all you want to know about cnns
play

All You Want To Know About CNNs Yukun Zhu Deep Learning Deep - PowerPoint PPT Presentation

Mar 24, 2023 •357 likes •1.15k views

All You Want To Know About CNNs Yukun Zhu Deep Learning Deep Learning Image from http://imgur.com/ Deep Learning Image from http://imgur.com/ Deep Learning Image from http://imgur.com/ Deep Learning Image from http://imgur.com/ Deep

  1. All You Want To Know About CNNs Yukun Zhu

  2. Deep Learning

  3. Deep Learning Image from http://imgur.com/

  4. Deep Learning Image from http://imgur.com/

  5. Deep Learning Image from http://imgur.com/

  6. Deep Learning Image from http://imgur.com/

  7. Deep Learning Image from http://imgur.com/

  8. Deep Learning Image from http://imgur.com/

  9. Deep Learning in Vision Object detection performance, PASCAL VOC 2010 33.4 DPM (2010)

  10. Deep Learning in Vision Object detection performance, PASCAL VOC 2010 40.4 33.4 DPM segDPM (2010) (2014)

  11. Deep Learning in Vision Object detection performance, PASCAL VOC 2010 53.7 40.4 33.4 DPM segDPM RCNN (2010) (2014) (2014)

  12. Deep Learning in Vision Object detection performance, PASCAL VOC 2010 62.9 53.7 40.4 33.4 DPM segDPM RCNN RCNN* (2010) (2014) (2014) (Oct 2014)

  13. Deep Learning in Vision Object detection performance, PASCAL VOC 2010 67.2 62.9 53.7 40.4 33.4 DPM segDPM RCNN RCNN* segRCNN (2010) (2014) (2014) (Oct 2014) (Jan 2015)

  14. Deep Learning in Vision Object detection performance, PASCAL VOC 2010 70.8 67.2 62.9 53.7 40.4 33.4 DPM segDPM RCNN RCNN* segRCNN Fast RCNN (2010) (2014) (2014) (Oct 2014) (Jan 2015) (Jun 2015)

  15. A Neuron Image from http://cs231n.github.io/neural-networks-1/

  16. A Neuron in Neural Network Image from http://cs231n.github.io/neural-networks-1/

  17. Activation Functions ● Sigmoid: f(x) = 1 / (1 + e -x ) ● ReLU: f(x) = max(0, x) ● Leaky ReLU: f(x) = max(ax, x) ● Maxout: f(x) = max(w 0 x + b 0 , w 1 x + b 1 ) ● and many others…

  18. Neural Network (MLP) The network simulates a function y = f(x; w) Image modified from http://cs231n.github.io/neural-networks-1/

  19. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -1.00 -3.00 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  20. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 -3.00 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  21. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 -3.00 6.00 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  22. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  23. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  24. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  25. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  26. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  27. x 0 Forward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  28. Loss Function Loss function measures how well prediction matches true value Commonly used loss function: ● Squared loss: (y - y’) 2 ● Cross-entropy loss: -sum i (y i ’ * log(y i )) ● and many others

  29. Loss Function During training, we would like to minimize the total loss on a set of training data ● We want to find w* = argmin{sum i [loss(f(x i ; w), y i )]}

  30. Loss Function During training, we would like to minimize the total loss on a set of training data ● We want to find w* = argmin{sum i [loss(f(x i ; w), y i )]} ● Usually we use gradient based approach w t+1 = w t - a ∇ w ○

  31. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 1.00 -3.00 Image and code modified from http://cs231n.github.io/optimization-2/

  32. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 -0.53 1.00 f = 1/x -3.00 df/dx = -1/x 2 Image and code modified from http://cs231n.github.io/optimization-2/

  33. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 -0.53 -0.53 1.00 f = x + 1 -3.00 df/dx = 1 Image and code modified from http://cs231n.github.io/optimization-2/

  34. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 -0.20 -0.53 -0.53 1.00 f = e x -3.00 df/dx = e x Image and code modified from http://cs231n.github.io/optimization-2/

  35. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 0.20 -0.20 -0.53 -0.53 1.00 f = -x -3.00 df/dx = -1 Image and code modified from http://cs231n.github.io/optimization-2/

  36. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 4.00 0.20 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 0.20 -0.20 -0.53 -0.53 1.00 f = x + a df/dx = 1 -3.00 0.20 Image and code modified from http://cs231n.github.io/optimization-2/

  37. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -2.00 -1.00 0.20 4.00 0.20 -3.00 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 0.20 0.20 -0.20 -0.53 -0.53 1.00 -3.00 0.20 Image and code modified from http://cs231n.github.io/optimization-2/

