3.8 Basis Functions in Neural Networks
Basis Functions in Neural Networks: -
- Basis functions are mathematical functions used to transform input data into a new form so that machine learning models can learn patterns more easily.
- Sometimes the relationship between input and output is not simple or linear. They help convert complex relationships into simpler forms and improve prediction accuracy.
- Examples include polynomial functions and activation functions such as ReLU and Sigmoid used in neural networks.
- In neural networks, activation functions act like basis functions. They transform the input values so the network can learn complex relationships.
Common examples include:
-
Sigmoid function
-
ReLU function
-
Tanh function
These functions convert the input into non-linear outputs, helping the network learn better.
Example
Suppose we want to predict student performance based on study hours.
Input: Study hours = 2, 4, 6, 8
But the relationship between study hours and marks may not be perfectly straight. A basis function can transform the data.
Example transformation:
-
Original input:
-
Basis function:
If study hours = 4
New value =
This transformation helps the model capture more complex patterns.