CanadaでのComputer ScienceのPhD進学に向けて/Towards CS PhD in Canada/

CanadaでのComputer ScienceのPhD進学に向けて/Towards CS PhD in Canada/

2020年「海外大学院留学説明会」での発表資料です
https://www.titech.ac.jp/event/2020/047271.html
https://gakuiryugaku.net/seminar/1963

1b6d585eaf0c1ff4d0cc7637295aa50f?s=128

Hiroki Naganuma

July 12, 2020
Tweet

Transcript

  1. Computer Science and Operations Research,Université de Montréal Mila - Quebec

    Artificial Intelligence Institute PhD Student Hiroki Naganuma/௕পେथ July.12.2020  Computer Science ઐ߈ͷւ֎ PhD ਐֶʹ޲͚ͯ @2020೥Ն ւ֎େֶӃཹֶઆ໌ձ(ถࠃେֶӃֶੜձओ࠵)
  2. Hiroki Naganuma/௕প େथ Sep 2020- Incoming Ph.D. student in Computer

    Science @Université de Montréal, Mila - Quebec Artificial Intelligence Institute Research Interests • (Application) Fast, Scalable and Robust Optimization Method for Deep Learning • (Theory) Theoretical and Algorithmic Foundations of Deep Learning 2013 2017 2019 B.E. in CS @Tokyo Institute of Technology M.E. in CS @Tokyo Institute of Technology Apr 2020 Education and Research Experience Student Fellow @ IBM Research Tokyo Student Internship @RIKEN AIP PhD Student in CS @Tokyo Institute of Technology Apr Sep Apr PhD Student in CS @Université de Montréal, and Mila - Quebec Artificial Intelligence Institute Apr 2 /18
  3. Université de Montréal • Χφμͷࠃཱେֶ(1878೥ʹઃஔ; ౦޻େ͸1881೥) • 14ͷֶ෦ͱ෇ଐߍ(ֶੜ਺66,500ਓҎ্; ౦޻େ͸໿10,000ਓ) •

    ੈքτοϓϨϕϧ(ಛʹਂ૚ֶशͷཧ࿦෼໺ͷϝοΧ) CS͸ੈք31Ґ(೥ʑ্ঢ)ɺMila - Quebec Artificial Intelligence Institute ͷڌ఺: ਂ૚ֶश෼໺ʹ͓͚Δޭ੷Ͱ2019೥ʹνϡʔϦϯά৆Λड৆ͨ͠ Yoshua Bengioࢯ͕ઃཱ) • Mila - Quebec Artificial Intelligence Instituteͷઃஔʹ൐͍Google, Facebook, Microsoft, IBM, DeepMindͳͲͷਂ૚ֶशؔ࿈ͷݚڀ෼໺ʹ ྗΛೖΕΔاۀͷݚڀڌ఺͕ઃஔ͞ΕΔ • ϑϥϯεޠͷେֶͱͯ͠ϑϥϯεͷύϦେֶʹ͍ͭͰੈքͰ2൪໨ͷن໛ 3 /18
  4. • Χφμͷਓޱɾܦࡁن໛Ͱୈೋͷ౎ࢢ(ҰҐ͸τϩϯτ) • έϕοΫभͳͷͰެ༻ޠ͸ϑϥϯεޠͷΈ • ๺ถͷύϦ(ொฒΈ΍จԽͳͲɺ৯ࣄ͕๺ถͷதͰ͸ඒຯ͍͠ํΒ͍͠) • Université de MontréalͷଞʹMcGill

    University͕༗໊ Montréal 4 /18
  5. ౦޻େͰͷࢦಋڭ׭ɺಌΕΔݚڀऀͷӨڹɾͦͷଞͷӨڹ ཹֶΛ໨ࢦͨ͠ཧ༝ ໘઀४උɾϝʔϧίϯλΫτ ग़ئޙͰ͖Δ͜ͱ ঑ֶۚɾਪનঢ়ɾޠֶࢼݧɾGPAɾCVɾSoPɾίϯλΫτ ւ֎େֶӃग़ئͷ४උ Outline 5 /18

