Member-only story
Gradient Descent Intuition — How Machines Learn
The algorithm to find the “lowest-cost” state in which the ML system can exist.
Gradient descent is an algorithm to find the minimum of a function. In machine learning, the difference between the predicted and the actual value is called error. All error, across all datapoints, when added together is known as the cost.
Naturally, we would want to minimize the function that represents this cost — Cost Function. When the cost function cannot be differentiated easily Gradient Descent does the job for us.
Step by step we descend,
GD is the means to our end
- The AI Poet
Where in Machine Learning does Gradient Descent figure?
Gradient descent figures in the training phase of any machine learning algorithm. For example, neural networks are trained by using a technique called Backpropagation. Gradient descent forms a very important part of backpropagation.
Gradient descent is a very popular method to tune the parameters of a machine learning model — so as to attain the minimum error state.
A machine learning algorithm is about telling the machine to learn data’s behavior. Given that a machine can only understand 0’s and 1’s it is…
