Upgrade to Pro — share decks privately, control downloads, hide ads and more …

How Many Vertices Does a Random Walk Miss in a ...

Avatar for Nobutaka Shimizu Nobutaka Shimizu
July 05, 2022
Research
190
0

How Many Vertices Does a Random Walk Miss in a Network with Moderately Increasing the Number of Vertices?

slide at SODA2021

Avatar for Nobutaka Shimizu

Nobutaka Shimizu

July 05, 2022

More Decks by Nobutaka Shimizu

See All by Nobutaka Shimizu
Planted Clique Conjectures are Equivalent
Avatar for Nobutaka Shimizu nobushimi
0
310
理論計算機科学における 数学の応用: 擬似ランダムネス
Avatar for Nobutaka Shimizu nobushimi
1
500
Hardness Self-Amplification: Simplified, Optimized, and Unified
Avatar for Nobutaka Shimizu nobushimi
0
360
Hardness Self-Amplification from Feasible Hard-Core Sets
Avatar for Nobutaka Shimizu nobushimi
0
260
Nearly Optimal Average-Case Complexity of Counting Bicliques Under SETH
Avatar for Nobutaka Shimizu nobushimi
0
210
Quasi-majority Functional Voting on Expander Graphs
Avatar for Nobutaka Shimizu nobushimi
0
110
Phase Transitions of Best-of-Two and Best-of-Three on Stochastic Block Models
Avatar for Nobutaka Shimizu nobushimi
0
180

Other Decks in Research

See All in Research
AIを叩き台として、 「検証」から「共創」へと進化するリサーチ
Avatar for めーら/Miki mela_dayo
0
280
明日から使える！研究効率化ツール入門
Avatar for Yusuke Matsui matsui_528
13
7.3k
羽田新ルート運用6年の検証
Avatar for マン点 1manken
0
160
Ankylosing Spondylitis
Avatar for Charles Holtzkampf ankh2054
0
170
LiDAR点群の地表面分類手法の比較・検証
Avatar for vegapunk-hiroshi vegapunkhiroshi79
0
120
多様なデータを許容し学習し続ける模倣学習 / Advanced Imitation Learning for VLA
Avatar for PRIN Lab prinlab
0
220
適応的スパムフィルタのための軽量な類似メッセージカウンタ / jsai2026-adaptive-spam-filter
Avatar for monochromegane monochromegane
0
3.5k
データセンター事業者を取り巻く近年の状況とその中での研究開発動向、テストベッドへの貢献の可能性
Avatar for KIKUCHI Shunsuke kikuzo
1
180
老舗ものづくり企業でリサーチが変革を起こすまで - 三菱重工DXの実践
Avatar for kazu skydats
0
190
2026 東京科学大 情報通信系 研究室紹介 (大岡山)
Avatar for Tokyo Tech ICT Dept. icttitech
0
3.7k
NII S. Koyama's Lab Research Overview AY2026
Avatar for NII S. Koyama's Lab skoyamalab
0
290
COFFEE-Japan PROJECT Impact Report（海ノ向こうコーヒー）
Avatar for 坂ノ途中 ontheslope
0
1.9k

Featured

See All Featured
ReactJS: Keep Simple. Everything can be a component!
Avatar for Pedro Nauck pedronauck
666
130k
Building a A Zero-Code AI SEO Workflow
Avatar for Ian Lurie portentint
PRO
0
570
The Director’s Chair: Orchestrating AI for Truly Effective Learning
Avatar for Mike Taylor tmiket
1
190
Leading Effective Engineering Teams in the AI Era
Avatar for Addy Osmani addyosmani
9
2k
Unsuck your backbone
Avatar for Amy Palamountain ammeep
672
58k
Marketing to machines
Avatar for Jono Alderson jonoalderson
1
5.4k
Kristin Tynski - Automating Marketing Tasks With AI
Avatar for Tech SEO Connect techseoconnect
PRO
0
270
10 Git Anti Patterns You Should be Aware of
Avatar for Lemi Orhan Ergin lemiorhan
PRO
659
62k
RailsConf & Balkan Ruby 2019: The Past, Present, and Future of Rails at GitHub
Avatar for Eileen M. Uchitelle eileencodes
141
35k
Are puppies a ranking factor?
Avatar for Jono Alderson jonoalderson
1
3.5k
Faster Mobile Websites
Avatar for Dean Hume deanohume
310
31k
Reality Check: Gamification 10 Years Later
Avatar for Sebastian Deterding codingconduct
0
2.2k

