How regression is different from artificial nueral networks?
Answers
Answered by
0
Ok, let’s start from where both are equals. Both of them are supervised Machine Learning algorithms that have two main challenges:
Training the model
Predicting from the model
Ok, now we already have limited our scoop, but it still not enough, there are a lot of different supervised ML algorithms. ANN(Artificial Neural Networks) and Logistic Regression are algorithms related to classification problems where you have a discrete number of possibilities(e.g not a continuous value ).
Logistic regression can have a single variable classification where many people use sigmoid function, and softmax function to multiclass-classification problems. Neural Networks has a very particular “structure” where you have one input layer, at least one hidden layer and finally an ouput layer.
Both algorithms have a similar problem: to find the best value for their parameters. Usually people use gradient descent in logistic regression to minimize the cost function and on each iteration better values to fit their data. In Neural Networks you have a different (but not that different) way, we use back propagation to find the parameters (or weights) to our inputs.
There is an interesting case when you have a Neural Network with one input node, hidden node and output node, when using softmax where you can check that this structure of Neural Network is equivalent to a Logistic Regression.
To make their predictions, each one uses a different way. Logistic Regression uses sigmoid function to find a probability P(Y =1| Given_X) where 1 is a class and 0 is the other possibility. On Neural Networks you need to use ForwardPropagation and then on the last part you will use softmax or another function to find the probability that that specific input is of a class 1,2,..,K. Notice that on Neural Networks you will usually use(if not most of the time) for multi class classification problems, like finding specific classes on images (numbers, letters, people, cars etc..) and on Logistic regression when not multi-class like with One-vs-All or softmax will be used on binary classification.
For a more detailed explanation I would suggest this Stanford guide for Supervised Learning Algorithms that goes from Linear Regression, Logistic Regression, Softmax Regression, Neural Networks and so on. I higly recommend this so you can have on your own the solid knowledge about this topics:
Training the model
Predicting from the model
Ok, now we already have limited our scoop, but it still not enough, there are a lot of different supervised ML algorithms. ANN(Artificial Neural Networks) and Logistic Regression are algorithms related to classification problems where you have a discrete number of possibilities(e.g not a continuous value ).
Logistic regression can have a single variable classification where many people use sigmoid function, and softmax function to multiclass-classification problems. Neural Networks has a very particular “structure” where you have one input layer, at least one hidden layer and finally an ouput layer.
Both algorithms have a similar problem: to find the best value for their parameters. Usually people use gradient descent in logistic regression to minimize the cost function and on each iteration better values to fit their data. In Neural Networks you have a different (but not that different) way, we use back propagation to find the parameters (or weights) to our inputs.
There is an interesting case when you have a Neural Network with one input node, hidden node and output node, when using softmax where you can check that this structure of Neural Network is equivalent to a Logistic Regression.
To make their predictions, each one uses a different way. Logistic Regression uses sigmoid function to find a probability P(Y =1| Given_X) where 1 is a class and 0 is the other possibility. On Neural Networks you need to use ForwardPropagation and then on the last part you will use softmax or another function to find the probability that that specific input is of a class 1,2,..,K. Notice that on Neural Networks you will usually use(if not most of the time) for multi class classification problems, like finding specific classes on images (numbers, letters, people, cars etc..) and on Logistic regression when not multi-class like with One-vs-All or softmax will be used on binary classification.
For a more detailed explanation I would suggest this Stanford guide for Supervised Learning Algorithms that goes from Linear Regression, Logistic Regression, Softmax Regression, Neural Networks and so on. I higly recommend this so you can have on your own the solid knowledge about this topics:
Similar questions