1
$\begingroup$

In spike-based neural networks, there is a learning rule called STDP (Spike-Timing-Dependent Plasticity). It's a completely unsupervised learning rule that works continuously when data is fed into the network.

I've been trying to find learning rules that work like STDP, but can be applied to a numerical-based network with multiple layers, and complex connections (e.g., multi-layer perceptron). I found some interesting techniques such as: Principal Component Analysis, Kohonen's Self-Organizing Map, Competitive Learning, Oja's Rule, and the Hebbian rule.

But none of these can be applied to a deeper neural network with multiple layers and more complex connections.

I'm looking for a learning rule that works unsupervised and online, similar to STDP, but that can be applied to traditional neural networks.

Improve this question
$\endgroup$
3
  • $\begingroup$ Have you considered auto encoders? $\endgroup$
    – Gaslight Deceive Subvert
    Commented Aug 4 at 19:40
  • $\begingroup$ What kind of auto encoders do you mean? Because the simple idea of auto encoders is to learn using supervised backpropagation to convert the input information to a smaller input space and then decode it back as best as possible. $\endgroup$
    – aarong
    Commented Aug 8 at 12:36
  • $\begingroup$ Autoencoding is a form of unsupervised machine learning. So it's unclear what you're looking for. Unsupervised and online is extremely broad is very broad. $\endgroup$
    – Gaslight Deceive Subvert
    Commented Aug 8 at 16:28

0

Reset to default

You must log in to answer this question.