  6. ཹֶΛ໨ࢦͨ͠ཧ༝ 1/3 ౦޻େͷֶ෦࣌୅͔Βͷࢦಋڭ׭(ԣాཧԝઌੜ)ͷӨڹ ւ֎ͰͷϙευΫɾݚڀ৬Λܦݧ͖ͯͨ͠ࢦಋڭ׭ʹӨڹΛड͚ͨ => ਪનঢ়Λॻ͍͍͖ͯͨͩɺֶ෦࢛೥ͰߴੑೳܭࢉͷτοϓձٞͷϓϩάϥϜʹࢀՃͤͯ͞΋Βͬͨ • ࠃࡍతʹ׆༂͢Δԣాઌੜͷ࢟ΛؒۙͰݟͯɺ୯७ʹಌΕͨ • ߴੑೳܭࢉτοϓձٞʹͯւ֎ͷ༗໊େͷಉ೥୅ͷ࣮ྗʹڻ͔͞Εͨ

    6 Reference: ԣాཧԝ, “GPU ίϯϐϡʔςΟϯά์࿘ه”, Ұൠࣾஂ๏ਓ ిࢠ৘ใ௨৴ֶձ ৘ใɾγεςϜιαΠΤςΟࢽ ୈ23ר ୈ2߸ʢ௨ר 91 ߸ʣ /18
  7. ཹֶΛ໨ࢦͨ͠ཧ༝ 2/3 ౦޻େ΁ਐֶ͖͔͚ͨͬ͠ͱͳͬͨݚڀऀ(ླଜ๛ଠ࿠ઌੜ)ͷӨڹ Reference: ླଜ๛ଠ࿠, “େن໛ฒྻ෼ࢄγεςϜʹ͓͚ΔϏοάσʔλղੳͱࣾձγϛϡϨʔγϣϯͷݚڀ”, Ұൠࣾஂ๏ਓ ਓ޻஌ೳֶձ ֶձࢽ ਓ޻஌ೳ

    31 ר 4 ߸ʢ2016 ೥ 7 ݄ʣ 2013೥౰࣌ͱͯ΋ඞཁͱ͞Ε͍ͯ ٕͨज़(ࠓ΋ࠓޙ΋ॏཁ) େن໛ͳ࣮ࣾձσʔλΛͲͷΑ͏ ʹߴ଎ʹϦΞϧλΠϜͰॲཧ͢Δ ͔ͷݚڀ ౦ژ޻ۀେֶ ٬һ।ڭत (2009-2013) IBM౦ژجૅݚڀॴ ݚڀһ 2015೥ΑΓIBM T. J. Watsonݚڀॴ ΋ͱ΋ͱ͸ɺླଜઌੜͷߦΘΕ͍ͯͨؔ࿈ͷݚڀ͕ͨͯ͘͠౦޻େʹAOೖࢼͰೖֶͨ͠(2013೥) ( ∵ ߴߍͷଔ࿦Ͱେن໛σʔλॲཧʹؔ͢ΔτϐοΫΛѻͬͨ) ͔͠͠ɺݚڀࣨ഑ଐલʹླଜઌੜ͕ւ֎ʹҠಈ͞ΕͨͷͰࢦಋ͸ड͚ΒΕͣ ͍͔ͭಉ͡ϑΟʔϧυʹཱͬͯΈ͍ͨͱ͍͏ಌΕͱɺݚڀऀͱͯ͠ͷ࢟੎ʹײ໏Λड͚ͨ Dublin, Ireland IBM Thomas J. Watson Research Center 7 /18
  8. ཹֶΛ໨ࢦͨ͠ཧ༝ 3/3 ͦͷଞͷӨڹ(ֶ಺֎ͷϓϩάϥϜͳͲ) 1. ϦΫϧʔτϗʔϧσΟϯάεओ࠵ Silicon Valley Worksop (ֶ෦3೥) =>

