You can think about it as that machine learning is the big umbrella that includes deep learning and other linear algorithms (e.g. linear regression). Now, supervised learning is when you explicitly configure the learning algorithm to learn towards a pre-defend target(s). While in unsupervised learning, the target of learning is not explicitly defined or known at the learning stage such as those clustering algorithms.

Now, back to your main question, transfer learning-based models are initially learned using supervised learning in which it captures latent features that can aid in accelerating the learning process in different problems/data. In which, you can utilise those pre-trained algorithms weights to solve a problem that it was not intended to solve at the first place. However, in most cases, you need to perform some type of fine-tuning. In which you freeze the earlier layers of the network (which theatrically learned the common features, e.g. lines, edges colours…etc.) and only re-train those last few layers on the new data domain that you actually want to use the model for.

You can think about it as that machine learning is the big umbrella that includes deep learning and other linear algorithms (e.g. linear regression). Now, supervised learning is when you explicitly configure the learning algorithm to learn towards a pre-defend target(s). While in unsupervised learning, the target of learning is not explicitly defined or known at the learning stage such as those clustering algorithms.

Now, back to your main question, transfer learning-based models are initially learned using supervised learning in which it captures latent features that can aid in accelerating the learning process in different problems/data. In which, you can utilise those pre-trained algorithms weights to solve a problem that it was not intended to solve at the first place. However, in most cases, you need to perform some type of fine-tuning. In which you freeze the earlier layers of the network (which theatrically learned the common features, e.g. lines, edges colours…etc.) and only re-train those last few layers on the new data domain that you actually want to use the model for.

Dear,

Machine learning models or classifiers such as SVM can be trained in both supervised and unsupervised, same for deep learning model, which are neural networks with different schems (recurrent, feedforward..). However Transfer Learning is a prcedure or method that use other training episods to build a new one but not from scratch. hope that help you

Both Transfer learning and Deep learning can be regarded as supervised and Unsupervised, while machine is the general name.

thanks for the replies, i need any real time algorithm for human activity detection using machine learning ? any Matlab algorithm ?

They can all be supervised or un-supervised.

U can easily get in Python than MATLAB. Visit GitHub and search using your desired keywords.

Thanks