Example: bankruptcy

Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 2 ...

Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 20171 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: PiazzaFor questions about midterm, poster session, projects, use Piazza instead of staff list!SCPD students: Use your address to register for Piazza; contact for Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: Assignment 1 Out tonight, due 4/18 11:59pm-K-Nearest Neighbor-Linear classifiers: SVM, Softmax-Two-layer neural network-Image features3 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: Python + Numpy4 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: Google Cloud5 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Image Classification: A core task in Computer Vision6cat(assume given set of discrete labels){dog, cat, truck, plane.}

Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 2 - April 6, 2017 Administrative: Piazza For questions about midterm, poster session, projects,

Fullscreen Download

Tags:

  Li amp justin johnson amp serena yeung lecture

Information

Domain:

Source:

Link to this page:

Please notify us if you found a problem with this document:

Other abuse

Transcription of Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 2 ...

1 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 20171 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: PiazzaFor questions about midterm, poster session, projects, use Piazza instead of staff list!SCPD students: Use your address to register for Piazza; contact for Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: Assignment 1 Out tonight, due 4/18 11:59pm-K-Nearest Neighbor-Linear classifiers: SVM, Softmax-Two-layer neural network-Image features3 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: Python + Numpy4 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Administrative: Google Cloud5 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Image Classification: A core task in Computer Vision6cat(assume given set of discrete labels){dog, cat, truck, plane.}

2 }This image by Nikita is licensed under CC-BY Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 This image by Nikita is licensed under CC-BY Problem: Semantic Gap7 What the computer seesAn image is just a big grid of numbers between [0, 255] 800 x 600 x 3(3 channels RGB) Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Challenges: Viewpoint variation8 All pixels change when the camera moves!This image by Nikita is licensed under CC-BY Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Challenges: Illumination9 This image is CC0 public domainThis image is CC0 public domainThis image is CC0 public domainThis image is CC0 public domainFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Challenges: Deformation10 This image by Umberto Salvagnin is licensed under CC-BY image by Tom Thai is licensed under CC-BY This image by sare bear is licensed under CC-BY image by Umberto Salvagnin is licensed under CC-BY Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Challenges.

3 Occlusion11 This image is CC0 public domainThis image by jonsson is licensed under CC-BY image is CC0 public domainFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201712 This image is CC0 public domainChallenges: Background ClutterThis image is CC0 public domainFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Challenges: Intraclass variation13 This image is CC0 public domainFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017An image classifier14 Unlike sorting a list of numbers, no obvious way to hard-code the algorithm for recognizing a cat, or other Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Attempts have been made15 John Canny, A Computational Approach to Edge Detection , IEEE TPAMI 1986 Find edgesFind corners? Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Data-Driven a dataset of images and Machine Learning to train a the classifier on new imagesExample training setFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 First classifier: Nearest Neighbor17 Memorize all data and labelsPredict the label of the most similar training imageFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Example Dataset: CIFAR1018 Alex Krizhevsky, Learning Multiple Layers of Features from Tiny Images , Technical Report, classes50,000 training images10,000 testing imagesFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Example Dataset.

4 CIFAR1019 Alex Krizhevsky, Learning Multiple Layers of Features from Tiny Images , Technical Report, classes50,000 training images10,000 testing imagesTest images and nearest neighborsFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Distance Metric to compare images20L1 distance:addFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201721 Nearest Neighbor classifierFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201722 Nearest Neighbor classifierMemorize training dataFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201723 Nearest Neighbor classifierFor each test image: Find closest train image Predict label of nearest imageFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201724 Nearest Neighbor classifierQ: With N examples, how fast are training and prediction? Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201725 Nearest Neighbor classifierQ: With N examples, how fast are training and prediction?

5 A: Train O(1), predict O(N) Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201726 Nearest Neighbor classifierQ: With N examples, how fast are training and prediction?A: Train O(1), predict O(N)This is bad: we want classifiers that are fast at prediction; slow for training is okFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 What does this look like?27 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017K-Nearest Neighbors28 Instead of copying label from nearest neighbor, take majority vote from K closest pointsK = 1K = 3K = 5 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201729 What does this look like? Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201730 What does this look like? Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017K-Nearest Neighbors: Distance Metric31L1 (Manhattan) distanceL2 (Euclidean) distanceFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017K-Nearest Neighbors: Distance Metric32L1 (Manhattan) distanceL2 (Euclidean) distanceK = 1K = 1 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017K-Nearest Neighbors: Demo Time33 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Hyperparameters34 What is the best value of k to use?

