Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Features
Speaker Deck
PRO
Sign in
Sign up for free
Search
Search
Today was a Good Day: The Daily Life of Softwar...
Search
Hayato Maki
November 20, 2019
Research
0
62
Today was a Good Day: The Daily Life of Software Developers
Hayato Maki
November 20, 2019
Tweet
Share
More Decks by Hayato Maki
See All by Hayato Maki
Billion-scale Embedding for E-commerce Recommendation in Alibaba
hamaki
0
60
論文紹介:Relaxed Softmax for PU Learning
hamaki
3
970
MIRU 2019 Lunch on Seminar
hamaki
1
240
コーディネート整合性を考慮したカテゴリ間推薦
hamaki
0
1.2k
Regularization_The Element of Statical Learning
hamaki
0
150
Neural Activity During Sentence Processing as Reflected in Theta, Alpha, Beta, and Gamma Oscillations
hamaki
0
160
【ICML読み会】Unsupervised Deep Embedding for Clustering Analysis
hamaki
0
1.3k
Other Decks in Research
See All in Research
文書画像のデータ化における VLM活用 / Use of VLM in document image data conversion
sansan_randd
2
320
第 2 部 11 章「大規模言語モデルの研究開発から実運用に向けて」に向けて / MLOps Book Chapter 11
upura
0
440
Embers of Autoregression: Understanding Large Language Models Through the Problem They are Trained to Solve
eumesy
PRO
7
1.2k
Weekly AI Agents News! 11月号 論文のアーカイブ
masatoto
0
180
[依頼講演] 適応的実験計画法に基づく効率的無線システム設計
k_sato
0
180
機械学習でヒトの行動を変える
hiromu1996
1
390
メールからの名刺情報抽出におけるLLM活用 / Use of LLM in extracting business card information from e-mails
sansan_randd
