Upgrade to Pro
— share decks privately, control downloads, hide ads and more …
Speaker Deck
Sign up for free
Menu
Search
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Features
All features
Private URLs
Password Protection
Custom URLS
Scheduled publishing
Remove Branding
Restrict embedding
Deck Collections
Notes
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Explore
Featured decks
Featured speakers
Programming
Technology
Storyboards
Pricing
Search
Sign in
Sign up for free
IP66_EvacuationLearning
Search
SatokiMasuda
November 18, 2024
Research
77
0
Share
Embed
Copy iframe code
Copy JS code
Copy link
Start on current slide
IP66_EvacuationLearning
第66回土木計画学研究発表会・秋大会の発表資料です。「異質性に着目した強化学習に基づく動的避難目的地選択モデル」
SatokiMasuda
November 18, 2024
More Decks by SatokiMasuda
See All by SatokiMasuda
TR-C2026
stkmsd
0
10
IP73_LUTI_dynamic_game
stkmsd
0
72
hksts2025
stkmsd
1
85
ieee2025
stkmsd
0
81
ip71_contraflow_reconfiguration
stkmsd
0
180
kanazawa2024
stkmsd
0
78
hksts2024
stkmsd
0
79
IP70_counterfactual_machine_learning
stkmsd
0
96
ip68_LocationGame
stkmsd
0
64
Other Decks in Research
See All in Research
20260624 NLP colloquium: 単一のhubテキストがCLIPを壊す:hubnessによる埋め込みの脆弱性特定
de9uch1
2
260
OWASP AISVS - C7
shiell
2
780
Cross-Media Human-Information Interaction
signer
PRO
0
240
GLIM とMegaParticles:正規分布近似の限界とタイトカップリング&パーティクルフィルタの進展 / GLIM and MegaParticles : Progress of the distribution representation in SLAM
koide3
0
860
秋葉原ウォーカブル基礎調査報告書
izumiyama_lab
1
110
SLAMはどこまで解決されたのか?
tomonom
0
1.4k
超効率化への挑戦:1bit LLMの現状と展望
yumaichikawa
0
720