6 What is the best distance to use?These are hyperparameters: choices about the algorithm that we set rather than learnFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Hyperparameters35 What is the best value of k to use?What is the best distance to use?These are hyperparameters: choices about the algorithm that we set rather than learnVery problem-dependent. Must try them all out and see what works Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters36 Idea #1: Choose hyperparameters that work best on the dataYour DatasetFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters37 Idea #1: Choose hyperparameters that work best on the dataBAD: K = 1 always works perfectly on training dataYour DatasetFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters38 Idea #1: Choose hyperparameters that work best on the dataBAD: K = 1 always works perfectly on training dataIdea #2: Split data into train and test, choose hyperparameters that work best on test dataYour DatasettraintestFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters39 Idea #1.

7 Choose hyperparameters that work best on the dataBAD: K = 1 always works perfectly on training dataIdea #2: Split data into train and test, choose hyperparameters that work best on test dataBAD: No idea how algorithm will perform on new dataYour DatasettraintestFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters40 Idea #1: Choose hyperparameters that work best on the dataBAD: K = 1 always works perfectly on training dataIdea #2: Split data into train and test, choose hyperparameters that work best on test dataBAD: No idea how algorithm will perform on new dataYour DatasettraintestIdea #3: Split data into train, val, and test; choose hyperparameters on val and evaluate on testBetter!traintestvalidationFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters41 Your Datasettestfold 1fold 2fold 3fold 4fold 5 Idea #4: Cross-Validation: Split data into folds, try each fold as validation and average the resultstestfold 1fold 2fold 3fold 4fold 5testfold 1fold 2fold 3fold 4fold 5 Useful for small datasets, but not used too frequently in deep learningFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Setting Hyperparameters42 Example of5-fold cross-validationfor the value of point: singleoutcome.

8 The line goesthrough the mean, barsindicated standarddeviation(Seems that k ~= 7 works bestfor this data) Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201743k-Nearest Neighbor on images never slow at test time-Distance metrics on pixels are not informative(all 3 images have same L2 distance to the one on the left)OriginalBoxedShiftedTintedOriginal image is CC0 public domainFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201744k-Nearest Neighbor on images never of dimensionalityDimensions = 1 Points = 4 Dimensions = 3 Points = 43 Dimensions = 2 Points = 42 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017K-Nearest Neighbors: SummaryIn Image classification we start with a training set of images and labels, and must predict labels on the test setThe K-Nearest Neighbors classifier predicts labels based on nearest training examplesDistance metric and K are hyperparametersChoose hyperparameters using the validation set; only run on the test set once at the very end!

9 45 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201746 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201747 This image is CC0 public domainNeural NetworkLinear classifiersFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201748 Man in black shirt is playing worker in orange safety vest is working on young girls are playing with lego is doing backflip on wakeboardKarpathy and Fei-Fei , Deep Visual-Semantic Alignments for Generating Image Descriptions , CVPR 2015 Figures copyright IEEE, 2015. Reproduced for educational Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201749 Man in black shirt is playing worker in orange safety vest is working on young girls are playing with lego is doing backflip on wakeboardKarpathy and Fei-Fei , Deep Visual-Semantic Alignments for Generating Image Descriptions , CVPR 2015 Figures copyright IEEE, 2015.

10 Reproduced for educational Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Recall CIFAR105050,000 training images each image is 32x32x310,000 test Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Parametric Approach51 Imagef(x,W)10 numbers giving class scoresArray of 32x32x3 numbers(3072 numbers total)parametersor weightsWFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Parametric Approach: Linear Classifier52 Imageparametersor weightsWf(x,W)10 numbers giving class scoresArray of 32x32x3 numbers(3072 numbers total)f(x,W) = WxFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Parametric Approach: Linear Classifier53 Imageparametersor weightsWf(x,W)10 numbers giving class scoresArray of 32x32x3 numbers(3072 numbers total)f(x,W) = Wx10x110x30723072x1 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Parametric Approach: Linear Classifier54 Imageparametersor weightsWf(x,W)10 numbers giving class scoresArray of 32x32x3 numbers(3072 numbers total)f(x,W) = Wx + b3072x110x110x307210x1 Fei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 201755 Example with an image with 4 pixels, and 3 classes (cat/dog/ship) image5623124256231242 Stretch pixels into + scoreDog scoreShip scoreFei-Fei Li & Justin Johnson & Serena YeungLecture 2 - April 6, 2017 Interpreting a Linear Classifier56 What is this thing doing?