2
270
Neural Fieldの紹介
nnchiba
1
420
Geospecific View Generation - Geometry-Context Aware High-resolution Ground View Inference from Satellite Views
satai
2
130
The many faces of AI and the role of mathematics
gpeyre
1
1.4k
LLM時代にLabは何をすべきか聞いて回った1年間
hargon24
1
530
ニュースメディアにおける事前学習済みモデルの可能性と課題 / IBIS2024
upura
3
690
Featured
See All Featured
Cheating the UX When There Is Nothing More to Optimize - PixelPioneers
stephaniewalter
280
13k
Building Applications with DynamoDB
mza
91
6.1k
Designing for humans not robots
tammielis
250
25k
Faster Mobile Websites
deanohume
305
30k
Put a Button on it: Removing Barriers to Going Fast.
kastner
59
3.6k
How to Create Impact in a Changing Tech Landscape [PerfNow 2023]
tammyeverts
48
2.2k
Helping Users Find Their Own Way: Creating Modern Search Experiences
danielanewman
29
2.3k
Easily Structure & Communicate Ideas using Wireframe
afnizarnur
191
16k
4 Signs Your Business is Dying
shpigford
182
21k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
eileencodes
132
33k
How to Ace a Technical Interview
jacobian
276
23k
A Philosophy of Restraint
colly
203
16k
Transcript
None
എܠ 3 ઌߦݚڀʹΑΔͱɺ৬ʹର͢Δຬ͕ߴ͍ιϑτΣ Ξ։ൃऀ 4%& ɺੜ࢈తͰྑ͍ίʔυΛॻ͘ɻ 4%&ʹɺ։ൃҎ֎ʹ༷ʑͳࣄ͕͋Δɻ
4%&ʹͱͬͯɺྑ͍ͱͲΜͳͩΖ͏͔ɻ 4%&ʹͱͬͯɺయܕతͳͱͲΜͳͩΖ͏͔ɻ
ߩݙ 4 4%&ʹͱͬͯͷྑ͍ɺయܕతͳΛੳ͢ΔͨΊͷ֓ ೦ϑϨʔϜϫʔΫΛ࡞ͨ͠ɻ .JDSPTPGUͰϑϧλΠϜͰಇ͘4%&ɺਓ͔Βಘͨ ΤϯέʔτௐࠪΛͱʹੳΛߦͬͨɻ Կʹ࣌ؒΛ͍ͬͯΔͱ͍͏ੳΛͱʹɺྑ͍ɺయ
ܕతͳΛఆྔతʹੳͨ͠ɻ ྑ͍Λయܕతʹ͢ΔͨΊͷΞυόΠεΛ༩͑ͨɻ
