perceptron

simple classifier $y=Wx+b$ transformation that converts feature vectors into recognition scores

for Computer Vision

• input (x): image features (pixels, Histogram of Oriented Gradients (HoG), etc.)
• weights (w): learned params
• output (y): scores for each class

interpretation

• each row of W is a template for a class
• High inner product (similarity) → high confidence

see linear classifier linear models are the foundation for neural networks