valse2015如何做研究之读博经验纵横谈（ppt课件）.pdfVIP

VALSE2015

如何做研究之读博经验纵横谈

刘偲

中国科学院信息工程研究所

ResearchExperience

•ObtainPh.DfromNLPRinChineseacademyofsciences

–Supervisor：Prof.LuHanqing

•NationalUniversityofSingaporeLearningVisionGroup(2009-2014)

–Supervisor：Prof.YanShuicheng

•InstituteofInformationEngineering,ChineseAcademyofSciences

Timespent

•TheemailfromPuMuming(蒲慕明)tohisstudent:

–Imentionedthatabout60hrworkingtimeperweekiswhatIconsidertheminimal

timeanaveragesuccessfulyoungscientistinthesedayshastoputintothelabwork

–Isuggestthateveryoneputsinatleast6hrconcentratedbenchworkand2+hrreading

andotherresearch-relatedactivityeachda

–Readingpapersandbooksshouldbedonemostlyafterwork

•《施一公谈如何做科研》：

•每天晚上做实验到半夜三点左右，回到住处躺下来睡觉时常常已是四点以后；

•九点左右又回到实验室开始了新的一天

•每天三餐都在实验室，分别在上午9点、下午3点和晚上9、10点

•这样的生活节奏持续11天

•在纽约工作整整两年，我从未迈进中心公园一步。

Change

the

way

of

thinking

•The

differences

between

undergraduate

and

graduate

students

–Undergraduate:

•Learnthe

existing

knowledge

–Graduate:

•Explore

and

create

new

knowledge

•New

Problem:

Clothes

Search,

Fashion

Recommendation

•Correctly

analyze

negative

results

is

the

key

to

the

success.

–Do

not

be

afraid

of

negative

results

Performance

from

10%+

to

60%+

Change

the

way

of

thinking

•Perfectionism

hinders

innovation

–Go

ahead

to

the

end

–Come

back

and

refine

each

step

The

performance

of

Semantic

SegmentationWe

can

design

an

human

parsing

algorithm

even

though

the

is

not

sensitive

to

the

accuracy

of

over-segmentationhuman

pose

estimation

is

inaccurate