Day01

Nearest Neighbor

Memorize all data and labels

Predict the label of the most similar training image

The nearest neighbor algorithm is the lazy algorithm

K-Nearest Neighbors

instead of copying label from nearest neighbors, take majority vote from K closest points.

K-Nearest Neighbors on images never used

• Very slow at test time
• Distance metrices on pixels are not informative

Distance metrices and K are hyperparameters

Day02

Linear Classification

• Loss function
• Optimization
• Convnets