Karkada さんの論文 × 2 の紹介: (1) Closed-Form Training Dynamics Reveal Learned Features and Linear Structure in Word2Vec-like Models, (2) Symmetry in language statistics shapes the geometry of model representations
eumesy
PRO
1
740
EIRによる不正端末のブロッキング 5G時代におけるデバイス識別と不正対策の進化
stellarcraft
0
130
論文紹介:Doc-to-LoRA: Learning to Instantly Internalize Contexts
yukako_nakano
0
170
Evaluating LLM Reliability Across Facts, Evidence, and Cultures
yukiar
0
170
実例から見るLLMのマンガ理解:実務VQAタスクによる長期的文脈と視覚情報の定性評価
kzmssk
0
140
Featured
See All Featured
Bioeconomy Workshop: Dr. Julius Ecuru, Opportunities for a Bioeconomy in West Africa
akademiya2063
PRO
1
360
Build The Right Thing And Hit Your Dates
maggiecrowley
39
3.4k
Darren the Foodie - Storyboard
khoart
PRO
4
3.9k
16th Malabo Montpellier Forum Presentation
akademiya2063
PRO
0
380
How GitHub (no longer) Works
holman
316
150k
Future Trends and Review - Lecture 12 - Web Technologies (1019888BNR)
signer
PRO
0
3.7k
Agile Leadership in an Agile Organization
kimpetersen
PRO
0
230
Designing for Timeless Needs
cassininazir
1
480
HTML-Aware ERB: The Path to Reactive Rendering @ RubyCon 2026, Rimini, Italy
marcoroth
5
670
The Organizational Zoo: Understanding Human Behavior Agility Through Metaphoric Constructive Conversations (based on the works of Arthur Shelley, Ph.D)
kimpetersen
PRO
0
450
Agile Actions for Facilitating Distributed Teams - ADO2019
mkilby
0
280
The Cult of Friendly URLs
andyhume
79
7k
Transcript
ҟ࣭ੑʹணͨ͠ڧԽֶशʹجͮ͘ ಈతආతબϞσϧ ˓ ૿ా ܛथ ౦ژେֶֶܥݚڀՊ ະདྷ גࣜձࣾ๛ాதԝݚڀॴ Ӌ౻
ӳೋ ౦ژେֶֶܥݚڀՊ 12 ަ௨ωοτϫʔΫੳηογϣϯ 2022.11.12 16:45-18:15 ୈ9ձ ୈճܭըֶݚڀൃදձɾळେձ!ླྀٿେֶ
ආͷωοτϫʔΫσβΠϯɾ੍ޚ 2 ࡂ࣌ʹಥൃతͳधཁ͕ൃੜ͠ɺωοτϫʔΫ༰ྔΛա ةݥେ ةݥத ةݥখ
ආͷωοτϫʔΫσβΠϯɾ੍ޚ Øࡂ࣌ͷใఏڙσβΠϯ Øࡂؒͷආ܇࿅ͷ࠷దઃܭ 3 બͷҟ࣭ੑΛ׆༻ͨ͠ආަ௨ͷΈ߹Θͤ࠷ద ةݥେ ةݥத ةݥখ ใͷֶशաఔͱߦಈม༰ΛϞσϧԽ͠ɺ੍ޚʹ׆༻͢Δ