    StanfordେֶͰͷϨΫνϟʔɺStanford Duck Syndromeͱ͍͏ݴ༿Λ஌Δ 2. ࠃࡍϓϩάϥϛϯάܥίϯϖ΁ͷࢀՃ (म࢜1೥) => Microsoft Imagine Cup΍ Hackathon@StanfordͰࣗ෼΋ւ֎Ͱ௨༻͢Δ͔΋ͱࣗ৴Λ͚ͭΔ 3. Mercari Πϯλʔϯ @San Francisco - US, @London - UK (म࢜1೥) => ݱ஍ֶੜͱσΟεΧογϣϯ͢Δػձɺ೔ຊͷେֶӃͱͷҧ͍ʹ৮ΕΔ 4. ౦޻େ δϣʔδΞ޻ՊେϦʔμʔγοϓɾϓϩάϥϜ (म࢜2೥) => δϣʔδΞ޻ՊେͰ̎िؒաͯ͝͠ΈΔɺࣗ෼ΑΓͣͬͱؤுͬͯΔ೔ຊਓʹܹࢗΛड͚Δ 5. JTओ࠵ ಥ͖ൈ͚Δਓࡒθϛ (म࢜2೥) => ໜ໦݈Ұ࿠ઌੜɾ೾಄྄ઌੜ(ܦࡁධ࿦Ո)ɾඌݪ࿨ܒ͞Μ(IT൷ධՈ)͔ΒͷϑΟʔυόοΫ ੒ޭ = ଧ਺ x ௕ଧ཰ ඌݪ࿨ܒ͞Μ ʮͲͷΑ͏ͳλάͰଧ੮ʹݺ͹ΕΔଘࡏʹͳΔ͔ɺ ͦͷͨΊͷλάઃܭɾ୹ظͰࣗ෼ͷλάͰҹ৅͚ͮ৴པؔ܎Λ࡞Δ͜ͱ͕ॏཁʯ ߴ͍௕ଧ཰Λૂ͑Δւ֎ͷPhDਐֶʹ޲͚ͯ ४උ͢Δ͜ͱΛܾҙ(म࢜2೥ͷ9݄) 8 /18
  9. ౦޻େͰͷࢦಋڭ׭ɺಌΕΔݚڀऀͷӨڹɾͦͷଞͷӨڹ ཹֶΛ໨ࢦͨ͠ཧ༝ ໘઀४උɾϝʔϧίϯλΫτ ग़ئޙͰ͖Δ͜ͱ ঑ֶۚɾਪનঢ়ɾޠֶࢼݧɾGPAɾCVɾSoPɾίϯλΫτ ւ֎େֶӃग़ئͷ४උ Outline 9 /18

  10. ւ֎େֶӃग़ئ४උ(શମεέδϡʔϧ) 2018 2019 2020 2021 Ұ೥ؒͷΠϯλʔϯͰͷ ੒ՌΛ·ͱΊΔ 2022 म࢜࿦จ ఏग़

    Χφμ େֶӃग़ئ ւ֎େֶӃ ത࢜՝ఔޙظ։࢝ 4 9 2 4 12 म࢜࿦จݚڀͱɺۀ੷ͮ͘ΓͷͨΊͷݚڀ ւ֎େֶӃ ത࢜՝ఔޙظमྃ 9 ത࢜࿦จ ࣥච ֶҐཹֶ(ݚڀظؒ) 9 ൒೥ͷΠϯλʔϯͰͷ ੒ՌΛ·ͱΊΔ 10 ཹֶ४උूதظؒ 2023 ͜ͷλΠϛϯάͰ ͜ͷܭըΛཱͯͨ 11 10 ւ֎େֶ ϝʔϧίϯλΫτ NeurIPS 2019 12 5 1 ΧφμେֶӃ ΦϯϥΠϯ໘઀ 1 2 ΧφμେֶӃ ߹֨௨஌ 3 Χφμ಺ఆडཧ ೖֶखଓ͖ ࠃࡍֶձͰւ֎େֶ ͷڭतͱ࿩͢ᶆ 3 ࠃࡍֶձͰւ֎େֶ ͷڭतͱ࿩͢ᶃ 1 7 IELTS͜͜·Ͱʹ 3ճड͚ͨ ࠃࡍֶձͰւ֎େֶ ͷڭतͱ࿩͢ᶄ 6 ࠃ಺େֶӃ ത࢜՝ఔޙظ։࢝ ࠃࡍαϚʔεΫʔϧͰ ւ֎େֶͷڭतͱ࿩͢ᶅ ਪનঢ়ͷ ͓ئ͍Λ͢Δ →ୈҰࢤ๬ʹ߹֨ͨ͠ͷͰɺ2݄຤͔Β3݄ʹग़ئΛ༧ఆ͍ͯͨ͠ΠΪϦεɾεΠεͷେֶӃʹ͸ग़ئͤͣ /18
  11. ւ֎େֶӃग़ئ४උ(TODO) 11 1. ΍Γ͍ͨݚڀΛܾΊΔ 2. ߦ͖͍ͨݚڀࣨɾେֶΛߜΔ [8݄·Ͱɺૣܾ͘·Ε͹ૣ͘ίϯλΫτ͕औΕΔ] => USͰ5೥PhDΛ͢ΔͷͰ͸ͳ͘ɺεΠεɾΧφμɾUKͷେֶӃͰ3-4೥ͰͷPhDऔಘΛ໨ඪͱ͠·ͨ͠ 3.