  38. x 0 Backward Computation x 1 sigmoid f(x 0 , x 1 ) = 1 / (1 + exp(-(w 0 x 0 + w 1 x 1 + w 2 ))) 1 2.00 -0.20 -2.00 -1.00 0.20 0.40 4.00 f = ax df/dx = a 0.20 -3.00 -0.40 6.00 1.00 -1.00 0.37 1.37 0.73 -2.00 0.20 0.20 -0.20 -0.53 -0.53 1.00 -0.60 -3.00 0.20 Image and code modified from http://cs231n.github.io/optimization-2/

  39. Why NNs?

  40. Universal Approximation Theorem A feed-forward network with a single hidden layer containing a finite number of neurons, can approximate continuous functions on compact subsets of R n , under mild assumptions on the activation function. https://en.wikipedia.org/wiki/Universal_approximation_theorem

  41. Stone’s Theorem ● Suppose X is a compact Hausdorff space and B is a subalgebra in C(X, R) such that: B separates points. ○ B contains the constant function 1. ○ If f ∈ B then a f ∈ B for all constants a ∈ R. ○ If f , g ∈ B, then f + g , max{ f, g } ∈ B. ○ ● Then every continuous function defined on C(X, R) can be approximated as closely as desired by functions in B

  42. Why CNNs?

  43. Problems of MLP in Vision For input as a 10 * 10 image: ● A 3 layer MLP with 200 hidden units contains ~100k parameters For input as a 100 * 100 image: ● A 1 layer MLP with 20k hidden units contains ~200m parameters

  44. Can We Do Better?

  45. Can We Do Better?

  46. Can We Do Better?

  47. Can We Do Better?

  48. Can We Do Better? Based on such observation, MLP can be improved in two ways: ● Locally connected instead of fully connected ● Sharing weights between neurons We achieve those by using convolution neurons

  49. Convolutional Layers Image from http://cs231n.github.io/convolutional-networks/

  50. Convolutional Layers width height depth Image from http://cs231n.github.io/convolutional-networks/. See this page for an excellent example of convolution.

  51. Pooling Layers Image from http://cs231n.github.io/convolutional-networks/

  52. Pooling Layers Example: Max Pooling Image from http://cs231n.github.io/convolutional-networks/

  53. Pooling Layers Commonly used pooling layers: ● Max pooling ● Average pooling Why pooling layers? ● Reduce activation dimensionality ● Robust against tiny shifts

  54. CNN Architecture: An Example Image from http://cs231n.github.io/convolutional-networks/

  55. Layer Activations for CNNs Conv:1 ReLU:1 Conv:2 ReLU:2 MaxPool:1 Conv:3 Image modified from http://cs231n.github.io/convolutional-networks/

  56. Layer Activations for CNNs MaxPool:2 Conv:5 ReLU:5 Conv:6 ReLU:6 MaxPool:3 Image modified from http://cs231n.github.io/convolutional-networks/

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

Understanding Geometry of Encoder-Decoder CNNs (E-D CNNs) Jong Chul Ye & Woon Kyoung Sung

Understanding Geometry of Encoder-Decoder CNNs (E-D CNNs) Jong Chul Ye & Woon Kyoung Sung

Understanding Geometry of Encoder-Decoder CNNs (E-D CNNs) Jong Chul Ye & Woon Kyoung Sung BISPL - BioImaging, Signal Processing and Learning Lab. Dept. Bio & Brain Engineering Dept. of Mathematical Sciences KAIST, Korea E-D CNN

911 views • 26 slides
Table of Contents Convolutional Neural Nets (CNNs) 1 Deep Q Learning 2 Lecture 6: CNNs and Deep

Table of Contents Convolutional Neural Nets (CNNs) 1 Deep Q Learning 2 Lecture 6: CNNs and Deep

Lecture 6: CNNs and Deep Q Learning 1 Emma Brunskill CS234 Reinforcement Learning. Winter 2019 1 With many slides for DQN from David Silver and Ruslan Salakhutdinov and some vision slides from Gianni Di Caro and images from Stanford CS231n,

1.17k views • 68 slides
Table of Contents Convolutional Neural Nets (CNNs) 1 Deep Q Learning 2 Lecture 6: CNNs and Deep

Table of Contents Convolutional Neural Nets (CNNs) 1 Deep Q Learning 2 Lecture 6: CNNs and Deep

Lecture 6: CNNs and Deep Q Learning 2 Emma Brunskill CS234 Reinforcement Learning. Winter 2018 2 With many slides for DQN from David Silver and Ruslan Salakhutdinov and some vision slides from Gianni Di Caro and images from Stanford CS231n,

850 views • 67 slides
Geirhos et al. (2019) Introduction ImageNet classifjcation with CNNs Which image cues are

Geirhos et al. (2019) Introduction ImageNet classifjcation with CNNs Which image cues are