ֶशաఔͷදݱ – ڧԽֶश ΤʔδΣϯτ͕ڧԽֶशΛߦ͏ͱ͢Δͱɺ ߦಈͱใुͷ֫ಘΛ܁Γฦ͠ɺ࠷దͳํࡦΛֶश͍ͯ͘͠ → ʮඇආʯঢ়ଶʹ͋Δ࣌ɺظใु͕࠷ߴ͍ঢ়ଶભҠΛֶश 4 ࣌ؒ ֶशᶃ
ආޮ༻ 𝑣 ආ ᶄ Ұํɺਓؒڥ͔Β࠷దํࡦΛֶͳ͍͜ͱɺֶशΛ٫ ͢Δ͜ͱ͕͋Δɻ
ֶशաఔͷදݱ – day-to-dayͷܦ࿏બ ܦ࿏બߦಈʹؔ͢Δ࣮ݧ࣮ࣨݧͷݚڀ͕ߦΘΕ͖ͯͨ • ܁Γฦ͠ʹΑΔश׳Խ Bogers, Bierlaire, Hoogendoorn (2007)
• ϕΠζϧʔϧʹΑΔೝͷߋ৽ Jha, Madanat, Peeta (1998) • Horowitz (1984) – ֶशʹΑΔཱྀ֮ߦ࣌ؒͷܗΛදݱ 5 𝑢!" = 𝛽#!$% $ & "'( 𝑤& 𝑇!& + 𝜖!" աڈͷཱྀߦ࣌ؒͷॏΈ͖ฏۉ 𝑤& ͷઃఆʹΑΓ͞·͟·ͳදݱ͕Մೳɻ ex) ͍ۙաڈͷܦݧ΄ͲॏΈ͚ ü ॏΈ 𝑤& ੳऀ͕ઃఆ͢Δ ü ࡂ܇࿅ຖ܁Γฦ͞ΕΔֶशͰͳ͍ ຊݚڀͰ𝑤! Λ٫ͱଊ͑ɺೝֶशաఔΛද͢ॏཁͳ ύϥϝʔλͱͯ͠ਪఆ͢Δ
ຊݚڀͷয త ใͷֶशͱ٫Λߟྀͨ͠ආωοτϫʔΫ੍ޚ nڧԽֶशɾday-to-dayͷܦ࿏બͷֶश • աڈͷܦݧ֎෦ใʹΑΔ֮ߦಈنൣܗ Øֶशִ͕͍ؒ߹ͷɺֶशͱ٫ͷهड़ͱ༧ଌ͕ඞཁ n ආߦಈͷੳ •
܇࿅ใఏڙલޙͷҙมԽͷੳ ØޮՌͷ࣋ଓͷੳ͕ෆՄܽ ࡂكগࣄΏֶ͑शͱ٫ΛϞσϦϯά͠ɺ࣮ݧσʔλʹΑΔ ύϥϝʔλਪఆͰֶशաఔͷಛΛ໌Β͔ʹ͢Δɻ 6
ֶशաఔͷදݱ 7 ࣌ؒ ֶशᶃ 𝑣) ࣌ؒͱͱʹޮՌݮ ٫ ආޮ༻ 𝑣
ආ
ֶशաఔͷදݱ 8 ࣌ؒ ੳ࣌ 𝑣) ࣌ؒͱͱʹޮՌݮ ٫ ආޮ༻ 𝑣 ආ
ֶशᶃ
ֶशաఔͷදݱ 9 ආޮ༻ 𝑣 ආ ࣌ؒ ੳ࣌ 𝑣) 𝜆 𝑣避難
= 𝑣" +𝜆𝛿学習 ֶशޮՌൈ͖ͷ ޮ༻ͷ֬ఆ߲ ֶशܦݧ͕͋Ε ͳ͚Εͷม ֶशᶃ
ֶशաఔͷදݱ 10 ࣌ؒ ੳ࣌ 𝑣) 𝜆 ਅͷޮՌ 𝑣避難 = 𝑣"
+𝜆𝛿学習 ֶशޮՌൈ͖ͷ ޮ༻ͷ֬ఆ߲ ֶशܦݧ͕͋Ε ͳ͚Εͷม ආޮ༻ 𝑣 ආ ܦݧΛ୯७ʹઆ໌มʹؚΊΔ͚ͩͰ ֶशͷޮՌΛաখධՁͯ͠͠·͏ ֶशᶃ
ֶशաఔͷදݱ – ఏҊ 11 ࣌ؒ ֶशᶃ ̅ 𝜆 ආޮ༻ 𝑣
ආ ੳ࣌ ୯Ґ࣌ؒ 𝑣)
ֶशաఔͷදݱ – ఏҊ 12 ࣌ؒ 𝑣) ̅ 𝜆 ආޮ༻ 𝑣
ආ ੳ࣌ 𝛾 ̅ 𝜆 𝛾* ̅ 𝜆 𝛾+ ̅ 𝜆 ੳ࣌Ͱͷ ֶशᶃͷޮՌ 𝑣避難 = 𝑣" +𝛾# ̅ 𝜆𝛿学習① ֶशᶃ͕͋Ε ͳ͚Εͷม ٫ͷఔΛද͢ม𝛾Λಋೖ ֶशᶃ ୯Ґ࣌ؒ
ֶशաఔͷදݱ ʻԾఆʼ • ԿֶशΛ܁Γฦͯ͠ɺͦͷޮՌ ̅ 𝜆Ͱݻఆ • ֶशͷִؒ΄΅ҰఆͰɺwaveؒͰ٫ 𝛾ͰޮՌ͕ݮ͢Δ 13
𝑣避難 = 𝑣" +𝛾!$% ̅ 𝜆𝛿&'() % + 𝛾!$* ̅ 𝜆𝛿&'() * + 𝛾!$# ̅ 𝜆𝛿&'() # + ⋯ = 𝑣" + ̅ 𝜆 * +,% ! 𝛾!$+𝛿+ • ਪఆରɺ ̅ 𝜆ʢֶशʹΑΔޮՌʣͱ 𝛾ʢ٫ʣ શ 𝑥 wave͋Δͱ͖ɺwave 𝑥ޙͷޮ༻ͷ֬ఆ߲ɺ
ֶशσʔλͷऔಘ • ෳwaveͷߦಈσʔλ͕͋Ε٫܇࿅ޮՌͷਪఆ͕Մೳ 14 ࠞࡶใ ආ܇࿅ ආ܇࿅ ආ܇࿅ ࠞࡶใ ਁਫใ
ආ܇࿅ ࠞࡶใ ਁਫใ ආߦಈ SPௐࠪᶃ 2022/3/2 ~ 4 ආߦಈ SPௐࠪᶄ 2022/3/11 ~ 15 ආߦಈ SPௐࠪᶅ 2022/3/25 ~ 29 ආߦಈ SPௐࠪᶆ 2022/4/14 ~ 20 ࠞࡶใ ਁਫใ 272໊
܇࿅ͱใఏڙͷ༷ࢠ 15 1 2 3 ᶃେౡஸஂ ϋβʔυใ ྟւ෦ͷ΄͏͕ਫ ʹରͯ҆͠શͱ͍͏ ͷײͱҧ͏ɻ
