Insights on assignment 1 - image classification:

Raven

• with white only background it definitely works best. Clean background.
• Some classes that are really different from other classes are recognized in a better way.

The output of the confidence number is 1000 neurons. If each one is a number the cat is light up.

Classification vs Regression

2 TYPES OF OUTPUT:

Classification: if given an input image is a dog or cat. Discrete set of possibilities. Like Mobile Net

Regression: Is a continuous number output. Stock price of a house, number of bedrooms.

What is a pre-trained model?

Is like a formula or a pre fabricated solution. Frozen state of that model. Most of the ML5 models are pre trained. You can think of them like API service, you add an input and delivers some output. We can also train our own model at some point. ChatGPT or Mobile net are a pre-trained model. You can think of it like a function.

Mobile Net

Convolutional Neural Networks for Mobile Vision Applications. Give an image and classifies it. Before was Alex Net. Small, fast and accurate.

ML5js would guive ups probability, class name and class ID of each image. .The output in raw is a one thousand number that takes the distribution of those classes. Sumarizing the high level semantic detail.

Clear number in the wave for the actual classes. Basically noise for the rest. Confidence is the actual number of the model outputs. These not normalized numbers are Logits.

AlexNet