constant curvature graph convolutional networks
play

Constant Curvature Graph Convolutional Networks Gregor Gary - PowerPoint PPT Presentation

Constant Curvature Graph Convolutional Networks Gregor Gary Octavian Bachmann B ecigneul Ganea ETH Z urich MIT MIT Overview Overview Embeddings of graphs into hyperbolic and spherical space and their products Overview


  1. Constant Curvature Graph Convolutional Networks Gregor Gary Octavian Bachmann B´ ecigneul Ganea ETH Z¨ urich MIT MIT

  2. Overview

  3. Overview • Embeddings of graphs into hyperbolic and spherical space and their products

  4. Overview • Embeddings of graphs into hyperbolic and spherical space and their products • Extend gyrovector framework to spherical geometry and provide a unifying formalism

  5. Overview • Embeddings of graphs into hyperbolic and spherical space and their products • Extend gyrovector framework to spherical geometry and provide a unifying formalism • Introduce graph neural networks producing embeddings in product spaces

  6. Overview • Embeddings of graphs into hyperbolic and spherical space and their products • Extend gyrovector framework to spherical geometry and provide a unifying formalism • Introduce graph neural networks producing embeddings in product spaces • Differentiable transitions in geometry during training in each component

  7. Graphs

  8. Graphs • Lots of data available in the form of graphs (social networks, railway tracks, phylogenetic trees etc.)

  9. Graphs • Lots of data available in the form of graphs (social networks, railway tracks, phylogenetic trees etc.)

  10. Graphs • Lots of data available in the form of graphs (social networks, railway tracks, phylogenetic trees etc.) • Node set V = { 1 , . . . , n } and adjacency matrix A ∈ R n × n

  11. Where to Embed Graphs?

  12. Where to Embed Graphs? • Euclidean geometry not suitable for many graphs

  13. Where to Embed Graphs? • Euclidean geometry not suitable for many graphs

  14. Where to Embed Graphs? • Euclidean geometry not suitable for many graphs • Graph distance d G ( i , j ) = ”Shortest path from i to j” not respected in Euclidean embedding

  15. Where to Embed Graphs? • Euclidean geometry not suitable for many graphs • Graph distance d G ( i , j ) = ”Shortest path from i to j” not respected in Euclidean embedding • Arbitrary low distortion in spherical and hyperbolic space

  16. Non-Euclidean Geometry

  17. Non-Euclidean Geometry • Focus on constant sectional curvature manifolds

  18. Non-Euclidean Geometry • Focus on constant sectional curvature manifolds • Well-studied in the field of Differential Geometry

  19. Non-Euclidean Geometry • Focus on constant sectional curvature manifolds • Well-studied in the field of Differential Geometry • Computationally attractive expressions for distance, exponential map etc.

  20. Hyperbolic Space as Poincar´ e Ball

  21. Hyperbolic Space as Poincar´ e Ball • H n = { x : || x || 2 ≤ 1 √ c } with curvature − c equipped with 4 Riemannian tensor g c x = (1 − c || x || 2 ) 2 1

  22. Hyperbolic Space as Poincar´ e Ball • H n = { x : || x || 2 ≤ 1 √ c } with curvature − c equipped with 4 Riemannian tensor g c x = (1 − c || x || 2 ) 2 1 • Projection of hyperboloid

  23. Hyperbolic Space as Poincar´ e Ball • H n = { x : || x || 2 ≤ 1 √ c } with curvature − c equipped with 4 Riemannian tensor g c x = (1 − c || x || 2 ) 2 1 • Projection of hyperboloid � � 2 c || x − y || 2 √ c cosh − 1 1 • d c H ( x , y ) = 1 + 2 ( 1 2 )( 1 2 ) c −|| x || 2 c −|| y || 2 Heatmap of d κ Projection of hyperboloid [4] H

  24. Gyrospace Structure

  25. Gyrospace Structure • Next best thing to a vector space

  26. Gyrospace Structure • Next best thing to a vector space • Vector addition x + y �→ x ⊕ c y

  27. Gyrospace Structure • Next best thing to a vector space • Vector addition x + y �→ x ⊕ c y • Scalar multiplication r x �→ r ⊗ c x

  28. Gyrospace Structure • Next best thing to a vector space • Vector addition x + y �→ x ⊕ c y • Scalar multiplication r x �→ r ⊗ c x • Geodesic γ x − → y ( t ) = x ⊕ c ( t ⊗ c ( − x ⊕ c y ))

  29. Spherical Space as Stereographic Projection

  30. Spherical Space as Stereographic Projection • Stereographic projection of S d +1 ∼ = R d + g c x where g c 4 x = (1+ c || x || 2 ) 2 1

  31. Spherical Space as Stereographic Projection • Stereographic projection of S d +1 ∼ = R d + g c x where g c 4 x = (1+ c || x || 2 ) 2 1 � � 2 c || x − y || 2 • d c √ c cos − 1 1 2 S ( x , y ) = 1 + ( 1 2 )( 1 c + || x || 2 c + || y || 2 2 )

  32. Spherical Space as Stereographic Projection • Stereographic projection of S d +1 ∼ = R d + g c x where g c 4 x = (1+ c || x || 2 ) 2 1 � � 2 c || x − y || 2 • d c √ c cos − 1 1 2 S ( x , y ) = 1 + ( 1 2 )( 1 c + || x || 2 c + || y || 2 2 )

  33. Our Contributions: 1) Unified Formalism

  34. Our Contributions: 1) Unified Formalism • κ - stereographic model for any κ ∈ R : κ = { x ∈ R d | − κ � x � 2 st d 2 < 1 }

