โ Image Classification
Image classification assigns a label or category to an image based on its visual content. It is a fundamental problem in computer vision and has numerous applications such as object recognition, face detection, and image retrieval.
โ Semantic gap, challenges (17:04)
The semantic gap refers to the difference between low-level visual features extracted from an image and high-level semantic concepts that humans associate with them. The challenges in image classification include dealing with variations in lighting, scale, and orientation, recognizing objects under partial occlusion, and distinguishing between objects with similar visual appearances.
Semantic Gap:
(ppt)Challenges๏ผ viewpoint variation, intraclass variation, deformation, illumination
โ Machine learning: a data-driven approach
The data-driven approach to machine learning involves training a model using a large dataset of labeled examples. The model learns to generalize patterns from the training data and can then be used to predict labels for new, unseen examples.
โ Nearest neighbor classifier
The nearest neighbor classifier is a simple but effective algorithm for image classification. It works by finding the nearest training image(s) to a test image based on some distance metric, and then assigning the label of the nearest training image(s) to the test image.
Not learning, just store
Distance Metric
ๆๅฏ่ฝๆ่ฏฏ๏ผๅ ไธบfocus on pixel level๏ผcolor๏ผ
Decision boundaries
Pixel color level difference not useful๏ผ
โ Hyperparameters
Hyperparameters are parameters in a machine learning model that are set before training and are not learned from the data. They control the complexity of the model and can have a significant impact on its performance. Examples of hyperparameters in a linear classifier include the regularization strength and the learning rate.
่ดต
ๅ
ถๅฎdeep learningๅจๆฒกๆๅพๅคdataๆถ work not well
โ Linear classifier
A linear classifier is a type of machine learning model that learns to separate data points into different classes using a linear decision boundary. This decision boundary can be represented algebraically, visually, or geometrically.
โ Algebraic, Visual, Geometric viewpoints
โ Loss functions: SVM, Softmax
In a linear classifier, the loss function is used to measure how well the model is able to predict the correct class labels. Two commonly used loss functions for linear classifiers are the support vector machine (SVM) loss and the softmax loss. The SVM loss encourages the model to have a large margin between different classes, while the softmax loss is used for multi-class classification problems and produces a probability distribution over all possible classes.
* svm๏ผๅฏน็class score ้ซ
A1: ๆ change๏ผๅ ไธบ4.9-0.5่ฟๆฏๆฏ๏ผ1.3+1๏ผๅ๏ผ2.0+1๏ผๅคง
A2๏ผmin๏ผ0๏ผmax๏ผๆญฃๆ ็ฉท
A3๏ผloop over ไธๆญฃ็กฎ็class๏ผๆC-1ไธชไธๆญฃ็กฎ๏ผmax๏ผ0๏ผ 0-0+1๏ผ=1๏ผๆไปฅ็ญๆกไธบC-1
A4: C
Q๏ผwhy should we skip the ground tree class๏ผ
A๏ผ่ฅ่่๏ผๅ
ถloss ไธบ1๏ผ0ๆฏ1ๆด่กจ็ฐๆญฃ็กฎ
A5:ๅชๆฏไธ็จmeanไบ๏ผๆฒกไปไนไธๅ
A6: bigger loss for bigger error
Linear-ใquadratic
๏ผๆ่ฎธๅคๅฐerror ๆฏๆไธไธชๅคงerrorๅฅฝ๏ผ
๏ผ้่ฆ
HOW to choose a unique W๏ผโฌ๏ธ
้ฒๆญขmodel doing to well on training data
SOFTMAX๏ผ่ฟไธชloss function็จไบmultinomial logistic regression/softmax classifier๏ผ
๏ผ็ฑไบsvmๅชๆฏไธไบscaler value็loss๏ผๅชๆฏ่ฆๆฑๅฏน็class score้ซ๏ผ่ฟๅนถไธ้ฃไนinterable๏ผ
๏ผๆณ่ฆinterpret raw classifier scores as probabilities๏ผ
A1: min๏ผ0//-log๏ผ1๏ผ=0
max๏ผๆญฃๆ ็ฉท//-log๏ผ0๏ผ=ๆญฃๆ ็ฉท
A2:
ๆป็ป๏ผ
โ Regularization
Regularization is a technique used to prevent overfitting in machine learning models. It involves adding a penalty term to the loss function that encourages the model to have smaller weights. Examples of regularization techniques include L1 and L2 regularization.
โSpread outโ the weight๏ผ ๆ็ๆฏ่ฎฉๆฏไธชweight้ฝๆ่ดก็ฎ