    ߦ͖͍ͨݚڀࣨʹίϯλΫτΛͱΔ[ASAP] => جຊతʹ࠷ॳ͸ϝʔϧͰɺ70%͘Β͍ͷฦ౴཰ 4. ޠֶࢼݧͷج४ΛΫϦΞ͢Δ(ͦΕͳΓʹ͕͔͔࣌ؒΔ)[ग़ئ·Ͱ(12݄)/UK͸ೖֶ·Ͱ] => ΞϝϦΧΛड͚ͳ͍ͷͰGRE͸ະडݧɺIELTSҰຊʹߜͬͨ(TOEFL΋Ұ౓ड͚͕ͨਆܦ࣭ͳํʹ͸߹Θͳ͍ͱࢥ͏) 5. CVΛڧ͘͢Δ(͕͔͔࣌ؒΔ) => PublicationͳͲΛ૿΍͢ɺCSܥͰ͸GithubͷϨϙδτϦఏग़ΛٻΊΔͱ͜Ζ΋ 6. ঑ֶۚΛ֬อ͢Δ [Ұൠʹ͸10݄ࠒ·Ͱ] => (ࢲͷ৔߹͸ಛघͰ)ଙਖ਼ٛҭӳࡒஂͷ঑ֶۚΛ֬อͰ͖͍ͯͨ 7. SoPΛॻ͘ [6݄ʹҰ౓هೖɺ11݄ࠒʹमਖ਼] => ֶৼ༻ͷ΋ͷΛӳޠͰ࠶ߏ੒͠ɺ֤େֶݚڀࣨ΁ͷग़ئ޲͚ʹॻ͖׵͑ͨ 8. ߴ͍GPAΛऔΔ => ͜Ε͸࠷ॳ͔Β΍ͬͯͳ͍ͱ೉͍͠ 9. ਪનঢ়Λॻ͍ͯ΋Β͑ΔઌੜΛݟ͚ͭΔ[8݄·Ͱɺ঑ֶۚͷԠืʹ΋ඞཁ] => ౦޻େͷࢦಋڭ׭ɺIBMɾRIKEN AIPͰͷࢦಋڭ׭ͷ߹ܭࡾ໊ʹॻ͍͍͖ͯͨͩ·ͨ͠ /18
  12. 1. ࣗ෼͕΍͍ͬͯΔݚڀͰ࠷લઢͷਓͨͪ 2. Α͘࿦จͰ໨ʹ͢Δஶऀͷ͍ΔݚڀࣨͳͲ Λɺࣗ෼ͷࢤ๬͢Δࠃ΍େֶΛؑΈͯީิΛ࡞Γɺ·ͣ͸ϝʔϧͰίϯλΫτ ΋͠ɺڭतͷίωͳͲ͕͋Ε͹ͦΕΛ׆༻Ͱ͖Δํ͕ྑ͍(͕ɺࢲͷ৔߹ͦΕ͕ͳ͔ͬͨ) ࢲͷ৔߹ɺ • PhDͱͯ͠ड͚ೖΕͯ΋Β͍͍ͨࢫ •

