How to visualize the representations learnt by hidden neurons ?

Question

0.00/5 (No votes)

See more:

I have trained a neural network to identify handwritten digits.It has 784 inputs,1 hidden layer with 50 neurons and output layer with 10 neurons.Now I want to visualize the representations learnt by hidden neurons.How can that be done ??

Posted 3-Apr-15 22:25pm

Member 11579730

Add a Solution

1 solution

Add a Solution

Add your solution here

Treat my content as plain text, not as HTML

Preview 0

…

Existing Members

Sign in to your account

...or Join us

Download, Vote, Comment, Publish.

Your Email
Password
Forgot your password?

Your Email
This email is in use. Do you need your password?
Optional Password

I have read and agree to the Terms of Service and Privacy Policy
Please subscribe me to the CodeProject newsletters

When answering a question please:

Read the question carefully.
Understand that English isn't everyone's first language so be lenient of bad spelling and grammar.
If a question is poorly phrased then either ask for clarification, ignore it, or edit the question and fix the problem. Insults are not welcome.
Don't tell someone to read the manual. Chances are they have and don't get it. Provide an answer or move on to the next question.

Let's work to help developers, not make them feel stupid.

This content, along with any associated source code and files, is licensed under The Code Project Open License (CPOL)

Peter Leow · Answer 1 · 2015-04-03T23:19:00

Solution 1

The knowledge learnt by an ANN is its snaptic weights, not the hidden neurons. Hidden neurons are computation nodes that take the inputs and their associated weights from the input layer, put them through some activation function to produce the outputs for the output layer. Check this out:
AI : Neural Network for beginners (Part 1 of 3)[^].
If you are looking for some machine learning algorithm that can produce graphical representation of its learning outcome, try SOM. Check this out: Self-organizing Map Demystified[^]

Posted 3-Apr-15 23:19pm

Peter Leow

Updated 3-Apr-15 23:21pm

v2

Comments

Member 11579730 4-Apr-15 6:22am

Actually,I want to look at which input images activate any given neuron most strongly. Do they have something in common? Has the neuron learnt something interpretable? So, for this I want to figure out and visualise the 'optimal stimulus' for a given neuron, i.e., the input image that would maximise its activation.The answers we get may not be precise; but it will be useful to get a sense for what kinds of patterns the neural network is actually picking up on.

Peter Leow 4-Apr-15 6:41am

ANN is back-box learning. It will produce outcome but does not tell you how and why. For decision making that requires how and why, then decision tree will be more appropriate.