Transcript

  1. )PXNBOZWFSUJDFTEPFTBSBOEPN XBMLNJTTJOBOFUXPSLXJUINPEFSBUFMZ JODSFBTJOHUIFOVNCFSPGWFSUJDFT 4IVKJ,JKJNB/PCVUBLB4IJNJ[V5BLFIBSV4IJSBHB ,ZVTIV6OJWFSTJUZ 5IF6OJWFSTJUZPG5PLZP $IVP6OJWFSTJUZ

  2. /28 2 3BOEPN8BML w #BTJD.BSLPWDIBJOPOBHSBQI w TJNQMFBOEMPXNFNPSZˠBQQMJDBUJPOJOOFUXPSLBOBMZTJT FH 1BHF3BOL .$.$

    DMVTUFSJOH FUD

  3. /28 3 )JUUJOH5JNFBOE$PWFS5JNF w )PXGBTUEPFTB38TQSFBET  w IJUUJOHUJNF  

    w DPWFSUJNF  w .BOZQSFWJPVTXPSLTTUVEZJOH38POTUBUJDHSBQIT thit max u,v E[38IJUTvTUBSUJOHGSPNV] tcov = max u E[38WJTJUTBMMWFSUJDFTTUBSUJOHGSPNu] [Aleliunas, Karp, Lipton, Lovász, Rackoff, 1979] [Feige, 1995] [Feige, 1995] [Matthews, 1988]

  4. /28 4 38PO%ZOBNJD(SBQIT 3FBMXPSMEOFUXPSLTDIBOHFUIFJSTUSVDUVSFPWFSUJNF FH 888 TPDJBMOFUXPSL DIFNJDBMOFUXPSL ʜ [Yu,

    McCann, 2016] 8FDBOBOBMZ[FEZOBNJDOFUXPSLTVTJOH38

  5. /28 5 38PO%ZOBNJD(SBQIT ɾ TFRVFODF PGHSBQITTU  ɾ*GBMM BSF SFHVMBS

      ɾ38POSFDVSSJOHGBNJMZ TQFDJ fi DUZQFPG  ɾ"MM BSFSBOEPNHSBQI ∃ (Gt ) tcov = 2Ω(n) Gt d thit = O(n2) Gt Gt [Avin, Kousky, Lotler, 2008] [Sauerwald, Zanetti, 2019] [Oksana and Luís, 2014] [Cai, Sauerwald, Zanetti, 2020] Most previous works consider the case of a static vertex set (i.e., for all ) V(Gt ) = V t Consider RW on a sequence of graphs. G0 , G1 , …

  6. /28 6 5IJT8PSL ɾ8FDPOTJEFS38POBHSPXJOHHSBQI G1 G2 G3 G4 Gn 6

  7. /28 7 0VS'SBNFXPSL ɾ8FGPDVTPONPEFSBUFMZHSPXJOHHSBQIT ɾLet be a function (duration ɾLet

    be graphs s.t. ɾ is obtained by adding a vertex and edges to . 𝔡 : ℕ → ℕ (G(i))i∈ℕ V(G(i)) = {v1 , …, vi } G(i+1) G(i) 7 ɾ PGVOWJTJUFEWFSUJDFTBGUFS38PO 5IFO  U = Un = Gn E[U] = ? SPVOE ɾ'PSFBDI QFSGPSN38PGMFOHUI POFBDI  i 𝔡 (i) G(i) *OUIJTTFUUJOH XFDBOBQQMZBOBMZTJTUFDIOJRVFTPG38

  8. /28 8 4JNVMBUJPO7JEFP 8 (SPXJOHDPNQMFUFHSBQI (SPXJOHQBUI

  9. /28 3FMBUFE8PSL w .PTUQSFWJPVTXPSLTTUVEJFEB38POBTUBUJDWFSUFYTFU 9 [Avin, Kousky, Lotler, 2008] [Sauerwald,

    Zanetti, 2019] [Oksana and Luís, 2014] [Cai, Sauerwald, Zanetti, 2020] ɾ5IFPOMZFYDFQUJPOJT[Cooper, Frieze, 2002] XIPDPOTJEFSFE38POHSPXJOH QSFGFSFOUJBMBUUBDINFOUNPEFMXJUIDPOTUBOU 5IFZQSPWFEUIBU  BT 𝔡 (i) E[U]/n → const n → ∞