ॳΊͯͬͨɻ ਁਫҬʹॅΉߴྸঁੑ
ආతબϞσϧ – 2ͭͷಈֶੑ 16 ࣗ ආॴ A ආॴ B wave1
wave2 ࡂ࣌ ಈతࢄબϞσϧ ආܦݧ ࡂؒ ޮ༻ͷߋ৽ 𝑝 𝑠!"# 𝑠! = 𝑒 # $ % 𝑠!"# 𝑠! ; 𝜽 "&'! ("#$ ∑ ("#$ % ∈* (" 𝑒 # $ % 𝑠!"# + 𝑠! ; 𝜽 "&'! ("#$ % 𝑣!"# = 𝑣$ + 𝜆 % %,&' % 𝛾%(%, 𝛿% wave間 wave
waveؒ ֶशύϥϝʔλͷਪఆ 17 wave1避難者 wave1非避難者 推定値 t値 推定値 t値 非避難効用の変化
(出発時刻選択) 48h前固有項 0.362 2.08* 0.224 0.73 24h前固有項 0.344 1.85 0.862 2.25* 12h前固有項 -0.303 -1.28 -1.541 -1.89 6h前固有項 -1.236 -3.39** 2.173 2.08* 目的地効用の変化 ハザードマップ内 -0.683 -1.40 -1.128 -1.59 避難訓練参加 (自宅選択時) -0.580 -1.05 0.004 0.01 目的地の混雑情報 (非自宅選択時) -0.433 -1.38 -2.017 -2.86** 記憶率 0.143 0.48 0.321 1.20 サンプル数 100 144 初期対数尤度 -723.9 -290.0 最終対数尤度 -682.2 -197.8 尤度比 0.058 0.318 修正済尤度比 0.047 0.290 *:5%有意, **1%有意 • wave1Ͱආ͢Δͱճͨ͠ਓͱͦ͏Ͱͳ͍ਓʹ͚ͯਪఆ ආͷબ͕ݩ͔ Β͍ਓɺ܇࿅ ࢀՃͷޮՌ͕ೝΊ ΒΕͳ͍͕ɺࠞࡶ Λආ͚Α͏ͱ͢Δ ̅ 𝜆 𝛾 忘却率
waveؒ ֶशύϥϝʔλͷਪఆ 18 EMクラス1 EMクラス2 推定値 t値 推定値 t値 非避難効用の変化
(出発時刻選択) 48h前固有項 11.871 0.10 -0.053 -0.30 24h前固有項 0.369 2.14* 0.237 1.11 12h前固有項 -8.180 -0.15 0.143 0.63 6h前固有項 8.740 0.15 -2.449 -5.46** 目的地効用の変化 ハザードマップ内 -2.144 -2.28* 0.847 1.41 避難訓練参加 (自宅選択時) -1.902 -2.20* -0.629 -0.74 目的地の混雑情報 (非自宅選択時) -3.598 -3.09** 0.028 0.08 記憶率 1 4.15** 0.213 1.17 サンプル数 100 初期対数尤度 -723.9 最終対数尤度 -571.7 尤度比 0.210 修正済尤度比 0.188 *:5%有意, **1%有意 Ϋϥε1 = • ਁਫ͢Δॴʹߦ͖ͨ ͘ͳ͍ • ܇࿅ࢀՃʹΑΓආ ޮ༻্͕ঢ͢Δ • ࠞࡶ͢Δॴʹߦ͖ͨ ͘ͳ͍ ͱֶश͢Δൣ Ϋϥε2 = ใఏڙʹײ͕ͳ͍ • wave1Ͱආ͢Δͱճͨ͠ਓͷதʹҟ࣭ੑ͕͋Δͱߟ͑ɺજ ࡏΫϥεϞσϧͰਪఆ ̅ 𝜆 𝛾 忘却率
ආ܇࿅ࢀՃʹΑΔආޮ༻ͷมԽ 19 ࣌ؒ ආ܇࿅ᶃ ආ܇࿅ᶄ 𝑣) ආޮ༻ 𝑣 ආ 2िؒ
−3.60 -0.63 Ϋϥε1 Ϋϥε2
ࠞࡶใఏڙʹΑΔతޮ༻ͷมԽ 20 ࣌ؒ ใఏڙᶃ ใఏڙᶄ ආޮ༻ 𝑣 ආ 2िؒ −1.90
0.03 Ϋϥε1 Ϋϥε2
·ͱΊ üكগࣄʹର͢Δֶशͱ٫ͷաఔΛϞσϧԽ ü࣮ݧσʔλʹΑΔύϥϝʔλਪఆͰ܇࿅ͱใͷֶशաఔΛੳ üใఏڙܦݧʹର͢Δֶशͷఔͱɺ٫ͷ͞ʹҟ࣭ੑ͕ ͋Δ͜ͱΛ໌Β͔ʹͨ͠ ࠓޙͷํੑ Øใఏڙʹର͢ΔԠͷҟ࣭ੑΛར༻ͯ͠ɺආަ௨ͷधཁɾܦ ࿏ɾతͷ࠷ద੍ޚൃల Ø٫ͷԾఆͷ؇ →
ܦա࣌ؒͷߏԽɺม͝ͱʹҟͳΔ٫ͷઃఆ 21