    ͳͥڵຯΛ࣋ͬͯΔͷ͔ • ࣗ෼ͷݚڀͱݚڀࣨͷݚڀʹͲͷΑ͏ʹؔ࿈͕͋Δͷ͔ • ࣗ෼ͷCVͱɺͲͷΑ͏ʹͦͷݚڀࣨʹContributionͰ͖Δ͔ Λɺ·ͱΊͯϝʔϧ͠·ͨ͠ ·ͨɺϝʔϧͰίϯλΫτ͕औΕΕ͹ɺ࣮ࡍʹ๚໰ͨ͠ΓɺֶձͰ࿩࣌ؒ͢Λ࡞ͬͯ΋Β͏͜ͱ͕͓͢͢ΊͰ͢ • ਓ޻஌ೳֶձओ࠵ NeurIPS Ϩϙʔλʔ೿ݣ • จ෦Պֶলओ࠵ ࠃࡍֶձ౳ࢀՃิॿاը ͳͲͷػձΛ׆༻ͯ͠ɺ࣮ࡍʹ࿩ͯ͠Έ͍ͨڭतͱ࿩͢໿ଋΛऔΓ͚ͭ·ͨ͠ ࣮ࡍʹɺ࿩ͯ͠Έͯɺ͜ͷ෼໺͸΄Μͱʹ΍Γ͍ͨ͜ͱͳͷ͔ͳͲɺࣗ෼ࣗ਎ͷؾ ෇͖ʹ΋ͳΓ·ͨ͠ ւ֎େֶӃग़ئ४උ(ݚڀࣨબͼˠίϯλΫτ) 12/18
  13. ւ֎େֶӃग़ئ४උ(ޠֶɾGPAɾਪનঢ়) ޠֶͱGPA͸ࣗ෼Ͱ஍ಓʹؤுΔ͔͠ͳ͍ ΞυόΠεͱͯ͠͸ɺֶ෦࣌୅ͷGPA͕ͦ͜·Ͱߴ͘ͳͯ͘΋म࢜ͰṢճ͢Ε͹νϟϯε͸͋Γ·͢ (ࢲͷ৔߹: ֶ෦3.4/4.0, म࢜4.0/4.0, Oxfordͱ͔͸3.2ͱ͔3.4Ͱ଍੾Γ͕͋ͬͨؾ͕͢Δ) ޠֶʹؔͯ͠͸ΠΪϦεͷେֶӃ͸Conditional Offerͱͯ͠ɺೖֶ࣌·Ͱʹޠֶཁ݅Λୡ੒Ͱ͖Ε͹ྑ͍ͱ͍͏ ৚݅෇͖߹͕֨͋Δͱ͜Ζ΋͋Γ·͢

    ਪનঢ়ʹ͍ͭͯ ࣗ෼ͷ͜ͱΛΑ͘஌ͬͯΔΞΧσϛΞͷํʹॻ͍ͯ΋Β͏ͷ͕ྑ͠ͱ͞Ε͍ͯΔͷͰɺͦ͏͍ͬͨػձΛ࡞Δ • B4-M2: ಉ͡౦޻େͷݚڀࣨʹॴଐ (౦޻େ ԣాཧԝ ।ڭत; ߴੑೳܭࢉ͕ઐ໳) • M1: IBM ౦ژجૅݚڀॴͰΠϯλʔϯ (ͦͷ࣌ͷࢦಋڭ׭ɺݱࡏ͸NII ؔࢁଠ࿕ ॿڭ; ج൫ιϑτ΢ΣΞ͕ઐ໳) • M2: RIKEN AIPͰΠϯλʔϯ (౦େ ླ໦େ࣊ ।ڭत; ਺ཧ౷ܭɾਂ૚ֶशཧ࿦͕ઐ໳) ͦΕͧΕҟͳΔ؍఺͔ΒධՁͯ͠΋Β͑ΔΑ͏ͳઌੜʹ͓ئ͍ͨ͠ 13/18
  14. ւ֎େֶӃग़ئ४උ(SoP) Կ౓΋࿅Γ௚͢(ֶৼͱಉ͡ཁྖ)Ͱਓ͔ΒϑΟʔυόοΫΛड͚ΔɺωΠςΟϒͷఴ࡟Λड͚Δࣄ͕ॏཁͰ͢ ֶৼͱҧ͏఺͸ɺࢤ๬ߍผʹ෦෼తʹ಺༰Λม͑Δ఺ɺ۩ମతͳSoPͷߏ੒౳͸ࢀߟϦϯΫΛࢀর͍ͩ͘͞ e.g. ࠨ;ఴ࡟લ / ӈ;ఴ࡟ޙ 14/18

