We want to use neural network for recognition purpose. Being a supervised learning approach, it requires both input and target. Now, if we have only input and no target then how to use the Neural network?
Hello,
The second type of neural networks is called unsupervised learning, which is based on classification of data. In type of learning algorithms, no need to specify the target, you have to plugin your data and specify the classes. You can use Matlab toolbox for that.
Regards.
Say you want to recognize cats in images but you don't have labels.
What you could try is to use autoencoders (AE) which are trained unsupervised, i.e. input and output are both the unlabeled images. The AE then learns to replicate the input but compresses the image to a (potentially) lower dimensional tensor.
Based on this compressed representation, you could cluster the data points. If it turns out that one cluster represents the class cat, you have your unsupervised classificator. In this case, the only supervision is done by you when assigning your target class to a cluster.
Hello,
If you have data and you want to determine the categories of the data that belong, you can use the recurrent neural network (unsupervised neural network) , that not need to have target but only determine the number of categories as output for the the neural network. you can feed the training inputs data and the train the network to divide the inputs into desired categories that you determined.
There is a technique called self organsing map for unsupervised learning...but still it is a topic of research and exploraton.
- Badges
- Science topic