Lecture #3 - Binary Classification

Question 1

When discussing Modulation Classification; What are the drawbacks?

Answer 1

1. We don’t know if the threshold 1.3 is adequate.
2. We don’t know of the feature F is suited for the classification.

Question 2

Explain what a feature vector is

Answer 2

A mathematical representation of an object or data point in ML or data analysis content.

e.g. If we have a vector as shown:

[5.2, 100, 0.3, 0, 1.2]

This represents input data for training models. Could be examples like email length, number of key strokes.

Question 3

What does the sigmoid function do?

Answer 3

The sigmoid function converts the problem to the probabilistic domain.

It restricts the output to [0, 1]

Question 4

What is automatic differentiation?

Answer 4

It takes advantage that a complex number can be broken down into multiplications, additions, subtraction and division.

Question 5

What is the softmax function?

Answer 5

The softmax function is a mathematical function that takes a vector of real numbers as input and outputs a vector of probabilities that sum up to 1.

Question 6

When do we use the soft max function?

Answer 6

We use the soft max function when we have more than classes. When it’s only one class, we can use the sigmoid function.

Question 7

What is a vector of logits?

Answer 7

The input to a soft max function. A vector of logits refers to a vector of unnormalized or raw model outputs before they are transformed into probabilities using a softmax function.

Question 8

Why can the softmax function be numerically challenging? How do we remedy this?

Answer 8

It’s numerically challenging if one element x_i, is very large.