We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PPTX, PDF, TXT or read online on Scribd
You are on page 1/ 19
Topic:
Binary Cross Entropy and
Categorical Cross Entropy Name – Lakshya Singh Roll.no – 2021UEA6511 Date – 05/03/2024 Introduction
Entropy is a concept from information theory. It
quantifies the amount of uncertainty involved in predicting the outcome of a random variable. In the context of machine learning, we often use entropy to quantify the “impurity” of an input set. Importance of Entropy
In machine learning, entropy is used to quantify the
uncertainty or randomness of a set. It’s used in various algorithms like decision trees and clustering. It also forms the basis for more advanced concepts like cross entropy and information gain. Binary Cross Entropy - Definition
Binary Cross Entropy is a loss function used in
binary classification tasks. If ‘p’ is the prediction and ‘y’ is the actual value, then the binary cross entropy is defined as: Binary Cross Entropy - Intuition
Binary Cross Entropy measures how far away our
predictions are from the actual values. In other words, it tells us the cost of our wrong predictions. Binary Cross Entropy - Use Cases
Binary Cross Entropy is used in binary classification
problems like spam detection, tumor detection etc. Binary Cross Entropy - Example
Let’s say we have a binary classification problem with the
following predictions and actual values: Predictions = [0.1, 0.9, 0.6], Actual Values = [0, 1, 1]. The Binary Cross Entropy can be calculated as follows: Binary Cross Entropy - Properties
Binary Cross Entropy is always positive, and tends
towards zero as we get better at correctly predicting the actual value. Categorical Cross Entropy - Definition
Categorical Cross Entropy is a loss function used in
multi-class classification tasks. If ‘p’ is the prediction and ‘y’ is the actual value, then the categorical cross entropy is defined as:
where ‘m’ is the number of classes.
Categorical Cross Entropy - Intuition
Categorical Cross Entropy measures how far away
our predictions are from the actual values for multi- class classification problems. It tells us the cost of our wrong predictions. Categorical Cross Entropy - Use Cases
Categorical Cross Entropy is used in multi-class
classification problems like digit recognition, animal classification etc. Categorical Cross Entropy - Example
Let’s say we have a 3-class classification problem with
the following predictions and actual values: Predictions = [[0.2, 0.3, 0.5], [0.1, 0.6, 0.3]], Actual Values = [[0, 0, 1], [0, 1, 0]]. The Categorical Cross Entropy can be calculated as follows: Categorical Cross Entropy - Properties
Categorical Cross Entropy is always positive, and tends
towards zero as we get better at correctly predicting the actual value. Comparison
Both Binary and Categorical Cross Entropy are used as a
measure of error in classification problems. While Binary Cross Entropy is used for binary classification problems, Categorical Cross Entropy is used for multi-class classification problems. Comparison Practical Tips
• Be careful with the base of the logarithm used in the
entropy calculations. In machine learning, we usually use the natural logarithm. • Make sure the actual values are one-hot encoded for multi-class classification problems. Common Pitfalls
• One common pitfall is thinking that a lower entropy
always means a better model. While it’s true that we want to minimize the entropy, a model with a very low entropy could be overfitting the data. • Another common pitfall is using Binary Cross Entropy for multi-class classification problems. Make sure to use the right loss function for the right problem. Summary
• Entropy is a measure of uncertainty or randomness. In
machine learning, it’s used to quantify the “impurity” of an input set. • Binary Cross Entropy and Categorical Cross Entropy are two types of entropy used in binary and multi-class classification problems respectively. • Both these measures tell us how far away our predictions are from the actual values, and we aim to minimize these quantities to build a good model. References
The Academic Performance of Grade 10 Mathematics Learners Exposed To Hybrid Learning and Printed Modular Distance Learning in San Andres District, Schools Division of Quezon