3FTFBSDI2VFTUJPOT 5 4%&ʹྑ͍ɺయܕతͳͱײͤ͡͞ΔཁҼԿ͔ɻ ·ͨɺͦΕΒͲͷΑ͏ʹӨڹ͍ͯ͠Δ͔ɻ ྑ͍ɺయܕతͳʹɺ4%&ͲͷΑ͏ͳ࣌ؒͷ͍ํ Λ͍ͯ͠Δ͔ बۀʹͲΜͳछྨ͕͋ΓɺྨͰ͖Δ͔
ڠಇʢϛʔςΟϯάͳͲʣྑ͍ɺయܕతͳʹͲͷ Α͏ʹӨڹ͢Δ͔
ઌߦݚڀ 6 4%&͕ίʔσΟϯάʹඅͨ࣌ؒ͠ 1FSSZ `$PEJOH (PODBMWFT `$PEJOH $PMMBCPMBUJPO *OGPSNBUJPO
4FFLJOH "TUSPNTLJT `$PEJOHBUNPTU ։ൃελΠϧίʔσΟϯά࣌ؒͷఆٛɺܭଌ๏ɺௐࠪͷλΠϛϯάʹ ΑΔࠩҟ͕ੜ͍ͯ͡ΔՄೳੑ͋Γ ίʔσΟϯά࣌ؒͷܭଌɺΞϯέʔτɺ؍ɺ࡞ۀϩάͳͲ༷ʑ
ઌߦݚڀ 7 ੜ࢈ੑʹӨڹ͢ΔཁҼɿதஅɺϝʔϧ ৬ʹର͢ΔຬʹӨڹ͢Δཁɿ৺ཧతͳঢ়ଶɺࣄͷָ͠͞ ຊݚڀɺશൠతͳຬͰͳ͘ࡢʹ͍ͭͯ๚Ͷ͍ͯΔ͕ઌߦݚ ڀͱҟͳΔ
ۈʹҹʹӨڹ͢ΔཁҼ w ͦͷ͕ͲΕ͘Β͍యܕత͔ w ઌߦݚڀͷؒͰໃ६ͨ݁͠Ռ͕ಘΒΕ͍ͯΔɻϧʔνϯϫʔΫ͕ଟ͍ ͱྑ͍ѱ͍
༧උ࣮ݧ 8 ຊ࣮ݧͰ༻͍ΔΞϯέʔτΛ࡞͢Δࡍͷࢀߟʹ͢ΔͨΊ ༷ʑͳ৬Ґͷ4%&ʹରͯ͠ແ࡞ҝʹίϯλΫτΛͱΓɺͷΠϯλʔ ωοτΛ࣮ࢪ લʢલճͷۈʣʹ͍ͭͯɺԿΛͲΕ͘Β͍࣌ؒߦ͔ͬͨਘͶͨ &ϝʔϧΧϨϯμʔΛ֬ೝ͠ͳ͕Β͑ͯΒͬͨɻ ৽ͨͳ͕ݟΒΕͳ͘ͳΔ·ͰඃݧऀΛՃɻ̓ਓͰऩଋ͕ͨ͠ɺ̍̌ ਓ·ͰՃͨ͠ɻ
ຊ࣮ݧΞϯέʔτͷߏ 9 ॴଐνʔϜɺ৬Ґɺ։ൃܦݧʹ͍ͭͯਘͶΔ ׆ಈͷϦετΛݟͤɺͦΕͧΕʹରͯ͠લʹͲΕ͘Β͍࣌ؒΛͬͨ ͔ਘͶΔɻϦετʹແ͍׆ಈΛͨ͠߹ࣗ༝هड़ લͷҹʹ͍ͭͯɺʮྑ͍ɾྑ͘ͳ͍ʯɺʮయܕతͳɾయܕత Ͱͳ͍ʯͷͲͪΒ͔ΛͦΕͧΕબΜͰΒ͏ ͦͷଞͷ࣭܈ɻલʹԿճதஅ͞Ε͔ͨɺෆཁͳϛʔςΟϯά͍͘ ͔ͭ͋ͬͨɺɻΞϯέʔτͷճ࣌ؒΛগͳ͘͢ΔͨΊɺͲͷ࣭ ඃݧऀͷˋ͔ΒճΛಘΒΕΔΑ͏ϥϯμϜʹදࣔͨ͠ɻ
ॴཁ࣌ؒͷதԝ ΧϨϯμʔɺϝʔϧɺࢽͳͲΛݟͳ͕Βճ͢ΔΑ͏ʹଅͨ͠
Ξϯέʔτͷૹ 10 ςετࢼߦͱͯ͠ɺ໊ʹΞϯέʔτΛૹ ܰඍͳมߋΛՃ͑ɺਓʹରͯ͠ΞϯέʔτΛૹ िʹਓͣͭɺϲ݄ʹΘͨͬͯૹ ճҎ্ड͚औΔ͜ͱ͋Γ͏Δ ࣮ࡍʹͷඃݧऀ͕ճҎ্ड৴ ɻ
Ξϯέʔτಗ໊ͰճՄೳ͕ͩɺͷඃݧऀ͕ݦ໊Ͱճ ेͳݕఆྗΛಘΔͨΊʹඞཁͳඃݧऀ໊ ݸͷճΛಘͨʢճɿʣɻ ճऀͷ৬ҐδϡχΞɿγχΞʹɿ ։ൃܦݧͷฏۉʢ NJO NBYʣ
ࣗ༝ճͷੳ 11 (SBOUFE5IFPSZʹج͍ͮͯੳ ࣭తௐࠪͰ͔ͭΘΕΔͷ ख࡞ۀͰճΛྨ͍ͯ͘͠Α͏ͳΠϝʔδ
݁Ռɿ࣭తੳ
݁Ռɿྑ͍ɾѱ͍ 13 ؆୯ͷͨΊɺʮྑ͍ʯҎ֎ͷճΛѱ͍ͱදه ݸͷҼࢠɺݸͷߴϨϕϧҼࢠΛಛఆ ճͷׂ߹(PPE#BE
7BMVF$SFBUJPO Ձ 14 ਐḿΛͩ͢ ίʔσΟϯάʹऔΓΉ ҙٛͷ͋ΔࣄΛ͢Δ
ݐઃతͳٞ ϛʔςΟϯάඞͣ͠ѱ͍ҹͰແ͍ ৽͍͜͠ͱΛֶͿɻ ଞਓΛॿ͚Δ ͕࣌ؒଟ͗͢Δͱѱҹʹస͡Δ
&⒏DJFOU6TFPG5JNF ༗ޮͳ࣌ؒͷ͍ํ 15 ظɾܭը௨ΓʹਐΉ ۓٸରԠѱҹΛ༩͑Δ ूதͯ͠ಇ͚Δ ಉ྅ʹΑΔΠϯλϥϓτɺڥϊΠζɺ։์తͳΦ ϑΟεڥѱҹɻ