  35. Our Contributions: 1) Unified Formalism • κ - stereographic model for any κ ∈ R : κ = { x ∈ R d | − κ � x � 2 st d 2 < 1 } R d st d κ (1 − 2 κ x T y − κ || y || 2 ) x +(1+ κ || x || 2 ) y x ⊕ κ y x + y 1 − 2 κ x T y + κ 2 || x || 2 || y || 2 tan κ ( r · tan − 1 κ || x || ) x r ⊗ κ x r x || x || γ x → y ( t ) x + t ( y − x ) x ⊕ κ ( t ⊗ κ ( − x ⊕ κ y ))

  36. Our Contributions: 1) Unified Formalism • κ - stereographic model for any κ ∈ R : κ = { x ∈ R d | − κ � x � 2 st d 2 < 1 } R d st d κ (1 − 2 κ x T y − κ || y || 2 ) x +(1+ κ || x || 2 ) y x ⊕ κ y x + y 1 − 2 κ x T y + κ 2 || x || 2 || y || 2 tan κ ( r · tan − 1 κ || x || ) x r ⊗ κ x r x || x || γ x → y ( t ) x + t ( y − x ) x ⊕ κ ( t ⊗ κ ( − x ⊕ κ y )) • More unifying expressions for distance , exponential map etc. in our paper!

  37. Our Contributions: 2) Matrix Multiplications

  38. Our Contributions: 2) Matrix Multiplications κ , W ∈ R d × k and A ∈ R n × n • Embeddings X where X i • ∈ st d

  39. Our Contributions: 2) Matrix Multiplications κ , W ∈ R d × k and A ∈ R n × n • Embeddings X where X i • ∈ st d • Right matrix multiplication XW acts on columns X • i Thus lift to tangent space at zero: ( X ⊗ κ W ) i • = exp κ 0 ((log κ 0 ( X ) W ) i • )

  40. Our Contributions: 2) Matrix Multiplications κ , W ∈ R d × k and A ∈ R n × n • Embeddings X where X i • ∈ st d • Right matrix multiplication XW acts on columns X • i Thus lift to tangent space at zero: ( X ⊗ κ W ) i • = exp κ 0 ((log κ 0 ( X ) W ) i • ) • Introduced in [2], we extended it to spherical spaces

  41. Our Contributions: 2) Matrix Multiplications

  42. Our Contributions: 2) Matrix Multiplications • Left matrix multiplication AX acts on rows X i • : ( AX ) i • = A i 1 X 1 • + · · · + A in X n •

  43. Our Contributions: 2) Matrix Multiplications • Left matrix multiplication AX acts on rows X i • : ( AX ) i • = A i 1 X 1 • + · · · + A in X n • • Idea: Reduce problem of linear combination to definition of a non-euclidean midpoint

  44. Our Contributions: 2) Matrix Multiplications • Left matrix multiplication AX acts on rows X i • : ( AX ) i • = A i 1 X 1 • + · · · + A in X n • • Idea: Reduce problem of linear combination to definition of a non-euclidean midpoint

  45. Our Contributions: 2) Matrix Multiplications

  46. Our Contributions: 2) Matrix Multiplications • Leverage gyromidpoint for hyperbolic space and extend it to st d κ : � n � m κ ( x 1 , · · · , x n ; α ) = 1 α i λ κ � x i 2 ⊗ κ x j − 1) x i � n j =1 α j ( λ κ i =1

  47. Our Contributions: 2) Matrix Multiplications • Leverage gyromidpoint for hyperbolic space and extend it to st d κ : � n � m κ ( x 1 , · · · , x n ; α ) = 1 α i λ κ � x i 2 ⊗ κ x j − 1) x i � n j =1 α j ( λ κ i =1 • Define left matrix multiplication row-wise: � ( A ⊠ κ X ) i • := ( A ij ) ⊗ κ m κ ( X 1 • , · · · , X n • ; A i • ) j

  48. Our Contributions: 2) Matrix Multiplications • Leverage gyromidpoint for hyperbolic space and extend it to st d κ : � n � m κ ( x 1 , · · · , x n ; α ) = 1 α i λ κ � x i 2 ⊗ κ x j − 1) x i � n j =1 α j ( λ κ i =1 • Define left matrix multiplication row-wise: � ( A ⊠ κ X ) i • := ( A ij ) ⊗ κ m κ ( X 1 • , · · · , X n • ; A i • ) j • Same scaling behaviour: d κ ( 0 , r ⊗ κ x ) = r · d κ ( 0 , x )

  49. Gyromidpoint for Varying Curvature

  50. Our Contributions: 3) Differentiable Interpolation

  51. Our Contributions: 3) Differentiable Interpolation → 0 ± • All quantities recover their Euclidean counterpart for κ −

  52. Our Contributions: 3) Differentiable Interpolation → 0 ± • All quantities recover their Euclidean counterpart for κ − • We proved an even stronger result:

  53. Our Contributions: 3) Differentiable Interpolation → 0 ± • All quantities recover their Euclidean counterpart for κ − • We proved an even stronger result: Differentiability of st d κ w.r.t. κ around 0 The first order derivatives at 0 − and 0 + w.r.t. to κ of all the mentioned quantities exist and are equal .

Download Presentation
Download Policy: The content available on the website is offered to you 'AS IS' for your personal information and use only. It cannot be commercialized, licensed, or distributed on other websites without prior consent from the author. To download a presentation, simply click this link. If you encounter any difficulties during the download process, it's possible that the publisher has removed the file from their server.

Recommend


More recommend