  10. /28 10 0CTFSWBUJPOT ɾ*G UIFO UIF38WJTJUTBMMWFSUJDFTBUFBDISPVOE 𝔡 (i) ≫ tcov

    (G(i)) 10 ɾ4VQQPTF GPSBMM 5IFO 𝔡 (i) ≥ (1 + ϵ)thit (G(i)) i E[U] = n ∑ i=1 Pr[vi JTVOWJTJUFE] = O(1) . Pr[vi JTVOWJTJUFE] = Pr ⋀ j≥i {vi JTVOWJTJUFEBUjUISPVOE} UIBSSJWBMWFSUFY JF  i V(Gi )∖V(Gi−1 ) ≤ (1 + ϵ)−(n−i+1) 5IFSFGPSF ɾ)PXBCPVU  𝔡 (i) ≪ thit (G(i))

  11. /28 11 3FTVMUT 11 Theorem 1 Consider a lazy and

    reversible RW on a moderately growing graph. If , then . 𝔡 (i) ≥ 3Cthit (G(i)) N + 2tmix (G(i)) E[U] ≤ 8N + 32 ɾ3FTVMUGPSHSBQITXJUI  FH FYQBOEFSHSBQIT thit (G(i)) ≫ tmix (G(i)) ɾ8FDBOOPUBQQMZ5IFPSFNJG  FH QBUI thit (G(i)) ≈ tmix (G(i)) ɾ"UFBDISPVOE 38NJYFTFOPVHITJODF 𝔡 (i) ≥ 2tmix (G(i))

  12. /28 12 3FTVMUT 12 Theorem 2 (informal) Consider a lazy

    simple RW on a growing graph. Suppose that and Then, . |E(G(i))| |E(G(i−1))| ≤ 1 + O(i−1) 𝔡 (i) ≥ Ω ( thit (Gi ) iγ ) E[U] = O(nγ) ɾ8FDBOBQQMZ5IFPSFNJGUIFFEHFTNPEFSBUFMZJODSFBTF FH QBUI

  13. /28 13 &YBNQMF 13 5IFSFJTBDPOTU TVDIUIBU GPSBOZ C > 0

    γ ∈ [0,1] 0OBHSPXJOHDPNQMFUFHSBQI  ɾ  ɾ 𝔡 (i) ≥ Ci1−γ ⇒ E[U] = O(nγ) 𝔡 (i) ≤ Ci1−γ ⇒ E[U] = Ω(nγ) 0OBHSPXJOHQBUI ɾ  ɾ 𝔡 (i) ≥ Ci2−γ ⇒ E[U] = O(nγ) 𝔡 (i) ≤ Ci2−γ ⇒ E[U] = Ω(nγ) ˎ-PXFSCPVOETBSFTIPXOCZBOBEIPDXBZ

  14. /28 14 *EFBPG1SPPG 14  "TBXBSNVQ DPOTJEFSBHSPXJOHDPNQMFUFHSBQI  8FFYUFOEUIFBSHVNFOUUPUIFHFOFSBMDBTF Theorem

    2 (reminder) Consider a lazy simple RW on a growing graph. Suppose that and Then, . |E(G(i))| |E(G(i−1))| ≤ 1 + O(i−1) 𝔡 (i) ≥ Ω ( thit (Gi ) iγ ) E[U] = O(nγ)

  15. /28 15 8BSN6Q$PNQMFUF(SBQI 15 ɾ4VQQPTF IBTBTFMGMPPQPOFBDIWFSUFY G(i) = Ki 38WJTJU

    XQ v ∈ V(G(i)) 1/i ɾPr[vJTVOWJTJUFEEVSJOHSPVOEi] = (1 − 1 i ) (i) Pr[vi JTVOWJTJUFEBMMUIFUJNF] = n ∏ j=i (1 − 1 j ) (j) E[U] = n ∑ i=1 n ∏ j=i (1 − 1 j ) 𝔡 (j)