͗͢ΔϛʔςΟϯάɻҰ͕ϛʔςΟϯάͰΊ ΒΕΔͱѱҹ
1FSDFQUJPO ֮ 16 ੜ࢈ੑʹର͢Δҹɺͦͷͷؾ కΊΓ͔Β͘ΔϓϨογϟʔ ۀ
17 యܕඇయܕత ׆ಈͦͷͷͰͳ͘ɺ࣌ؒͷ͍ํ ʹؔ͢Δظͱݱ࣮ͷΪϟοϓ͕ೝࣝ ʹେ͖͘Өڹ͍ͯ͠Δ ֎ଆͷ࢛͕֯ଆͷ࢛֯ʹӨڹ͢Δ ݁Ռɿయܕతͳ͔
18 ઃܭϑΣʔζ͔ɺίʔσΟϯάϑΣʔ ζ͔ 1SPKFDU1IBTF ϓϩδΣΫτͷఔ
19 /PNFFUJOHEBZɻΑ͘ײँ͞Ε͍ͯΔ ༵ 5ZQFPG8PSLEBZ ۈͷछྨ
20 ϛʔςΟϯάɺΠϯϑϥࣄɺτϨʔχϯά &YUFSOBM ֎෦ཁҼ Ͳ͜Ͱಇ͍͔ͨ -PDBUJPO ͍ͭͱҧ͏λεΫΛ͍ͯ͠Δ .BJO8PSL5BTL 1FSTPOBM ݸਓతͳࣄ
݁Ռɿྔతੳ
ྔతղੳ 22 యܕతͳɺͦ͏Ͱͳ͍ΑΓ౷ܭతʹ༏Ґʹ ྑ͍ʹͳΓ͍͢
ྔతղੳ 23 ҰฏۉͰ࣌ؒࠓಇ͍͍ͯΔʢɺฏۉ͕ٳܜʣ ྑ͍ɺయܕతͳ։ൃؔʹ͕͍ͬͨ࣌ؒ ίʔσΟϯάʹ͏࣌ؒɺ͍͍ͩͨલޙ
ۈͷΫϥελϦϯά 24
ۈͷΫϥελϦϯά 25 5FTUJOH%BZճ্࠷యܕతɻ̎൪ʹྑ͍ɻδϡχΞΫϥεͷ։ൃऀ͕ଟ͍ #VHpYJOH%BZδϡχΞΫϥεͷ։ൃऀ͕ଟ͍ $PEJOH%BZ࠷ྑ͍ɺ͔ͭओ؍తʹ࠷యܕత $PMMBCPSBUJPO%BZϝʔϧɺΠϯλϥϓγϣϯɺखॿ͚͕ଟ͍ɻγχΞ։ൃऀ͕ଟ͍ɻ .FFUJOH%BZϛʔςΟϯάͷɻฏۉͯ࣌ؒ͠Ҏ্
γχΞ։ൃऀ͕ଟ͍ɻ7BSJPVT%BZҎ֎Ͱ࠷ѱ͍ɻ 7BSJPVT%BZͦͷଞͷɻͬͱྫ֎తɻ
ϛʔςΟϯάͱதஅ 26 ෆඞཁɺ͍ϛʔςΟϯάۈͷҹΛѱԽͤ͞Δ ྑ͍ͱײ͡ΒΕͨͰ͢Βɺ༗ҙٛͳϛʔςΟϯάͱײ͡ΒΕׂͨ߹ׂڧ
ʮྑ͍ʯΛ૿͢ 27 ҰൠతʹɺதஅɾϛʔςΟϯάɾࣄ࡞ۀɾΠϯϑϥෆௐΛ࠷খݶʹ͢Δ͖ ։ൃΛओʹ୲͍ͯ͠ΔδϡχΞΫϥεͷ4%&ʹͱͬͯಛʹॏཁ ࠷ۙྲྀߦͷ։์తͳΦϑΟεͰͳ͘ɺਓͷಉ྅ͱγΣΞ͞ΕΔখ͍͞ΦϑΟε͕· ͍͠ɻ ࣗۈΛೝΊΔɻ/PONFFUJOH%BZΛ࡞ΔɻϛʔςΟϯάΛҰʹ·ͱΊΔɻ
ΤΩεύʔτൃݟɺࣝڞ༗ɺޮతͳϝʔϧγεςϜʹΑͬͯதஅΛ࠷খݶʹ͢Δɻ యܕతͳΛྑ͘͢Δ యܕతͰͳ͍Λྑ͘͢Δ
ࣄऀҙࣝूதͱڠۀͷؔ 28 ݸਓͷࣄ74νʔϜͷࣄͷϛʔςΟϯάτϨʔυΦϑ ίʔσΟϯάϑΣʔζʹ͓͚ΔϛʔςΟϯάɾதஅҹΛѱ͘͢Δ ઃܭϑΣʔζʹ͓͚ΔɾதஅҹΛྑ͘͢Δ ৬Ґ্͕͕Δ΄Ͳڠۀͷҹ͕ྑ͘ͳΔ ڠۀͷՁΛߴ͘͢Δ
ධՁγεςϜͷதʹɺଞͷνʔϜɾϓϩδΣΫτʢ044ؚΉʣΛॿ͚Δ͜ͱΛؚΊΔɻ ্࢘Ҏ֎͔ΒධՁ͞ΕΔγεςϜʹ͢Δɻ ϐΞɾϘʔφεΛ׆༻͢Δɻ ࣄऀҙࣝΛײͤ͡͞Δ͜ͱ͕ॏཁɻظɾܭըͱݱ࣮ͷဃΛখ͘͢͞Δɻ ֎෦ཁҼʢϛʔςΟϯάɺதஅʣͷίϯτϩʔϧ͕ॏཁɻ
݁ 29 ֎ཚཁҼΛ࠷খݶʹͯ͠ɺࣄऀҙࣝΛ࣋ͨͤΔ͜ͱ͕ॏཁɻ ܭըɾظ௨ΓʹਐΉͱҹ͕ྑ͘ͳΔ ίʔσΟϯάͷ࣌ؒΛ͘औΓɺதஅϛʔςΟϯά͕গͳ͍ͱ৬ʹ ର͢Δҹ͕ྑ͘ͳΔ ϛʔςΟϯάͷҹɺ։ൃͷϑΣʔζʹΑͬͯมΘΔ
ઃܭϑΣʔζͷϛʔςΟϯάΉ͠ΖҹΛྑ͘͢Δ