Show more

Documents from same domain

Convolutional Neural Networks for Visual Recognition

Convolutional Neural Networks for Visual Recognition

cs231n.stanford.edu

Progressive GAN, Karras 2018. Models from Single RGB Images”, ECCV 2018 Beyond recognition: Segmentation, 2D/3D Generation. Fei-Fei Li, Ranjay Krishna, Danfei Xu Lecture 1 - 15 March 30, 2021 Scene Graphs Krishna et al., Visual Genome: Connecting Vision and Language using Crowdsourced Image Annotations, IJCV 2017

  Network, Visual, Recognition, Neural, Convolutional, Karar, Convolutional neural networks for visual recognition

Lecture 9: CNN Architectures

Lecture 9: CNN Architectures

cs231n.stanford.edu

Lecture 9 - 22 May 2, 2017 ImageNet Large Scale Visual Recognition Challenge (ILSVRC) winners First CNN-based winner. Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 9 - 23 May 2, 2017 ImageNet Large Scale Visual Recognition Challenge (ILSVRC) winners ZFNet: Improved hyperparameters over AlexNet. Fei-Fei Li & Justin Johnson & Serena Yeung ...

  2017, Challenges, Scale, Visual, Recognition, Ilsvrc, Scale visual recognition challenge

Attention and Transformers Lecture 11

Attention and Transformers Lecture 11

cs231n.stanford.edu

graph with shared weights h 0 f W h 1 f W h 2 f W h 3 x 3 y T ... Extract spatial features from a pretrained CNN Image Captioning using spatial features 11 CNN Features: H x W x D h 0 [START] Xu et al, “Show, Attend and Tell: Neural Image Caption Generation with Visual Attention”, ICML 2015 z 0,0 z 0,1 z 0,2 z 1,0 z 1,1 z 1,2 z 2,0 z 2,1 z ...

  Transformers, Attention, Graph, Spatial, Attention and transformers

NaveenAppiah SagarVare - Stanford University

NaveenAppiah SagarVare - Stanford University

cs231n.stanford.edu

NaveenAppiah Mechanical Engineering [email protected] SagarVare Stanford ICME [email protected] ... the popular mobile game - Flappy Bird. It involves navi-gating a bird through a bunch of obstacles. Though, this ... the game emulator and learns to make good decisions over time. It is this simple learning framework and their

  Make, Games, Flappy, Naveenappiah

CNNs for Face Detection and Recognition

CNNs for Face Detection and Recognition

cs231n.stanford.edu

development of object classification, localization and detec-tion techniques. 2.1. Sliding Window In the early development of face detection, researchers tended to treat it as a repetitive task of object classifica-tion, by imposing sliding windows and performing object classification with the neural networks on the window re-gion.

  Technique, Faces, Recognition, Object, Detection, For face detection and recognition

Vector, Matrix, and Tensor Derivatives

Vector, Matrix, and Tensor Derivatives

cs231n.stanford.edu

Erik Learned-Miller The purpose of this document is to help you learn to take derivatives of vectors, matrices, and higher order tensors (arrays with three dimensions or more), and to help you take ... At this point, we have reduced the original matrix equation (Equation 1) …

  Have, Learned

Lecture 14: Reinforcement Learning

Lecture 14: Reinforcement Learning

cs231n.stanford.edu

Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 14 - May 23, 2017 Markov Decision Process 19 - Mathematical formulation of the RL problem - Markov property: Current state completely characterises the state of the

  Learning, Reinforcement, Reinforcement learning

Lecture 11: Detection and Segmentation

Lecture 11: Detection and Segmentation

cs231n.stanford.edu

Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 11 - 1 May 10, 2017 Lecture 11: Detection and Segmentation

  Detection, Segmentation

Lecture 13: Generative Models

Lecture 13: Generative Models

cs231n.stanford.edu

Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 13 - May 18, 2017 Generative Models 17 Training data ~ p data (x) Generated samples ~ p model (x) Want to learn p

  Generative

Lecture 10: Recurrent Neural Networks

Lecture 10: Recurrent Neural Networks

cs231n.stanford.edu

image -> sequence of words. Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 10 - 13 May 4, 2017 Recurrent Neural Networks: Process Sequences e.g. Sentiment Classification sequence of words -> sentiment. Fei-Fei Li & Justin Johnson & Serena Yeung Lecture 10 - 14 May 4, 2017

  Image, Recurrent

Related documents

Long Short-Term Memory - University of Wisconsin–Madison

Long Short-Term Memory - University of Wisconsin–Madison

pages.cs.wisc.edu

Karpathy, Andrej, and Li Fei-Fei. "Deep visual-semantic alignments for generating image descriptions." Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition. 2015. (Paper introducing image captioning using ConvNet + LSTM)

  Visual, Generating, Alignment, Semantics, Visual semantic alignments for generating

Related search queries

Visual-semantic alignments for generating