana.edu/how.2b/how.2b.html9. Howtobeagoodgraduatestudent:

http://www.deirdremccloskey.com/docs/pdf/Article_315.pdf10. TheNationalPhysicalScienceConsortiumGraduateFellowStudentHandbook(Someis

administrativestuffapplicableonlyforNPSCfellows,butalotofgeneraladvicetoo)http://www.npsc.org/resources/GradStudentHbk.pdf

11. CRA-Wgraduatestudentinformationguide(shouldyougo,howtoapply,whatgradschoolislike):https://cra.org/wp-content/uploads/2015/02/grad-guide.pdf

12. Adviceonoutlinesandpaper-writingfromGeorgeWhitesides(highesth-indexofanylivingchemist):http://www.ee.ucr.edu/~rlake/Whitesides_writing_res_paper.pdf