IMAGENET-TRAINED CNNS ARE BIASED TOWARDS TEXTURE; INCREASING SHAPE BIAS IMPROVES ACCURACY AND ROBUSTNESS Geirhos et al. (2019) Introduction ImageNet classifjcation with CNNs Which image cues are learned How infmuential they are

221 views • 18 slides
Deep Learning for Geometry Processing 3D Representations View-Based and Volumetric CNNs 3D

Deep Learning for Geometry Processing 3D Representations View-Based and Volumetric CNNs 3D

Deep Learning for Geometry Processing 3D Representations View-Based and Volumetric CNNs 3D Representations for Object Classification Multi-View CNNs Su et al. 2015 Multi-View CNNs Su et al. 2015 Multi-View CNNs Su et al. 2015 Multi-View

876 views • 86 slides
Distributed Optimization of CNNs and RNNs GTC 2015 William Chan williamchan.ca

Distributed Optimization of CNNs and RNNs GTC 2015 William Chan williamchan.ca

Distributed Optimization of CNNs and RNNs GTC 2015 William Chan williamchan.ca williamchan@cmu.edu March 19, 2015 Outline 1. Motivation 2. Distributed ASGD 3. CNNs 4. RNNs 5. Conclusion Carnegie Mellon University 2 Motivation Why

357 views • 22 slides
Training R-CNNs of various velocities Slow, fast, and faster

Training R-CNNs of various velocities Slow, fast, and faster

Training R-CNNs of various velocities Slow, fast, and faster Ross Girshick Facebook AI Research (FAIR) Tools for Efficient Object Detection, ICCV 2015 Tutorial Section

641 views • 51 slides
Introduction to CNNs and RNNs with PyTorch Introduction to CNNs and RNNs with PyTorch Presented

Introduction to CNNs and RNNs with PyTorch Introduction to CNNs and RNNs with PyTorch Presented

Introduction to CNNs and RNNs with PyTorch Introduction to CNNs and RNNs with PyTorch Presented by: Adam Balint Presented by: Adam Balint Email: balint@uoguelph.ca Email: balint@uoguelph.ca Working with more complex data Working with more

608 views • 25 slides
Part-based R-CNNs for Fine-grained Category Detec7on

Part-based R-CNNs for Fine-grained Category Detec7on

Part-based R-CNNs for Fine-grained Category Detec7on Ning Zhang Jeff Donahue Ross Girshick Trevor Darrell

741 views • 39 slides
OPENING 15 Sept 2018 Bloombergs Where to go in 2018 CNNs Metropolises on

OPENING 15 Sept 2018 Bloombergs Where to go in 2018 CNNs Metropolises on

OPENING 15 Sept 2018 Bloombergs Where to go in 2018 CNNs Metropolises on the up: 7 design - savvy cities to watch in 2018 Over 40 local and regional newspapers named Dundee as a top 2018 holiday destination

417 views • 25 slides
Learn earn CN CNNs Ns fr from om Lar arge ge-scale cale We Web b Im Images ages wi

Learn earn CN CNNs Ns fr from om Lar arge ge-scale cale We Web b Im Images ages wi

Learn earn CN CNNs Ns fr from om Lar arge ge-scale cale We Web b Im Images ages wi without hout Hu Human an An Anno notations tations Weilin Huang Malong Technologies Ho How t w to Trai ain a H a High-Perf erfor orma

548 views • 31 slides
Volumetric and Multi-View CNNs for Object Classification on 3D Data Charles R. Qi*, Hao Su*,

Volumetric and Multi-View CNNs for Object Classification on 3D Data Charles R. Qi*, Hao Su*,

Volumetric and Multi-View CNNs for Object Classification on 3D Data Charles R. Qi*, Hao Su*, Matthias Niener, Angela Dai, MengyuanYan, Leonidas J.Guibas Rich Applications of 3D Augmented Robot Reality Perception 3D Representations for

402 views • 25 slides
Advanced CNN Architectures Akshay Mishra, Hong Cheng CNNs are everywhere... Recommendation

Advanced CNN Architectures Akshay Mishra, Hong Cheng CNNs are everywhere... Recommendation

Advanced CNN Architectures Akshay Mishra, Hong Cheng CNNs are everywhere... Recommendation Systems Drug Discovery Physics simulations Heng-Tze Cheng, Levent Koc, Jeremiah Harmsen, Tal Shaked, Tushar Chandra, Hrishi Aradhye, Glen Anderson,

1.59k views • 103 slides
I ncremental N etwork Q uantization: Towards Lossless CNNs With Low-Precision Weights Aojun Zhou,

I ncremental N etwork Q uantization: Towards Lossless CNNs With Low-Precision Weights Aojun Zhou,

I ncremental N etwork Q uantization: Towards Lossless CNNs With Low-Precision Weights Aojun Zhou, Anbang Yao, Yiwen Guo, Lin Xu, Yurong Chen Presented by Zhuangwei Zhuang South China University of Technology June 6, 2017 1 Outline

