Analogator

This page describes a series of experiments performed in the PhD thesis Learning to See Analogies: A Connectionist Exploration, by Doug Blank and available here.

This work was originally done in 1997. The experiments have been re-done and documented here using the pyrobot.conx Python library for neural networks.

Visual Analogies

The following experiments require a recent Pyrobot version 5.0.2 or above, and the files in Visual-analogies.zip.

For more information on the Python conx neural network simulator, please see:

• http://pyrorobotics.org/?page=PyroModuleNeuralNetworks

To get started:

1. Unzip the files into a directory, say visual-analogies
2. You may need to set your PYTHONPATH environment variable to include the pyrobot directory
3. run python -i visual-analogies.py

You can interrupt the processing (control+c) and interactively explore the processing:

>>> n.interactive = 1
>>> n.propagate(input=all(letters[i]), 
                context=body(letters[i]), 
                figure=body(letters[i]), 
                ground=brim(letters[i]), 
                )

More soon... see the file visual-analogies.py for more details.