  16. /28 16 8BSN6Q$PNQMFUF(SBQI 16 ɾ6QQFSCPVOE 𝔡 (i) ≥ Ci1−γ ⇒

    E[U] = O(nγ) &WBMVBUFE[U] = n ∑ i=1 n ∏ j=i (1 − 1 j ) 𝔡 (j) n ∏ j=i (1 − 1 j ) 𝔡 (i) ≤ exp − n ∑ j=i C jγ ≤ exp (−(n − i + 1) C nγ ) . E[U] ≤ n ∑ i=1 exp (−(n − i + 1) C nγ ) = O(nγ) . )FODF □

  17. /28 17 8BSN6Q$PNQMFUF(SBQI 17 ɾ-PXFSCPVOE 𝔡 (i) ≤ Ci1−γ ⇒

    E[U] = Ω(nγ) &WBMVBUFE[U] = n ∑ i=1 n ∏ j=i (1 − 1 j ) 𝔡 (j) n ∏ j=i (1 − 1 j ) 𝔡 (i) ≥ n ∏ j=i (1 − 1 j ) Cn1−γ = ( i − 1 n ) Cn1−γ . E[U] ≥ n ∑ i=1 ( i − 1 n ) Cn1−γ = Ω(nγ) )FODF □

  18. /28 18 &YUFOEUP(FOFSBM(SBQI 18 E[U] = n ∑ i=1 n

    ∏ j=i (1 − 1 j ) 𝔡 (j) 0O GPSBOZ BOE  Ki u, v ∈ V(Ki ) t Pr[τu,v > t] = (1 − 1 i ) t ˎ UJNFGPS38UPWJTJU TUBSUJOHGSPN τu,v v u

  19. /28 19 &YUFOEUP(FOFSBM(SBQI 19 E[U] = n ∑ i=1 n

    ∏ j=i (1 − 1 j ) 𝔡 (j) 0O GPSBOZ BOE  Ki u, v ∈ V(Ki ) t Pr[τu,v > t] = (1 − 1 i ) t ˎ UJNFGPS38UPWJTJU TUBSUJOHGSPN τu,v v u ˎ-B[Z JSSFEVDJCMF BOEMB[Z38 ˎ JTUIFTUBUJPOBSZEJTU π ∈ [0,1]V [Aldous, Fill, 2002], [Oliveira, Peres, 2019] HFOFSBMJ[F 'PSBOZ BOE  v ∈ V(G) t Pr u∼π [τu,v > t] ≤ ( 1 − 1 thit ) t

  20. /28 20 &YUFOEUP(FOFSBM(SBQI 20 E[U] = n ∑ i=1 n

    ∏ j=i (1 − 1 j ) 𝔡 (j) 0O GPSBOZ BOE  Ki u, v ∈ V(Ki ) t Pr[τu,v > t] = (1 − 1 i ) t ˎ UJNFGPS38UPWJTJU TUBSUJOHGSPN τu,v v u ˎ-B[Z JSSFEVDJCMF BOEMB[Z38 ˎ JTUIFTUBUJPOBSZEJTU π ∈ [0,1]V [Aldous, Fill, 2002], [Oliveira, Peres, 2019] E[U] ≤ n ∑ i=1 n ∏ j=i ( 1 − 1 thit (G(j))) 𝔡 (j) 🤔 HFOFSBMJ[F 'PSBOZ BOE  v ∈ V(G) t Pr u∼π [τu,v > t] ≤ ( 1 − 1 thit ) t

  21. /28 E[U] ≤ n ∑ i=1 n ∏ j=i (

    1 − 1 thit (G(j))) 𝔡 (j) 21 &YUFOEUP(FOFSBM(SBQI 21 E[U] = n ∑ i=1 n ∏ j=i (1 − 1 j ) 𝔡 (j) 0O GPSBOZ BOE  Ki u, v ∈ V(Ki ) t Pr[τu,v > t] = (1 − 1 i ) t ˎ UJNFGPS38UPWJTJU TUBSUJOHGSPN τu,v v u ˎ-B[Z JSSFEVDJCMF BOEMB[Z38 ˎ JTUIFTUBUJPOBSZEJTU π ∈ [0,1]V [Aldous, Fill, 2002], [Oliveira, Peres, 2019] 😓 HFOFSBMJ[F 'PSBOZ BOE  v ∈ V(G) t Pr u∼π [τu,v > t] ≤ ( 1 − 1 thit ) t 0VSHSBQIJTEZOBNJD DIBOHFTPWFSUJNF π

