deep neural networks reveal a gradient in the complexity
play

Deep Neural Networks Reveal a Gradient in the Complexity of Neural - PowerPoint PPT Presentation

Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Umut Gcl and Marcel A. J. van Gerven Article overview by Ilya Kuzovkin Computational Neuroscience Seminar University of Tartu


  1. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Umut Güclü and Marcel A. J. van Gerven Article overview by Ilya Kuzovkin Computational Neuroscience Seminar University of Tartu 2015

  2. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  3. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream classes “cat” “spider” … pixels Linear

  4. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream classes “cat” hidden layer … “spider” … pixels Non-linear

  5. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream classes “cat” hidden layer … hidden layer … “spider” … pixels Deep

  6. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream “cat” important feature “spider”

  7. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream “cat” important feature “spider” RUN!

  8. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream “cat” important feature “spider” Convolutional filter RUN!

  9. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Convolutional (and pooling) layer

  10. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream classes “cat” hidden layer … hidden layer … convolutional layer “spider” … … pixels Deep Convolutional Neural Network

  11. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  12. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  13. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  14. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Matthew D. Zeiler, Rob Fergus Visualizing and Understanding Convolutional Networks 2013

  15. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Matthew D. Zeiler, Rob Fergus Visualizing and Understanding Convolutional Networks 2013

  16. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Matthew D. Zeiler, Rob Fergus Visualizing and Understanding Convolutional Networks 2013

  17. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Matthew D. Zeiler, Rob Fergus Visualizing and Understanding Convolutional Networks 2013

  18. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Matthew D. Zeiler, Rob Fergus Visualizing and Understanding Convolutional Networks 2013

  19. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  20. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream Two-stream hypothesis

  21. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  22. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  23. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  24. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  25. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  26. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  27. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream ?

  28. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  29. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  30. 96 x 37x 37 = 131,424

  31. 96 x 37x 37 = 131,424

  32. 96 x 37x 37 = 131,424 256 x 17x 17 = 73,984

  33. 96 x 37x 37 = 131,424 256 x 17x 17 = 73,984

  34. 96 x 37x 37 = 131,424 256 x 17x 17 = 73,984

  35. 96 x 37x 37 = 131,424 256 x 17x 17 = 73,984

  36. Train linear regression model 96 x 37x 37 = 131,424 256 x 17x 17 = 73,984

  37. Train linear regression model Test it 96 x 37x 37 = 131,424 256 x 17x 17 = 73,984

  38. 96 x 37x 37 = 131,424 Train linear regression model r = 0.22 Test it 256 x 17x 17 = 73,984

  39. 96 x 37x 37 = 131,424 Train linear regression model r = 0.22 Test it 256 x 17x 17 = 73,984 Train linear regression model Test it

  40. 96 x 37x 37 = 131,424 Train linear regression model r = 0.22 Test it 256 x 17x 17 = 73,984 Train linear regression model r = 0.67 Test it

  41. 96 x 37x 37 = 131,424 Train linear regression model r = 0.22 Test it 256 x 17x 17 = 73,984 Train linear regression model r = 0.67 Test it

  42. Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream

  43. N EXT COOL THING : CATEGORIES OF FEATURES ImageNet validation set …

  44. N EXT COOL THING : CATEGORIES OF FEATURES ImageNet validation set ... . … 1888

  45. N EXT COOL THING : CATEGORIES OF FEATURES ImageNet validation set ... . … 1888

  46. N EXT COOL THING : CATEGORIES OF FEATURES ImageNet validation set ... . … 1888 .

  47. N EXT COOL THING : CATEGORIES OF FEATURES ImageNet validation set ... . … 1888 . . deconvolution

  48. N EXT COOL THING : CATEGORIES OF FEATURES ImageNet validation set ... . … 1888 . . deconvolution human-assigned to 9 categories Low Mid High • blob • contour • pattern • contrast • shape • object • edge • texture • object part

  49. N EXT COOL THING : CATEGORIES OF FEATURES 1. Divide 1888 neurons into 9 ImageNet validation set ... . categories … 1888 . . deconvolution human-assigned to 9 categories Low Mid High • blob • contour • pattern • contrast • shape • object • edge • texture • object part

  50. N EXT COOL THING : CATEGORIES OF FEATURES 1. Divide 1888 neurons into 9 ImageNet validation set ... . categories … 2. Predict activity of each voxel 1888 from group-by-group . . deconvolution human-assigned to 9 categories Low Mid High • blob • contour • pattern • contrast • shape • object • edge • texture • object part

  51. N EXT COOL THING : CATEGORIES OF FEATURES 1. Divide 1888 neurons into 9 ImageNet validation set ... . categories … 2. Predict activity of each voxel 1888 from group-by-group . 3. For each voxel find the group, which best predicts . voxel’s activity deconvolution human-assigned to 9 categories Low Mid High • blob • contour • pattern • contrast • shape • object • edge • texture • object part

  52. N EXT COOL THING : CATEGORIES OF FEATURES 1. Divide 1888 neurons into 9 ImageNet validation set ... . categories … 2. Predict activity of each voxel 1888 from group-by-group . 3. For each voxel find the group, which best predicts . voxel’s activity deconvolution human-assigned 4. Assign each of 1888 DNN to 9 categories neurons to a visual layer: V1, V2, V4, LO Low Mid High • blob • contour • pattern • contrast • shape • object • edge • texture • object part

  53. N EXT COOL THING : CATEGORIES OF FEATURES 1. Divide 1888 neurons into 9 ImageNet validation set ... . categories … 2. Predict activity of each voxel 1888 from group-by-group . 3. For each voxel find the group, which best predicts . voxel’s activity deconvolution human-assigned 4. Assign each of 1888 DNN to 9 categories neurons to a visual layer: V1, V2, V4, LO Low Mid High • blob • contour • pattern 5. Map visual layers to • contrast • shape • object categories • edge • texture • object part

  54. N EXT COOL THING : CATEGORIES OF FEATURES

  55. O THER RESULTS Correlation between predicted responses between pairs of voxel groups

  56. O THER RESULTS Selectivity of visual areas to feature maps of varying complexity

  57. O THER RESULTS Distribution of the receptive field centers

  58. O THER RESULTS Biclustering of voxels and feature maps

  59. S UMMARY

  60. An intracranial dataset we have. How to repeat the result?

  61. An intracranial dataset we have. How to repeat the result? vs.

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