arxiv 2006 w h y hu y z you ucsd arxiv 2005 02636 w t
play

ArXiv:2006.????? w/ H.Y.Hu, Y.Z.You (UCSD) - PowerPoint PPT Presentation

Deep Learning and Physics 2020 Online seminar, 28 May 2020; Strings, Geometry and Data science workshop at Simons Center, Stony Brook NY, 6 Jan 2020; MRM2019 conference; Yokohama, 11 Dec 2019; Workshop on Discrete Geometry 2019, Tokyo


  1. “Deep Learning and Physics 2020 Online” seminar, 28 May 2020; “Strings, Geometry and Data science” workshop at Simons Center, Stony Brook NY, 6 Jan 2020; MRM2019 conference; Yokohama, 11 Dec 2019; Workshop on Discrete Geometry 2019, Tokyo CIC, 7 Dec 2019; Niigata U., 27 Nov 2019; NCTS-Kyoto joint workshop, Taiwan NCTS, Sep 6 2019; APCTP workshop on quantum matter, Aug 25 2019; Nordita workshop on holographic QCD, July 24 2019; QCD club, Keio U., July 8 2019; Arnold Sommerfeld Center, Munich, LMU, June 13 2019; Max Planck Institute for gravitational Physics, June 11 2019; Microsoft research, 25 April 2019; U. Washington, seminar, 24 April 2019; AdS/CFT workshop at OIST, 3 April 2019; “Machine learning landscape” workshop, ICTP Trieste, 10 Dec 2018; East Asian string workshop, KIAS, Seoul, 7 Nov 2018; KMI colloquium, Nagoya U, 25 Oct 2018; “QG meets lattice QCD workshop”, ECT*, Trento, 3 Sep 2018; APCTP focus program, Hanyang U, Seoul, 15 Aug 2018; QFT workshop, YITP, Kyoto, 31 July 2018; Machine learning workshop, TSiMF, China, 15 June, 2018; Paris QCD workshop, 11 June, 2018; DLAP2018 workshop, Osaka, 1 June, 2018; MPI, AEI, virtual seminar, 13 Apr, 2018; MIT, CTP, seminar, 4 Apr, 2018; CQuest, Sogang u., seminar, 29 March, 2018; KIAS, seminar, 26 March, 2018 深層学習と時空 橋本幸士 (阪大理) ArXiv:2006.????? w/ H.Y.Hu, Y.Z.You (UCSD) ArXiv:2005.02636 w/ T. Akutagawa, T. Sumimoto (Osaka u) ArXiv:1903.04951 ArXiv:1802.08313, 1809.10536 w/ S. Sugishita (Kentucky), A. Tanaka, A. Tomiya (RIKEN)

  2. 2

  3. 時空はニューラルネットワークか? 3

  4. 時空は NN か? 動機 1) 離散化した時空を NN と同一視して、 時空上の未知の力学系を、 時空発展のデータから学習できるか? 動機 2) あるデータで学習された NN 自体を、 時空そのものと見ることができるか? 4

  5. 時空はニューラルネットワーク 1. 空間は NN 2. 時間は NN 1802.08313 3. ホログラフィーは NN 1903.04951 4. 創発空間は NN 1802.08313 1809.10536 2006.????? 5. 創発時空は NN 2005.02636 6. 量子時空は NN?

  6. <latexit sha1_base64="aMToDsLjA4CF46hDv5BvReckLKU=">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</latexit> <latexit sha1_base64="YV9+ILxbpAg2lodAY9kyCUg13Xo=">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</latexit> 1. 空間は NN 一般の NN は空間で 一般の は空間ではな はない 局所性の不在:どのユニットがどれに「近い」? ボルツマンマシン パーセプトロン模型 [Ackley, Hinton, Sejnowski 1985] [Rosenblatt 1958] [Rumelhart, McClelland 1986] 2 3 ⇣ ⌘ f ( x i ) = W (2) W (1) X X jk x k P ( x i ) = exp w ij x i h j ϕ 4 − 5 j ij h j ∈ { 0 , 1 } 6