  22. /28 22 &YUFOEUP(FOFSBM(SBQI 22 E[U] ≤ n ∑ i=1 n

    ∏ j=i ( 1 − 1 thit (G(j))) 𝔡 (j) Lemma E[U] ≤ O(1) ⋅ n ∑ i=1 n ∏ j=i max v∈V(G(j)) π(j−1)(v) π(j)(v) ( 1 − 1 thit (G(j))) (j) JTTUBUJPOBSZEJTUPG π(j) G(j) *G DIBOHFTNPEFSBUFMZ UIFOXFDBOFWBMVBUF  π(i) E[U] 😓 😄

  23. /28 23 1SPPG4LFUDI 23  Pr[vi JTVOWJTJUFE] ≈ ∏ j≥i

    SBUJPCFUXFFO BOE   Xt π(j) × Pr u∼π [τu,v > t] *G  UIJTUFSN  𝔡 (j) ≥ tmix (G(j)) ≈ 1 [Aldous, Fill, 2002], [Oliveira, Peres, 2019] 'PSBOZ BOE  v ∈ V(G) t Pr u∼π [τu,v > t] ≤ ( 1 − 1 thit ) t ˎ DBOCFNVDIMFTTUIBOUIFNJYJOHUJNF😓 𝔡 (j) ˎ JTTUBUJPOBSZEJTUPG π(j) G(j)

  24. /28 24 #FIBWJPSPG38 24 0OBTUBUJDHSBQI  EJTUSJCVUJPOPG38 → π π

    [0,1]V EJTUPGXt t → ∞ 0OBEZOBNJDHSBQI  DIBOHFTPWFSUJNF π π(i) EJTUPGXt t → ∞ π(i+1)

  25. /28 25 #FIBWJPSPG38 25 0OBTUBUJDHSBQI  EJTUSJCVUJPOPG38 → π π

    [0,1]V EJTUPGXt t → ∞ 0OBEZOBNJDHSBQI  DIBOHFTPWFSUJNF π π(i) EJTUPGXt t → ∞ π(i+1) *G DIBOHFTNPEFSBUFMZ  XFDBOCPVOEUIFEJTUBODF π

  26. /28 26 &YUFOEUP(FOFSBM(SBQI 26 Lemma E[U] ≤ O(1) ⋅ n

    ∑ i=1 n ∏ j=i max v∈V(G(j)) π(j−1)(v) π(j)(v) ( 1 − 1 thit (G(j))) (j) JTTUBUJPOBSZEJTUPG π(j) G(j) *G DIBOHFTNPEFSBUFMZ UIFOXFDBOFWBMVBUF  π(i) E[U]

  27. /28 27 1SPPGWJB-FNNB 27 'SPNBTTVNQUJPO Theorem 2 (reminder) Consider a

    lazy simple RW on a growing graph. Suppose that and . Then, . |E(G(i))| |E(G(i−1))| ≤ 1 + O(i−1) 𝔡 (i) ≥ Ω ( thit (Gi ) iγ ) E[U] = O(nγ) π(i−1)(v) π(i)(v) = deg(i−1)(v) 2|E(G(i−1))| ⋅ 2|E(G(i))| deg(i)(v) ≤ |E(G(i))| |E(G(i−1) | ≤ 1 + O(i−1) 5IFO max v∈V(G(j)) π(j−1)(v) π(j)(v) ( 1 − 1 thit (G(j))) 𝔡 (j) ≤ ( 1 − 1 thit (G(j))) (j)−O( thit(G(j)) j ) ≈ ( 1 − 1 thit (G(j))) (j) )FODF E[U] ≤ O(1) ⋅ n ∑ i=1 n ∏ j=i ( 1 − 1 thit (G(j))) 𝔡 (j)

  28. /28 28 $PODMVTJPO 28 ɾ8FJOUSPEVDF38POHSPXJOHHSBQIT ɾ"OBMZTJTJTUSBDUBCMF*G DIBOHFTNPEFSBUFMZ π ɾ$BOXFBQQMZUFDIOJRVFTPGQSFWJPVTXPSLTUPUIFHSPXJOHTFUUJOH 'VUVSF%JSFDUJPO

    ɾ)JUUJOHUJNF NJYJOHUJNF FUDPOHSPXJOHHSBQIT [Avin, Kousky, Lotler, 2008] [Sauerwald, Zanetti, 2019] [Oksana and Luís, 2014] [Cai, Sauerwald, Zanetti, 2020]