466 views • 23 slides
CNNs Applications M. Soleymani Sharif University of Technology Spring 2019 Most slides have

CNNs Applications M. Soleymani Sharif University of Technology Spring 2019 Most slides have

CNNs Applications M. Soleymani Sharif University of Technology Spring 2019 Most slides have been adopted from Fei Fei Li and colleagues lectures, cs231n, Stanford 2017. AlexNet [Krizhevsky, Sutskever, Hinton, 2012] ImageNet Classification

939 views • 68 slides
Interpreting chest X-rays via CNNs that exploit hierarchical disease dependencies and uncertainty

Interpreting chest X-rays via CNNs that exploit hierarchical disease dependencies and uncertainty

Interpreting chest X-rays via CNNs that exploit hierarchical disease dependencies and uncertainty labels by Hieu H. Pham, Tung T. Le, Dat Q. Tran, Dat T. Ngo, and Ha Q. Nguyen Medical Imaging Department Vingroup Big Data Institute (VinBDI),

193 views • 5 slides
Spa$ally Coupled Turbo-Like Codes: Convolu$onal Codes on Graphs

Spa$ally Coupled Turbo-Like Codes: Convolu$onal Codes on Graphs

Spa$ally Coupled Turbo-Like Codes: Convolu$onal Codes on Graphs Saeedeh Moloudi Lund University, Sweden Joint work with Michael Lentmaier, Lund

63 views • 5 slides
Lecture 04 Reliable Communication I-Hsiang Wang ihwang@ntu.edu.tw National Taiwan University

Lecture 04 Reliable Communication I-Hsiang Wang ihwang@ntu.edu.tw National Taiwan University

Principle of Communications, Fall 2017 Lecture 04 Reliable Communication I-Hsiang Wang ihwang@ntu.edu.tw National Taiwan University 2017/10/25,26 Channel Coding Binary Interface x ( t ) { c i } { u m } x b ( t ) ECC Symbol Pulse Up { b

1.08k views • 60 slides
Sem Semanti tic c segm segmen enta tati tion on CV3DST | Prof. Leal-Taix 1 Ta Task d

Sem Semanti tic c segm segmen enta tati tion on CV3DST | Prof. Leal-Taix 1 Ta Task d

Sem Semanti tic c segm segmen enta tati tion on CV3DST | Prof. Leal-Taix 1 Ta Task d defin init itio ion: s semantic ic s segm gmentatio ion Classify the main object in the image. CAT , GRASS, TREE, SKY No objects, just

1.29k views • 95 slides
Overcoming Delay, Synchronization and Cyclic Paths Meir Feder Tel Aviv University joint work

Overcoming Delay, Synchronization and Cyclic Paths Meir Feder Tel Aviv University joint work

Overcoming Delay, Synchronization and Cyclic Paths Meir Feder Tel Aviv University joint work with Elona Erez and Cyclic Paths 1 Root of the problem In much of network coding assume instantaneous coding

651 views • 44 slides
Extracting and Modeling Relations with Graph Convolutional Networks Ivan Titov with Diego

Extracting and Modeling Relations with Graph Convolutional Networks Ivan Titov with Diego

Extracting and Modeling Relations with Graph Convolutional Networks Ivan Titov with Diego Marcheggiani, Michael Schlichtkrull, Thomas Kipf, Max Welling, Rianne van den Berg and Peter Bloem 1 Inferring missing facts in knowledge bases: link

827 views • 66 slides
Assignment 5 Zahra Sheikhbahaee Zeou Hu & Colin Vandenhof April 2020 1 Convolutional

Assignment 5 Zahra Sheikhbahaee Zeou Hu & Colin Vandenhof April 2020 1 Convolutional

Assignment 5 Zahra Sheikhbahaee Zeou Hu & Colin Vandenhof April 2020 1 Convolutional Neural Networks Basics Consider the following CNN architecture: conv: conv: input: 100 filters 50 filters pooling: size 3 3 size 5 5 256

110 views • 6 slides
BRDF Evolution BRDFs have evolved historically 1970s: Empirical models Phongs

BRDF Evolution BRDFs have evolved historically 1970s: Empirical models Phongs

BRDF Evolution BRDFs have evolved historically 1970s: Empirical models Phongs illumination model 1980s: Physically based models Microfacet models (e.g. Cook Torrance model) 1990s Physically based

876 views • 42 slides
Computer Graphics - Material Models - Philipp Slusallek & Arsne Prard-Gayot Overview

Computer Graphics - Material Models - Philipp Slusallek & Arsne Prard-Gayot Overview

Computer Graphics - Material Models - Philipp Slusallek & Arsne Prard-Gayot Overview Last time Light: radiance & light sources Light transport: rendering equation & formal solutions Today Reflectance

777 views • 62 slides

More recommend