16. Multi Class Neural Networks

Multi-Class Neural Networks

---

Topic: Multi-Class Neural Networks

Course: GMLC

Date: 3 April 2019   

Professor: Not specified

---

Resources

---

• https://developers.google.com/machine-learning/crash-course/multi-class-neural-networks/video-lecture

• https://developers.google.com/machine-learning/crash-course/multi-class-neural-networks/one-vs-all

• https://developers.google.com/machine-learning/crash-course/multi-class-neural-networks/softmax

• https://developers.google.com/machine-learning/crash-course/multi-class-neural-networks/programming-exercise

Key Points

---

• One vs all

  • Leverages binary classification

  • N possible solutions

  • N separate binary classifiers (one for each outcome)

  • Inefficient with large number of classes

• Softmax

  • Assigns decimal probabilities to each class

  • Must have same number of nodes as the output

  • Options

    • Full Softmax

    • Calculates probability for every possible class

  • Candidate sampling

    • Calculates a probability for all possible labels but only for a random sample of negative labels

    • Becomes inefficient as number of classes rise

• Multi class neural network

  • Opposed to binary neural network, multi class calculates the output between many classes

Check your understanding

---

• Explain what is a multi class neural network

• Know types of multi class neural networks

Summary of Notes

---

• Multi class neural networks work by setting algorithms on nodes for each possible class of our model

• We can choose between brute force class possibility or a candidate sampling type of algorithm