  7. <latexit sha1_base64="nwzfFCLRwUCM0jUcWfpi10n7t8=">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</latexit> <latexit sha1_base64="OAM1XRpoTDK6qJPFgGfv9fB7zD0=">ACv3ichVFNSxtBGH7cWhtj1VgvgpfFYLEUwsQqlp7UeujRr6hoNMxOxjhksrvsToIx+Af8Ax48VZAi/gnBi3/Ag39AEA8WLPTSQ9sFmzrR9lZ53vd5kZx9cqNIxdtlkv2l92vEp0Jrted/f0pvreLIVeNRAyJztBSsOD6VWrswZbRc8QPJK46Wy075czO/XJNBqDx30dR9uV7hJVdtKsENUYXUKrfz/pYacfMzUhtub7+z39sOcTwiuvC9iNp8Ud6Y/SRgrwoeiYspNIsw6KwH4JsDNKIY9ZLfUMeRXgQqKICReGsAZHSN8asmDwiVtHg7iAkIryErtIkrZKVZIqOLFlGku0WotZl9bNnmGkFrSLpj8gpY1hdsGO2R07Zyfsmv16slcj6tH0UqfZaWmlX+jdG1j4+V9VhWaDrXvVs54NvEx8qrIux8xzVOIlr62s3+38Gl+uPGWHbIb8v+VXbIzOoFb+yGO5uT8AZL0ANl/r/shWBrNZD9kxufG0pPT8VMkMIghjNB9T2ASXzCLHO17ivc4rs1ZUs1/JbpVZbrOnHX2HVfwPAU6ox</latexit> 1. 空間は NN sparse な NN は空間にな は空間になる 局所性なし 局所性適用 = 平行 移動 畳み込み層 全結合 [K. Fukushima `80] インプット : φ ( n ∆ x ) アウトプット : a φ ( n ∆ x ) + b ∂ x φ ( n ∆ x ) + c ∂ 2 x φ ( n ∆ x ) + · · · 7

  8. 2. 時間は NN ResNET と ODE ResNET (Residual network) [K.He et al.,1512.03385] x ( n +1) = f ( W ij x ( n ) j ) + x ( n ) i i “Skip connection” x ( n ) x ( n +1) f W i i 時間が離散化された力学系に等価 · x i = f i ( x ( t )) x i ( t n +1 ) = x i ( t n ) + Δ t ⋅ f i ( x ( t n )) t n +1 = t n + Δ t 時間を離散化 8

  9. 2. 時間は NN 力学は NN ハミ ハミルト ルトン力学は 1802.08313 q = ∂ H p = − ∂ H · · ∂ p , ∂ q 次の NN を試してみる : φ 1 q q ( t + Δ t ) = φ 1 ( W 11 q ( t ) + W 12 p ( t )) φ 2 p p ( t + Δ t ) = φ 2 ( W 21 q ( t ) + W 22 p ( t )) Δ t → 0 W ij = δ ij + Δ tw ij 極限の無矛盾性より φ i ( x ) = x + Δ tg i ( x ) · q = w 11 q + w 12 p + g 1 ( q ) · p = w 21 q + w 22 p + g 2 ( p ) トリビアルな線形ハミルトニアン … 9

  10. 2. 時間は NN 力学は NN ハミ ハミルト ルトン力学は 2 w 21 q 2 + λ 1 v F ( vp ) − λ 2 H = w 11 pq + 1 2 w 12 p 2 − 1 u G ( uq ) ( F ′ � = f , G ′ � = g ) 0 0 0 v 0 1 + Δ tw 11 Δ tw 12 0 W = 0 Δ tw 21 1 + Δ tw 12 0 q 0 0 0 u p Δ tf ( x ) 0 0 0 0 λ 1 1 0 0 1 ˜ φ i = W = 1 0 0 1 λ 2 ˜ W ij φ i W ij Δ tg ( x ) 0 0 0 0 1 0

  11. 2. 時間は NN 波動関数の解釈? 波動関数の解釈? N 個のスピン系の基底状態の波動関数? ψ ( s 1 , s 2 , ⋯ , s N ) 「ハミルトニアン H が与えられた時、エネルギー E を最小にせよ」 制限ボルツマンマシン状態 [Carleo Troyer `17], [Nomura, Darmawan, Yamaji, Imada `17], .. E 例:2次元反強磁性 ハイゼンベルグ模型 隠れ層の数

  12. <latexit sha1_base64="AvgKTR8IQdExyctru1rwnMXbrXc=">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</latexit> <latexit sha1_base64="/eEqvXEnTni57AUEIhrBngQ2w0=">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</latexit> <latexit sha1_base64="XagScydK2LYSdBnKEe0o0+PseEo=">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</latexit> 2. 時間は NN 深層ボルツマンマシンはユーク 深層ボルツ はユークリッド化時間 化時間 ハミルトニアンが与えられた時の基底状態の波動関数は 深層ボルツマンマシンと同定される [Carleo, Nomura, Imada `18], .. τ →∞ e − τ H | any ⟩ = e −Δ τ H e −Δ τ H ⋯ | any ⟩ | ψ ⟩ = lim H τ 2 3 w (0) w ( n ) ij h ( n ) h ( n +1) X X X X ψ ( x i ) = exp ij x i h j − 4 − 5 i j h ( n ) ij n ij 1 ∈ { 0 , 1 } j 2

Recommend


More recommend