  15. ւ֎େֶӃग़ئͷ४උ(CV, ঑ֶۚ) M1ͷͱ͖ʹ5೥ؒࠃ಺֎Ͱֶඅɾੜ׆අʹ༗ޮͳଙਖ਼ٛҭӳࡒஂͷ঑ֶ͕ۚऔΕ͍ͯͨ =>໨҆: HPΛࢀর (https://masason-foundation.org/scholars/) 15/18

  16. ౦޻େͰͷࢦಋڭ׭ɺಌΕΔݚڀऀͷӨڹɾͦͷଞͷӨڹ ཹֶΛ໨ࢦͨ͠ཧ༝ ໘઀४උɾϝʔϧίϯλΫτ ग़ئޙͰ͖Δ͜ͱ ঑ֶۚɾਪનঢ়ɾޠֶࢼݧɾGPAɾCVɾSoPɾίϯλΫτ ւ֎େֶӃग़ئͷ४උ Outline 16/18

  17. ग़ئޙͰ͖Δ͜ͱ(ϝʔϧɾ໘઀४උ) ग़ئޙʹɺग़ئͨ͠ࢫΛ఻͑Δ ॻྨͰͷҰ࣍બߟʹ௨Ε͹ɺڭतͱͷ໘ஊ͕͋ΔͷͰɺཧ࿦෢૷ͱɺݚڀ঺հ΍໘઀͕εϜʔεʹਐΊΒΕΔ ͨΊͷεϥΠυΛࣄલʹϝʔϧͰૹͬͨΓ͠·ͨ͠ ໘઀ͷલʹૹͬͨϝʔϧ 17/18

  18. • ւ֎େֶӃཹֶ͸࣮ࡍʹࢤͯ͠΋ɺग़ئ·ͰʹఘΊͯ͠·͏ਓ͕΄ͱΜͲͳҹ৅ • ͨͩ͠ɺࡢࠓͷ৘ใܥ(ಛʹMLܥ)ͷڝ૪͸ͱͯ΋ݫ͘͠ɺ४උͯ͠΋શམͪͱ͍͏ྫ΋ଟ͘ݟ͖ͯ·ͨ͠ • ࡢ೥ͷܦݧ͔Βɺಉ࢜Λݟ͚ͭ੾᛭ୖຏ͢Δ͜ͱ͕ɺւ֎େֶӃਐֶʹ͸ॏཁͰ͋Δͱײ͡·ͨ͠ • ग़ئʹؔͯ͠αϙʔτ͍͍ͯͨͩͨ͢͠΂ͯͷํʹײँ͍ͨ͠·͢ ࠷ޙʹ ࢀߟαΠτɾஂମ౳

    ·ͱΊ • େࣲߦਓ, “ཹֶʹࢸΔ·ͰͷܦҢ” https://www.funaifoundation.jp/scholarship/201905oshibakojin.pdf • ઙҪ໌ཬ, “೔ຊͷֶ෦͔ΒΞϝϦΧͷίϯϐϡʔλʔαΠΤϯεത࢜՝ఔʹग़ئ͢Δ” https:// akaria.hatenablog.com/entry/2019/08/22/185244 • Philip Guo A Five-Minute Guide to Ph.D. Program Applications https://pg.ucsd.edu/PhD-application-tips.htm • XPLANE http://xplane.seldoon.net/ (ޙԉ: ެӹࡒஂ๏ਓ ଙਖ਼ٛҭӳࡒஂ) • ถࠃେֶӃֶੜձ https://gakuiryugaku.net/ (ޙԉ: ެӹࡒஂ๏ਓ ધҪ৘ใՊֶৼڵࡒஂ) 18/18